GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/vp3.c Lines: 1245 1542 80.7 %
Date: 2021-01-20 23:14:43 Branches: 815 1116 73.0 %

Line Branch Exec Source
1
/*
2
 * Copyright (C) 2003-2004 The FFmpeg project
3
 * Copyright (C) 2019 Peter Ross
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21
22
/**
23
 * @file
24
 * On2 VP3/VP4 Video Decoder
25
 *
26
 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
27
 * For more information about the VP3 coding process, visit:
28
 *   http://wiki.multimedia.cx/index.php?title=On2_VP3
29
 *
30
 * Theora decoder by Alex Beregszaszi
31
 */
32
33
#include <stdio.h>
34
#include <stdlib.h>
35
#include <string.h>
36
37
#include "libavutil/imgutils.h"
38
#include "libavutil/mem_internal.h"
39
40
#include "avcodec.h"
41
#include "get_bits.h"
42
#include "hpeldsp.h"
43
#include "internal.h"
44
#include "mathops.h"
45
#include "thread.h"
46
#include "videodsp.h"
47
#include "vp3data.h"
48
#include "vp4data.h"
49
#include "vp3dsp.h"
50
#include "xiph.h"
51
52
#define VP3_MV_VLC_BITS     6
53
#define VP4_MV_VLC_BITS     6
54
#define SUPERBLOCK_VLC_BITS 6
55
56
#define FRAGMENT_PIXELS 8
57
58
// FIXME split things out into their own arrays
59
typedef struct Vp3Fragment {
60
    int16_t dc;
61
    uint8_t coding_method;
62
    uint8_t qpi;
63
} Vp3Fragment;
64
65
#define SB_NOT_CODED        0
66
#define SB_PARTIALLY_CODED  1
67
#define SB_FULLY_CODED      2
68
69
// This is the maximum length of a single long bit run that can be encoded
70
// for superblock coding or block qps. Theora special-cases this to read a
71
// bit instead of flipping the current bit to allow for runs longer than 4129.
72
#define MAXIMUM_LONG_BIT_RUN 4129
73
74
#define MODE_INTER_NO_MV      0
75
#define MODE_INTRA            1
76
#define MODE_INTER_PLUS_MV    2
77
#define MODE_INTER_LAST_MV    3
78
#define MODE_INTER_PRIOR_LAST 4
79
#define MODE_USING_GOLDEN     5
80
#define MODE_GOLDEN_MV        6
81
#define MODE_INTER_FOURMV     7
82
#define CODING_MODE_COUNT     8
83
84
/* special internal mode */
85
#define MODE_COPY             8
86
87
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb);
88
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
89
90
91
/* There are 6 preset schemes, plus a free-form scheme */
92
static const int ModeAlphabet[6][CODING_MODE_COUNT] = {
93
    /* scheme 1: Last motion vector dominates */
94
    { MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
95
      MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
96
      MODE_INTRA,            MODE_USING_GOLDEN,
97
      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
98
99
    /* scheme 2 */
100
    { MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
101
      MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
102
      MODE_INTRA,            MODE_USING_GOLDEN,
103
      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
104
105
    /* scheme 3 */
106
    { MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
107
      MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
108
      MODE_INTRA,            MODE_USING_GOLDEN,
109
      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
110
111
    /* scheme 4 */
112
    { MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
113
      MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
114
      MODE_INTRA,            MODE_USING_GOLDEN,
115
      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
116
117
    /* scheme 5: No motion vector dominates */
118
    { MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
119
      MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
120
      MODE_INTRA,            MODE_USING_GOLDEN,
121
      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
122
123
    /* scheme 6 */
124
    { MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
125
      MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
126
      MODE_INTER_PLUS_MV,    MODE_INTRA,
127
      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
128
};
129
130
static const uint8_t hilbert_offset[16][2] = {
131
    { 0, 0 }, { 1, 0 }, { 1, 1 }, { 0, 1 },
132
    { 0, 2 }, { 0, 3 }, { 1, 3 }, { 1, 2 },
133
    { 2, 2 }, { 2, 3 }, { 3, 3 }, { 3, 2 },
134
    { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 }
135
};
136
137
enum {
138
    VP4_DC_INTRA  = 0,
139
    VP4_DC_INTER  = 1,
140
    VP4_DC_GOLDEN = 2,
141
    NB_VP4_DC_TYPES,
142
    VP4_DC_UNDEFINED = NB_VP4_DC_TYPES
143
};
144
145
static const uint8_t vp4_pred_block_type_map[8] = {
146
    [MODE_INTER_NO_MV]      = VP4_DC_INTER,
147
    [MODE_INTRA]            = VP4_DC_INTRA,
148
    [MODE_INTER_PLUS_MV]    = VP4_DC_INTER,
149
    [MODE_INTER_LAST_MV]    = VP4_DC_INTER,
150
    [MODE_INTER_PRIOR_LAST] = VP4_DC_INTER,
151
    [MODE_USING_GOLDEN]     = VP4_DC_GOLDEN,
152
    [MODE_GOLDEN_MV]        = VP4_DC_GOLDEN,
153
    [MODE_INTER_FOURMV]     = VP4_DC_INTER,
154
};
155
156
typedef struct {
157
    int dc;
158
    int type;
159
} VP4Predictor;
160
161
#define MIN_DEQUANT_VAL 2
162
163
typedef struct HuffEntry {
164
    uint8_t len, sym;
165
} HuffEntry;
166
167
typedef struct HuffTable {
168
    HuffEntry entries[32];
169
    uint8_t   nb_entries;
170
} HuffTable;
171
172
typedef struct Vp3DecodeContext {
173
    AVCodecContext *avctx;
174
    int theora, theora_tables, theora_header;
175
    int version;
176
    int width, height;
177
    int chroma_x_shift, chroma_y_shift;
178
    ThreadFrame golden_frame;
179
    ThreadFrame last_frame;
180
    ThreadFrame current_frame;
181
    int keyframe;
182
    uint8_t idct_permutation[64];
183
    uint8_t idct_scantable[64];
184
    HpelDSPContext hdsp;
185
    VideoDSPContext vdsp;
186
    VP3DSPContext vp3dsp;
187
    DECLARE_ALIGNED(16, int16_t, block)[64];
188
    int flipped_image;
189
    int last_slice_end;
190
    int skip_loop_filter;
191
192
    int qps[3];
193
    int nqps;
194
    int last_qps[3];
195
196
    int superblock_count;
197
    int y_superblock_width;
198
    int y_superblock_height;
199
    int y_superblock_count;
200
    int c_superblock_width;
201
    int c_superblock_height;
202
    int c_superblock_count;
203
    int u_superblock_start;
204
    int v_superblock_start;
205
    unsigned char *superblock_coding;
206
207
    int macroblock_count; /* y macroblock count */
208
    int macroblock_width;
209
    int macroblock_height;
210
    int c_macroblock_count;
211
    int c_macroblock_width;
212
    int c_macroblock_height;
213
    int yuv_macroblock_count; /* y+u+v macroblock count */
214
215
    int fragment_count;
216
    int fragment_width[2];
217
    int fragment_height[2];
218
219
    Vp3Fragment *all_fragments;
220
    int fragment_start[3];
221
    int data_offset[3];
222
    uint8_t offset_x;
223
    uint8_t offset_y;
224
    int offset_x_warned;
225
226
    int8_t (*motion_val[2])[2];
227
228
    /* tables */
229
    uint16_t coded_dc_scale_factor[2][64];
230
    uint32_t coded_ac_scale_factor[64];
231
    uint8_t base_matrix[384][64];
232
    uint8_t qr_count[2][3];
233
    uint8_t qr_size[2][3][64];
234
    uint16_t qr_base[2][3][64];
235
236
    /**
237
     * This is a list of all tokens in bitstream order. Reordering takes place
238
     * by pulling from each level during IDCT. As a consequence, IDCT must be
239
     * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
240
     * otherwise. The 32 different tokens with up to 12 bits of extradata are
241
     * collapsed into 3 types, packed as follows:
242
     *   (from the low to high bits)
243
     *
244
     * 2 bits: type (0,1,2)
245
     *   0: EOB run, 14 bits for run length (12 needed)
246
     *   1: zero run, 7 bits for run length
247
     *                7 bits for the next coefficient (3 needed)
248
     *   2: coefficient, 14 bits (11 needed)
249
     *
250
     * Coefficients are signed, so are packed in the highest bits for automatic
251
     * sign extension.
252
     */
253
    int16_t *dct_tokens[3][64];
254
    int16_t *dct_tokens_base;
255
#define TOKEN_EOB(eob_run)              ((eob_run) << 2)
256
#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) * 512) + ((zero_run) << 2) + 1)
257
#define TOKEN_COEFF(coeff)              (((coeff) * 4) + 2)
258
259
    /**
260
     * number of blocks that contain DCT coefficients at
261
     * the given level or higher
262
     */
263
    int num_coded_frags[3][64];
264
    int total_num_coded_frags;
265
266
    /* this is a list of indexes into the all_fragments array indicating
267
     * which of the fragments are coded */
268
    int *coded_fragment_list[3];
269
270
    int *kf_coded_fragment_list;
271
    int *nkf_coded_fragment_list;
272
    int num_kf_coded_fragment[3];
273
274
    /* The first 16 of the following VLCs are for the dc coefficients;
275
       the others are four groups of 16 VLCs each for ac coefficients. */
276
    VLC coeff_vlc[5 * 16];
277
278
    VLC superblock_run_length_vlc; /* version < 2 */
279
    VLC fragment_run_length_vlc; /* version < 2 */
280
    VLC block_pattern_vlc[2]; /* version >= 2*/
281
    VLC mode_code_vlc;
282
    VLC motion_vector_vlc; /* version < 2 */
283
    VLC vp4_mv_vlc[2][7]; /* version >=2 */
284
285
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
286
     * index into them */
287
    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     ///< qmat[qpi][is_inter][plane]
288
289
    /* This table contains superblock_count * 16 entries. Each set of 16
290
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
291
     * An entry will be -1 to indicate that no entry corresponds to that
292
     * index. */
293
    int *superblock_fragments;
294
295
    /* This is an array that indicates how a particular macroblock
296
     * is coded. */
297
    unsigned char *macroblock_coding;
298
299
    uint8_t *edge_emu_buffer;
300
301
    /* Huffman decode */
302
    HuffTable huffman_table[5 * 16];
303
304
    uint8_t filter_limit_values[64];
305
    DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
306
307
    VP4Predictor * dc_pred_row; /* dc_pred_row[y_superblock_width * 4] */
308
} Vp3DecodeContext;
309
310
/************************************************************************
311
 * VP3 specific functions
312
 ************************************************************************/
313
314
24
static av_cold void free_tables(AVCodecContext *avctx)
315
{
316
24
    Vp3DecodeContext *s = avctx->priv_data;
317
318
24
    av_freep(&s->superblock_coding);
319
24
    av_freep(&s->all_fragments);
320
24
    av_freep(&s->nkf_coded_fragment_list);
321
24
    av_freep(&s->kf_coded_fragment_list);
322
24
    av_freep(&s->dct_tokens_base);
323
24
    av_freep(&s->superblock_fragments);
324
24
    av_freep(&s->macroblock_coding);
325
24
    av_freep(&s->dc_pred_row);
326
24
    av_freep(&s->motion_val[0]);
327
24
    av_freep(&s->motion_val[1]);
328
24
}
329
330
12
static void vp3_decode_flush(AVCodecContext *avctx)
331
{
332
12
    Vp3DecodeContext *s = avctx->priv_data;
333
334
12
    if (s->golden_frame.f)
335
12
        ff_thread_release_buffer(avctx, &s->golden_frame);
336
12
    if (s->last_frame.f)
337
12
        ff_thread_release_buffer(avctx, &s->last_frame);
338
12
    if (s->current_frame.f)
339
12
        ff_thread_release_buffer(avctx, &s->current_frame);
340
12
}
341
342
12
static av_cold int vp3_decode_end(AVCodecContext *avctx)
343
{
344
12
    Vp3DecodeContext *s = avctx->priv_data;
345
    int i, j;
346
347
12
    free_tables(avctx);
348
12
    av_freep(&s->edge_emu_buffer);
349
350
12
    s->theora_tables = 0;
351
352
    /* release all frames */
353
12
    vp3_decode_flush(avctx);
354
12
    av_frame_free(&s->current_frame.f);
355
12
    av_frame_free(&s->last_frame.f);
356
12
    av_frame_free(&s->golden_frame.f);
357
358
972
    for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++)
359
960
        ff_free_vlc(&s->coeff_vlc[i]);
360
361
12
    ff_free_vlc(&s->superblock_run_length_vlc);
362
12
    ff_free_vlc(&s->fragment_run_length_vlc);
363
12
    ff_free_vlc(&s->mode_code_vlc);
364
12
    ff_free_vlc(&s->motion_vector_vlc);
365
366
36
    for (j = 0; j < 2; j++)
367
192
        for (i = 0; i < 7; i++)
368
168
            ff_free_vlc(&s->vp4_mv_vlc[j][i]);
369
370
36
    for (i = 0; i < 2; i++)
371
24
        ff_free_vlc(&s->block_pattern_vlc[i]);
372
12
    return 0;
373
}
374
375
/**
376
 * This function sets up all of the various blocks mappings:
377
 * superblocks <-> fragments, macroblocks <-> fragments,
378
 * superblocks <-> macroblocks
379
 *
380
 * @return 0 is successful; returns 1 if *anything* went wrong.
381
 */
382
12
static int init_block_mapping(Vp3DecodeContext *s)
383
{
384
    int sb_x, sb_y, plane;
385
12
    int x, y, i, j = 0;
386
387
48
    for (plane = 0; plane < 3; plane++) {
388
36
        int sb_width    = plane ? s->c_superblock_width
389
36
                                : s->y_superblock_width;
390
36
        int sb_height   = plane ? s->c_superblock_height
391
36
                                : s->y_superblock_height;
392
36
        int frag_width  = s->fragment_width[!!plane];
393
36
        int frag_height = s->fragment_height[!!plane];
394
395
551
        for (sb_y = 0; sb_y < sb_height; sb_y++)
396
26730
            for (sb_x = 0; sb_x < sb_width; sb_x++)
397
445655
                for (i = 0; i < 16; i++) {
398
419440
                    x = 4 * sb_x + hilbert_offset[i][0];
399
419440
                    y = 4 * sb_y + hilbert_offset[i][1];
400
401

419440
                    if (x < frag_width && y < frag_height)
402
404850
                        s->superblock_fragments[j++] = s->fragment_start[plane] +
403
404850
                                                       y * frag_width + x;
404
                    else
405
14590
                        s->superblock_fragments[j++] = -1;
406
                }
407
    }
408
409
12
    return 0;  /* successful path out */
410
}
411
412
/*
413
 * This function sets up the dequantization tables used for a particular
414
 * frame.
415
 */
416
62
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
417
{
418
62
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
419
    int i, plane, inter, qri, bmi, bmj, qistart;
420
421
186
    for (inter = 0; inter < 2; inter++) {
422
496
        for (plane = 0; plane < 3; plane++) {
423
372
            int dc_scale_factor = s->coded_dc_scale_factor[!!plane][s->qps[qpi]];
424
372
            int sum = 0;
425
408
            for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
426
408
                sum += s->qr_size[inter][plane][qri];
427
408
                if (s->qps[qpi] <= sum)
428
372
                    break;
429
            }
430
372
            qistart = sum - s->qr_size[inter][plane][qri];
431
372
            bmi     = s->qr_base[inter][plane][qri];
432
372
            bmj     = s->qr_base[inter][plane][qri + 1];
433
24180
            for (i = 0; i < 64; i++) {
434
23808
                int coeff = (2 * (sum     - s->qps[qpi]) * s->base_matrix[bmi][i] -
435
23808
                             2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
436
23808
                             s->qr_size[inter][plane][qri]) /
437
23808
                            (2 * s->qr_size[inter][plane][qri]);
438
439
23808
                int qmin   = 8 << (inter + !i);
440
23808
                int qscale = i ? ac_scale_factor : dc_scale_factor;
441
23808
                int qbias = (1 + inter) * 3;
442
47616
                s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
443
23808
                    (i == 0 || s->version < 2) ? av_clip((qscale * coeff) / 100 * 4, qmin, 4096)
444
1512
                                               : (qscale * (coeff - qbias) / 100 + qbias) * 4;
445
            }
446
            /* all DC coefficients use the same quant so as not to interfere
447
             * with DC prediction */
448
372
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
449
        }
450
    }
451
62
}
452
453
/*
454
 * This function initializes the loop filter boundary limits if the frame's
455
 * quality index is different from the previous frame's.
456
 *
457
 * The filter_limit_values may not be larger than 127.
458
 */
459
62
static void init_loop_filter(Vp3DecodeContext *s)
460
{
461
62
    ff_vp3dsp_set_bounding_values(s->bounding_values_array, s->filter_limit_values[s->qps[0]]);
462
62
}
463
464
/*
465
 * This function unpacks all of the superblock/macroblock/fragment coding
466
 * information from the bitstream.
467
 */
468
140
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
469
{
470
140
    int superblock_starts[3] = {
471
140
        0, s->u_superblock_start, s->v_superblock_start
472
    };
473
140
    int bit = 0;
474
140
    int current_superblock = 0;
475
140
    int current_run = 0;
476
140
    int num_partial_superblocks = 0;
477
478
    int i, j;
479
    int current_fragment;
480
    int plane;
481
140
    int plane0_num_coded_frags = 0;
482
483
140
    if (s->keyframe) {
484
4
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
485
    } else {
486
        /* unpack the list of partially-coded superblocks */
487
136
        bit         = get_bits1(gb) ^ 1;
488
136
        current_run = 0;
489
490

8681
        while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
491

8545
            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
492
                bit = get_bits1(gb);
493
            else
494
8545
                bit ^= 1;
495
496
8545
            current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
497
                                   SUPERBLOCK_VLC_BITS, 2);
498
8545
            if (current_run == 34)
499
437
                current_run += get_bits(gb, 12);
500
501
8545
            if (current_run > s->superblock_count - current_superblock) {
502
                av_log(s->avctx, AV_LOG_ERROR,
503
                       "Invalid partially coded superblock run length\n");
504
                return -1;
505
            }
506
507
8545
            memset(s->superblock_coding + current_superblock, bit, current_run);
508
509
8545
            current_superblock += current_run;
510
8545
            if (bit)
511
4237
                num_partial_superblocks += current_run;
512
        }
513
514
        /* unpack the list of fully coded superblocks if any of the blocks were
515
         * not marked as partially coded in the previous step */
516
136
        if (num_partial_superblocks < s->superblock_count) {
517
136
            int superblocks_decoded = 0;
518
519
136
            current_superblock = 0;
520
136
            bit                = get_bits1(gb) ^ 1;
521
136
            current_run        = 0;
522
523

2240
            while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
524
1052
                   get_bits_left(gb) > 0) {
525

1052
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
526
6
                    bit = get_bits1(gb);
527
                else
528
1046
                    bit ^= 1;
529
530
1052
                current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
531
                                       SUPERBLOCK_VLC_BITS, 2);
532
1052
                if (current_run == 34)
533
218
                    current_run += get_bits(gb, 12);
534
535
105952
                for (j = 0; j < current_run; current_superblock++) {
536
104900
                    if (current_superblock >= s->superblock_count) {
537
                        av_log(s->avctx, AV_LOG_ERROR,
538
                               "Invalid fully coded superblock run length\n");
539
                        return -1;
540
                    }
541
542
                    /* skip any superblocks already marked as partially coded */
543
104900
                    if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
544
88554
                        s->superblock_coding[current_superblock] = 2 * bit;
545
88554
                        j++;
546
                    }
547
                }
548
1052
                superblocks_decoded += current_run;
549
            }
550
        }
551
552
        /* if there were partial blocks, initialize bitstream for
553
         * unpacking fragment codings */
554
136
        if (num_partial_superblocks) {
555
128
            current_run = 0;
556
128
            bit         = get_bits1(gb);
557
            /* toggle the bit because as soon as the first run length is
558
             * fetched the bit will be toggled again */
559
128
            bit ^= 1;
560
        }
561
    }
562
563
    /* figure out which fragments are coded; iterate through each
564
     * superblock (all planes) */
565
140
    s->total_num_coded_frags = 0;
566
140
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
567
568
280
    s->coded_fragment_list[0] = s->keyframe ? s->kf_coded_fragment_list
569
140
                                            : s->nkf_coded_fragment_list;
570
571
560
    for (plane = 0; plane < 3; plane++) {
572
420
        int sb_start = superblock_starts[plane];
573
420
        int sb_end   = sb_start + (plane ? s->c_superblock_count
574
420
                                         : s->y_superblock_count);
575
420
        int num_coded_frags = 0;
576
577
420
        if (s->keyframe) {
578
12
            if (s->num_kf_coded_fragment[plane] == -1) {
579
10538
                for (i = sb_start; i < sb_end; i++) {
580
                    /* iterate through all 16 fragments in a superblock */
581
178942
                    for (j = 0; j < 16; j++) {
582
                        /* if the fragment is in bounds, check its coding status */
583
168416
                        current_fragment = s->superblock_fragments[i * 16 + j];
584
168416
                        if (current_fragment != -1) {
585
162492
                            s->coded_fragment_list[plane][num_coded_frags++] =
586
                                current_fragment;
587
                        }
588
                    }
589
                }
590
12
                s->num_kf_coded_fragment[plane] = num_coded_frags;
591
            } else
592
                num_coded_frags = s->num_kf_coded_fragment[plane];
593
        } else {
594

105594
            for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
595
105186
                if (get_bits_left(gb) < plane0_num_coded_frags >> 2) {
596
                    return AVERROR_INVALIDDATA;
597
                }
598
                /* iterate through all 16 fragments in a superblock */
599
1788162
                for (j = 0; j < 16; j++) {
600
                    /* if the fragment is in bounds, check its coding status */
601
1682976
                    current_fragment = s->superblock_fragments[i * 16 + j];
602
1682976
                    if (current_fragment != -1) {
603
1598148
                        int coded = s->superblock_coding[i];
604
605
1598148
                        if (coded == SB_PARTIALLY_CODED) {
606
                            /* fragment may or may not be coded; this is the case
607
                             * that cares about the fragment coding runs */
608
241844
                            if (current_run-- == 0) {
609
70601
                                bit        ^= 1;
610
70601
                                current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2);
611
                            }
612
241844
                            coded = bit;
613
                        }
614
615
1598148
                        if (coded) {
616
                            /* default mode; actual mode will be decoded in
617
                             * the next phase */
618
370084
                            s->all_fragments[current_fragment].coding_method =
619
                                MODE_INTER_NO_MV;
620
370084
                            s->coded_fragment_list[plane][num_coded_frags++] =
621
                                current_fragment;
622
                        } else {
623
                            /* not coded; copy this fragment from the prior frame */
624
1228064
                            s->all_fragments[current_fragment].coding_method =
625
                                MODE_COPY;
626
                        }
627
                    }
628
                }
629
            }
630
        }
631
420
        if (!plane)
632
140
            plane0_num_coded_frags = num_coded_frags;
633
420
        s->total_num_coded_frags += num_coded_frags;
634
27300
        for (i = 0; i < 64; i++)
635
26880
            s->num_coded_frags[plane][i] = num_coded_frags;
636
420
        if (plane < 2)
637
280
            s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] +
638
                                                num_coded_frags;
639
    }
640
140
    return 0;
641
}
642
643
#define BLOCK_X (2 * mb_x + (k & 1))
644
#define BLOCK_Y (2 * mb_y + (k >> 1))
645
646
#if CONFIG_VP4_DECODER
647
/**
648
 * @return number of blocks, or > yuv_macroblock_count on error.
649
 *         return value is always >= 1.
650
 */
651
7007
static int vp4_get_mb_count(Vp3DecodeContext *s, GetBitContext *gb)
652
{
653
7007
    int v = 1;
654
    int bits;
655
7007
    while ((bits = show_bits(gb, 9)) == 0x1ff) {
656
        skip_bits(gb, 9);
657
        v += 256;
658
        if (v > s->yuv_macroblock_count) {
659
            av_log(s->avctx, AV_LOG_ERROR, "Invalid run length\n");
660
            return v;
661
        }
662
    }
663
#define body(n) { \
664
    skip_bits(gb, 2 + n); \
665
    v += (1 << n) + get_bits(gb, n); }
666
#define thresh(n) (0x200 - (0x80 >> n))
667
#define else_if(n) else if (bits < thresh(n)) body(n)
668
7007
    if (bits < 0x100) {
669
2784
        skip_bits(gb, 1);
670
4223
    } else if (bits < thresh(0)) {
671
1763
        skip_bits(gb, 2);
672
1763
        v += 1;
673
    }
674
2460
    else_if(1)
675
1466
    else_if(2)
676
799
    else_if(3)
677
316
    else_if(4)
678
92
    else_if(5)
679
4
    else_if(6)
680
    else body(7)
681
#undef body
682
#undef thresh
683
#undef else_if
684
7007
    return v;
685
}
686
687
4556
static int vp4_get_block_pattern(Vp3DecodeContext *s, GetBitContext *gb, int *next_block_pattern_table)
688
{
689
4556
    int v = get_vlc2(gb, s->block_pattern_vlc[*next_block_pattern_table].table, 3, 2);
690
4556
    *next_block_pattern_table = vp4_block_pattern_table_selector[v];
691
4556
    return v + 1;
692
}
693
694
24
static int vp4_unpack_macroblocks(Vp3DecodeContext *s, GetBitContext *gb)
695
{
696
    int plane, i, j, k, fragment;
697
    int next_block_pattern_table;
698
    int bit, current_run, has_partial;
699
700
24
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
701
702
24
    if (s->keyframe)
703
2
        return 0;
704
705
22
    has_partial = 0;
706
22
    bit         = get_bits1(gb);
707
4154
    for (i = 0; i < s->yuv_macroblock_count; i += current_run) {
708
4132
        if (get_bits_left(gb) <= 0)
709
            return AVERROR_INVALIDDATA;
710
4132
        current_run = vp4_get_mb_count(s, gb);
711
4132
        if (current_run > s->yuv_macroblock_count - i)
712
            return -1;
713
4132
        memset(s->superblock_coding + i, 2 * bit, current_run);
714
4132
        bit ^= 1;
715
4132
        has_partial |= bit;
716
    }
717
718
22
    if (has_partial) {
719
22
        if (get_bits_left(gb) <= 0)
720
            return AVERROR_INVALIDDATA;
721
22
        bit  = get_bits1(gb);
722
22
        current_run = vp4_get_mb_count(s, gb);
723
20086
        for (i = 0; i < s->yuv_macroblock_count; i++) {
724
20064
            if (!s->superblock_coding[i]) {
725
8403
                if (!current_run) {
726
2853
                    bit ^= 1;
727
2853
                    current_run = vp4_get_mb_count(s, gb);
728
                }
729
8403
                s->superblock_coding[i] = bit;
730
8403
                current_run--;
731
            }
732
        }
733
22
        if (current_run) /* handle situation when vp4_get_mb_count() fails */
734
            return -1;
735
    }
736
737
22
    next_block_pattern_table = 0;
738
22
    i = 0;
739
88
    for (plane = 0; plane < 3; plane++) {
740
        int sb_x, sb_y;
741
66
        int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
742
66
        int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
743
66
        int mb_width = plane ? s->c_macroblock_width : s->macroblock_width;
744
66
        int mb_height = plane ? s->c_macroblock_height : s->macroblock_height;
745
66
        int fragment_width = s->fragment_width[!!plane];
746
66
        int fragment_height = s->fragment_height[!!plane];
747
748
418
        for (sb_y = 0; sb_y < sb_height; sb_y++) {
749
5456
            for (sb_x = 0; sb_x < sb_width; sb_x++) {
750
25520
                for (j = 0; j < 4; j++) {
751
20416
                    int mb_x = 2 * sb_x + (j >> 1);
752
20416
                    int mb_y = 2 * sb_y + (j >> 1) ^ (j & 1);
753
                    int mb_coded, pattern, coded;
754
755

20416
                    if (mb_x >= mb_width || mb_y >= mb_height)
756
352
                        continue;
757
758
20064
                    mb_coded = s->superblock_coding[i++];
759
760
20064
                    if (mb_coded == SB_FULLY_CODED)
761
11661
                        pattern = 0xF;
762
8403
                    else if (mb_coded == SB_PARTIALLY_CODED)
763
4556
                        pattern = vp4_get_block_pattern(s, gb, &next_block_pattern_table);
764
                    else
765
3847
                        pattern = 0;
766
767
100320
                    for (k = 0; k < 4; k++) {
768

80256
                        if (BLOCK_X >= fragment_width || BLOCK_Y >= fragment_height)
769
                            continue;
770
80256
                        fragment = s->fragment_start[plane] + BLOCK_Y * fragment_width + BLOCK_X;
771
80256
                        coded = pattern & (8 >> k);
772
                        /* MODE_INTER_NO_MV is the default for coded fragments.
773
                           the actual method is decoded in the next phase. */
774
80256
                        s->all_fragments[fragment].coding_method = coded ? MODE_INTER_NO_MV : MODE_COPY;
775
                    }
776
                }
777
            }
778
        }
779
    }
780
22
    return 0;
781
}
782
#endif
783
784
/*
785
 * This function unpacks all the coding mode data for individual macroblocks
786
 * from the bitstream.
787
 */
788
164
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
789
{
790
    int i, j, k, sb_x, sb_y;
791
    int scheme;
792
    int current_macroblock;
793
    int current_fragment;
794
    int coding_mode;
795
    int custom_mode_alphabet[CODING_MODE_COUNT];
796
    const int *alphabet;
797
    Vp3Fragment *frag;
798
799
164
    if (s->keyframe) {
800
169794
        for (i = 0; i < s->fragment_count; i++)
801
169788
            s->all_fragments[i].coding_method = MODE_INTRA;
802
    } else {
803
        /* fetch the mode coding scheme for this frame */
804
158
        scheme = get_bits(gb, 3);
805
806
        /* is it a custom coding scheme? */
807
158
        if (scheme == 0) {
808
207
            for (i = 0; i < 8; i++)
809
184
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
810
207
            for (i = 0; i < 8; i++)
811
184
                custom_mode_alphabet[get_bits(gb, 3)] = i;
812
23
            alphabet = custom_mode_alphabet;
813
        } else
814
135
            alphabet = ModeAlphabet[scheme - 1];
815
816
        /* iterate through all of the macroblocks that contain 1 or more
817
         * coded fragments */
818
2038
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
819
73930
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
820
72050
                if (get_bits_left(gb) <= 0)
821
                    return -1;
822
823
360250
                for (j = 0; j < 4; j++) {
824
288200
                    int mb_x = 2 * sb_x + (j >> 1);
825
288200
                    int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
826
288200
                    current_macroblock = mb_y * s->macroblock_width + mb_x;
827
828
288200
                    if (mb_x >= s->macroblock_width ||
829
287276
                        mb_y >= s->macroblock_height)
830
8466
                        continue;
831
832
                    /* coding modes are only stored if the macroblock has
833
                     * at least one luma block coded, otherwise it must be
834
                     * INTER_NO_MV */
835
1073976
                    for (k = 0; k < 4; k++) {
836
878591
                        current_fragment = BLOCK_Y *
837
878591
                                           s->fragment_width[0] + BLOCK_X;
838
878591
                        if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
839
84349
                            break;
840
                    }
841
279734
                    if (k == 4) {
842
195385
                        s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
843
195385
                        continue;
844
                    }
845
846
                    /* mode 7 means get 3 bits for each coding mode */
847
84349
                    if (scheme == 7)
848
2750
                        coding_mode = get_bits(gb, 3);
849
                    else
850
81599
                        coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
851
852
84349
                    s->macroblock_coding[current_macroblock] = coding_mode;
853
421745
                    for (k = 0; k < 4; k++) {
854
337396
                        frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
855
337396
                        if (frag->coding_method != MODE_COPY)
856
307251
                            frag->coding_method = coding_mode;
857
                    }
858
859
#define SET_CHROMA_MODES                                                      \
860
    if (frag[s->fragment_start[1]].coding_method != MODE_COPY)                \
861
        frag[s->fragment_start[1]].coding_method = coding_mode;               \
862
    if (frag[s->fragment_start[2]].coding_method != MODE_COPY)                \
863
        frag[s->fragment_start[2]].coding_method = coding_mode;
864
865
84349
                    if (s->chroma_y_shift) {
866
84349
                        frag = s->all_fragments + mb_y *
867
84349
                               s->fragment_width[1] + mb_x;
868

84349
                        SET_CHROMA_MODES
869
                    } else if (s->chroma_x_shift) {
870
                        frag = s->all_fragments +
871
                               2 * mb_y * s->fragment_width[1] + mb_x;
872
                        for (k = 0; k < 2; k++) {
873
                            SET_CHROMA_MODES
874
                            frag += s->fragment_width[1];
875
                        }
876
                    } else {
877
                        for (k = 0; k < 4; k++) {
878
                            frag = s->all_fragments +
879
                                   BLOCK_Y * s->fragment_width[1] + BLOCK_X;
880
                            SET_CHROMA_MODES
881
                        }
882
                    }
883
                }
884
            }
885
        }
886
    }
887
888
164
    return 0;
889
}
890
891
3758
static int vp4_get_mv(Vp3DecodeContext *s, GetBitContext *gb, int axis, int last_motion)
892
{
893
3758
    int v = get_vlc2(gb, s->vp4_mv_vlc[axis][vp4_mv_table_selector[FFABS(last_motion)]].table,
894
                     VP4_MV_VLC_BITS, 2);
895
3758
    return last_motion < 0 ? -v : v;
896
}
897
898
/*
899
 * This function unpacks all the motion vectors for the individual
900
 * macroblocks from the bitstream.
901
 */
902
164
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
903
{
904
    int j, k, sb_x, sb_y;
905
    int coding_mode;
906
    int motion_x[4];
907
    int motion_y[4];
908
164
    int last_motion_x = 0;
909
164
    int last_motion_y = 0;
910
164
    int prior_last_motion_x = 0;
911
164
    int prior_last_motion_y = 0;
912
164
    int last_gold_motion_x = 0;
913
164
    int last_gold_motion_y = 0;
914
    int current_macroblock;
915
    int current_fragment;
916
    int frag;
917
918
164
    if (s->keyframe)
919
6
        return 0;
920
921
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme; 2 is VP4 code scheme */
922
158
    coding_mode = s->version < 2 ? get_bits1(gb) : 2;
923
924
    /* iterate through all of the macroblocks that contain 1 or more
925
     * coded fragments */
926
2038
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
927
73930
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
928
72050
            if (get_bits_left(gb) <= 0)
929
                return -1;
930
931
360250
            for (j = 0; j < 4; j++) {
932
288200
                int mb_x = 2 * sb_x + (j >> 1);
933
288200
                int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
934
288200
                current_macroblock = mb_y * s->macroblock_width + mb_x;
935
936
288200
                if (mb_x >= s->macroblock_width  ||
937
287276
                    mb_y >= s->macroblock_height ||
938
279734
                    s->macroblock_coding[current_macroblock] == MODE_COPY)
939
8466
                    continue;
940
941

279734
                switch (s->macroblock_coding[current_macroblock]) {
942
711
                case MODE_GOLDEN_MV:
943
711
                    if (coding_mode == 2) { /* VP4 */
944
337
                        last_gold_motion_x = motion_x[0] = vp4_get_mv(s, gb, 0, last_gold_motion_x);
945
337
                        last_gold_motion_y = motion_y[0] = vp4_get_mv(s, gb, 1, last_gold_motion_y);
946
337
                        break;
947
                    } /* otherwise fall through */
948
                case MODE_INTER_PLUS_MV:
949
                    /* all 6 fragments use the same motion vector */
950
14558
                    if (coding_mode == 0) {
951
13074
                        motion_x[0] = get_vlc2(gb, s->motion_vector_vlc.table,
952
                                               VP3_MV_VLC_BITS, 2);
953
13074
                        motion_y[0] = get_vlc2(gb, s->motion_vector_vlc.table,
954
                                               VP3_MV_VLC_BITS, 2);
955
1484
                    } else if (coding_mode == 1) {
956
42
                        motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
957
42
                        motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
958
                    } else { /* VP4 */
959
1442
                        motion_x[0] = vp4_get_mv(s, gb, 0, last_motion_x);
960
1442
                        motion_y[0] = vp4_get_mv(s, gb, 1, last_motion_y);
961
                    }
962
963
                    /* vector maintenance, only on MODE_INTER_PLUS_MV */
964
14558
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
965
14184
                        prior_last_motion_x = last_motion_x;
966
14184
                        prior_last_motion_y = last_motion_y;
967
14184
                        last_motion_x       = motion_x[0];
968
14184
                        last_motion_y       = motion_y[0];
969
                    }
970
14558
                    break;
971
972
63
                case MODE_INTER_FOURMV:
973
                    /* vector maintenance */
974
63
                    prior_last_motion_x = last_motion_x;
975
63
                    prior_last_motion_y = last_motion_y;
976
977
                    /* fetch 4 vectors from the bitstream, one for each
978
                     * Y fragment, then average for the C fragment vectors */
979
315
                    for (k = 0; k < 4; k++) {
980
252
                        current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
981
252
                        if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
982
252
                            if (coding_mode == 0) {
983
48
                                motion_x[k] = get_vlc2(gb, s->motion_vector_vlc.table,
984
                                                       VP3_MV_VLC_BITS, 2);
985
48
                                motion_y[k] = get_vlc2(gb, s->motion_vector_vlc.table,
986
                                                       VP3_MV_VLC_BITS, 2);
987
204
                            } else if (coding_mode == 1) {
988
104
                                motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
989
104
                                motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
990
                            } else { /* VP4 */
991
100
                                motion_x[k] = vp4_get_mv(s, gb, 0, prior_last_motion_x);
992
100
                                motion_y[k] = vp4_get_mv(s, gb, 1, prior_last_motion_y);
993
                            }
994
252
                            last_motion_x = motion_x[k];
995
252
                            last_motion_y = motion_y[k];
996
                        } else {
997
                            motion_x[k] = 0;
998
                            motion_y[k] = 0;
999
                        }
1000
                    }
1001
63
                    break;
1002
1003
42307
                case MODE_INTER_LAST_MV:
1004
                    /* all 6 fragments use the last motion vector */
1005
42307
                    motion_x[0] = last_motion_x;
1006
42307
                    motion_y[0] = last_motion_y;
1007
1008
                    /* no vector maintenance (last vector remains the
1009
                     * last vector) */
1010
42307
                    break;
1011
1012
11934
                case MODE_INTER_PRIOR_LAST:
1013
                    /* all 6 fragments use the motion vector prior to the
1014
                     * last motion vector */
1015
11934
                    motion_x[0] = prior_last_motion_x;
1016
11934
                    motion_y[0] = prior_last_motion_y;
1017
1018
                    /* vector maintenance */
1019
11934
                    prior_last_motion_x = last_motion_x;
1020
11934
                    prior_last_motion_y = last_motion_y;
1021
11934
                    last_motion_x       = motion_x[0];
1022
11934
                    last_motion_y       = motion_y[0];
1023
11934
                    break;
1024
1025
210535
                default:
1026
                    /* covers intra, inter without MV, golden without MV */
1027
210535
                    motion_x[0] = 0;
1028
210535
                    motion_y[0] = 0;
1029
1030
                    /* no vector maintenance */
1031
210535
                    break;
1032
                }
1033
1034
                /* assign the motion vectors to the correct fragments */
1035
1398670
                for (k = 0; k < 4; k++) {
1036
1118936
                    current_fragment =
1037
1118936
                        BLOCK_Y * s->fragment_width[0] + BLOCK_X;
1038
1118936
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1039
252
                        s->motion_val[0][current_fragment][0] = motion_x[k];
1040
252
                        s->motion_val[0][current_fragment][1] = motion_y[k];
1041
                    } else {
1042
1118684
                        s->motion_val[0][current_fragment][0] = motion_x[0];
1043
1118684
                        s->motion_val[0][current_fragment][1] = motion_y[0];
1044
                    }
1045
                }
1046
1047
279734
                if (s->chroma_y_shift) {
1048
279734
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1049
63
                        motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
1050
                                             motion_x[2] + motion_x[3], 2);
1051
63
                        motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
1052
                                             motion_y[2] + motion_y[3], 2);
1053
                    }
1054
279734
                    if (s->version <= 2) {
1055
266358
                        motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1056
266358
                        motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
1057
                    }
1058
279734
                    frag = mb_y * s->fragment_width[1] + mb_x;
1059
279734
                    s->motion_val[1][frag][0] = motion_x[0];
1060
279734
                    s->motion_val[1][frag][1] = motion_y[0];
1061
                } else if (s->chroma_x_shift) {
1062
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1063
                        motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
1064
                        motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
1065
                        motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
1066
                        motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
1067
                    } else {
1068
                        motion_x[1] = motion_x[0];
1069
                        motion_y[1] = motion_y[0];
1070
                    }
1071
                    if (s->version <= 2) {
1072
                        motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1073
                        motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
1074
                    }
1075
                    frag = 2 * mb_y * s->fragment_width[1] + mb_x;
1076
                    for (k = 0; k < 2; k++) {
1077
                        s->motion_val[1][frag][0] = motion_x[k];
1078
                        s->motion_val[1][frag][1] = motion_y[k];
1079
                        frag += s->fragment_width[1];
1080
                    }
1081
                } else {
1082
                    for (k = 0; k < 4; k++) {
1083
                        frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
1084
                        if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1085
                            s->motion_val[1][frag][0] = motion_x[k];
1086
                            s->motion_val[1][frag][1] = motion_y[k];
1087
                        } else {
1088
                            s->motion_val[1][frag][0] = motion_x[0];
1089
                            s->motion_val[1][frag][1] = motion_y[0];
1090
                        }
1091
                    }
1092
                }
1093
            }
1094
        }
1095
    }
1096
1097
158
    return 0;
1098
}
1099
1100
164
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
1101
{
1102
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
1103
164
    int num_blocks = s->total_num_coded_frags;
1104
1105

164
    for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
1106
        i = blocks_decoded = num_blocks_at_qpi = 0;
1107
1108
        bit        = get_bits1(gb) ^ 1;
1109
        run_length = 0;
1110
1111
        do {
1112
            if (run_length == MAXIMUM_LONG_BIT_RUN)
1113
                bit = get_bits1(gb);
1114
            else
1115
                bit ^= 1;
1116
1117
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table,
1118
                                  SUPERBLOCK_VLC_BITS, 2);
1119
            if (run_length == 34)
1120
                run_length += get_bits(gb, 12);
1121
            blocks_decoded += run_length;
1122
1123
            if (!bit)
1124
                num_blocks_at_qpi += run_length;
1125
1126
            for (j = 0; j < run_length; i++) {
1127
                if (i >= s->total_num_coded_frags)
1128
                    return -1;
1129
1130
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
1131
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
1132
                    j++;
1133
                }
1134
            }
1135
        } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
1136
1137
        num_blocks -= num_blocks_at_qpi;
1138
    }
1139
1140
164
    return 0;
1141
}
1142
1143
259052
static inline int get_eob_run(GetBitContext *gb, int token)
1144
{
1145
259052
    int v = eob_run_table[token].base;
1146
259052
    if (eob_run_table[token].bits)
1147
33652
        v += get_bits(gb, eob_run_table[token].bits);
1148
259052
    return v;
1149
}
1150
1151
2755489
static inline int get_coeff(GetBitContext *gb, int token, int16_t *coeff)
1152
{
1153
    int bits_to_get, zero_run;
1154
1155
2755489
    bits_to_get = coeff_get_bits[token];
1156
2755489
    if (bits_to_get)
1157
1617513
        bits_to_get = get_bits(gb, bits_to_get);
1158
2755489
    *coeff = coeff_tables[token][bits_to_get];
1159
1160
2755489
    zero_run = zero_run_base[token];
1161
2755489
    if (zero_run_get_bits[token])
1162
342682
        zero_run += get_bits(gb, zero_run_get_bits[token]);
1163
1164
2755489
    return zero_run;
1165
}
1166
1167
/*
1168
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1169
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1170
 * data. This function unpacks all the VLCs for either the Y plane or both
1171
 * C planes, and is called for DC coefficients or different AC coefficient
1172
 * levels (since different coefficient types require different VLC tables.
1173
 *
1174
 * This function returns a residual eob run. E.g, if a particular token gave
1175
 * instructions to EOB the next 5 fragments and there were only 2 fragments
1176
 * left in the current fragment range, 3 would be returned so that it could
1177
 * be passed into the next call to this same function.
1178
 */
1179
26880
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1180
                       VLC *table, int coeff_index,
1181
                       int plane,
1182
                       int eob_run)
1183
{
1184
26880
    int i, j = 0;
1185
    int token;
1186
26880
    int zero_run  = 0;
1187
26880
    int16_t coeff = 0;
1188
    int blocks_ended;
1189
26880
    int coeff_i = 0;
1190
26880
    int num_coeffs      = s->num_coded_frags[plane][coeff_index];
1191
26880
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
1192
1193
    /* local references to structure members to avoid repeated dereferences */
1194
26880
    int *coded_fragment_list   = s->coded_fragment_list[plane];
1195
26880
    Vp3Fragment *all_fragments = s->all_fragments;
1196
26880
    VLC_TYPE(*vlc_table)[2] = table->table;
1197
1198
26880
    if (num_coeffs < 0) {
1199
        av_log(s->avctx, AV_LOG_ERROR,
1200
               "Invalid number of coefficients at level %d\n", coeff_index);
1201
        return AVERROR_INVALIDDATA;
1202
    }
1203
1204
26880
    if (eob_run > num_coeffs) {
1205
3487
        coeff_i      =
1206
3487
        blocks_ended = num_coeffs;
1207
3487
        eob_run     -= num_coeffs;
1208
    } else {
1209
23393
        coeff_i      =
1210
23393
        blocks_ended = eob_run;
1211
23393
        eob_run      = 0;
1212
    }
1213
1214
    // insert fake EOB token to cover the split between planes or zzi
1215
26880
    if (blocks_ended)
1216
2019
        dct_tokens[j++] = blocks_ended << 2;
1217
1218

2087120
    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
1219
        /* decode a VLC into a token */
1220
2060240
        token = get_vlc2(gb, vlc_table, 11, 3);
1221
        /* use the token to get a zero run, a coefficient, and an eob run */
1222
2060240
        if ((unsigned) token <= 6U) {
1223
220061
            eob_run = get_eob_run(gb, token);
1224
220061
            if (!eob_run)
1225
                eob_run = INT_MAX;
1226
1227
            // record only the number of blocks ended in this plane,
1228
            // any spill will be recorded in the next plane.
1229
220061
            if (eob_run > num_coeffs - coeff_i) {
1230
1656
                dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
1231
1656
                blocks_ended   += num_coeffs - coeff_i;
1232
1656
                eob_run        -= num_coeffs - coeff_i;
1233
1656
                coeff_i         = num_coeffs;
1234
            } else {
1235
218405
                dct_tokens[j++] = TOKEN_EOB(eob_run);
1236
218405
                blocks_ended   += eob_run;
1237
218405
                coeff_i        += eob_run;
1238
218405
                eob_run         = 0;
1239
            }
1240
1840179
        } else if (token >= 0) {
1241
1840179
            zero_run = get_coeff(gb, token, &coeff);
1242
1243
1840179
            if (zero_run) {
1244
601916
                dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
1245
            } else {
1246
                // Save DC into the fragment structure. DC prediction is
1247
                // done in raster order, so the actual DC can't be in with
1248
                // other tokens. We still need the token in dct_tokens[]
1249
                // however, or else the structure collapses on itself.
1250
1238263
                if (!coeff_index)
1251
241967
                    all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
1252
1253
1238263
                dct_tokens[j++] = TOKEN_COEFF(coeff);
1254
            }
1255
1256
1840179
            if (coeff_index + zero_run > 64) {
1257
                av_log(s->avctx, AV_LOG_DEBUG,
1258
                       "Invalid zero run of %d with %d coeffs left\n",
1259
                       zero_run, 64 - coeff_index);
1260
                zero_run = 64 - coeff_index;
1261
            }
1262
1263
            // zero runs code multiple coefficients,
1264
            // so don't try to decode coeffs for those higher levels
1265
4381268
            for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1266
2541089
                s->num_coded_frags[plane][i]--;
1267
1840179
            coeff_i++;
1268
        } else {
1269
            av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1270
            return -1;
1271
        }
1272
    }
1273
1274
26880
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1275
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1276
1277
    // decrement the number of blocks that have higher coefficients for each
1278
    // EOB run at this level
1279
26880
    if (blocks_ended)
1280
381061
        for (i = coeff_index + 1; i < 64; i++)
1281
372769
            s->num_coded_frags[plane][i] -= blocks_ended;
1282
1283
    // setup the next buffer
1284
26880
    if (plane < 2)
1285
17920
        s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1286
8960
    else if (coeff_index < 63)
1287
8820
        s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1288
1289
26880
    return eob_run;
1290
}
1291
1292
static void reverse_dc_prediction(Vp3DecodeContext *s,
1293
                                  int first_fragment,
1294
                                  int fragment_width,
1295
                                  int fragment_height);
1296
/*
1297
 * This function unpacks all of the DCT coefficient data from the
1298
 * bitstream.
1299
 */
1300
140
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1301
{
1302
    int i;
1303
    int dc_y_table;
1304
    int dc_c_table;
1305
    int ac_y_table;
1306
    int ac_c_table;
1307
140
    int residual_eob_run = 0;
1308
    VLC *y_tables[64];
1309
    VLC *c_tables[64];
1310
1311
140
    s->dct_tokens[0][0] = s->dct_tokens_base;
1312
1313
140
    if (get_bits_left(gb) < 16)
1314
        return AVERROR_INVALIDDATA;
1315
1316
    /* fetch the DC table indexes */
1317
140
    dc_y_table = get_bits(gb, 4);
1318
140
    dc_c_table = get_bits(gb, 4);
1319
1320
    /* unpack the Y plane DC coefficients */
1321
140
    residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_y_table], 0,
1322
                                   0, residual_eob_run);
1323
140
    if (residual_eob_run < 0)
1324
        return residual_eob_run;
1325
140
    if (get_bits_left(gb) < 8)
1326
        return AVERROR_INVALIDDATA;
1327
1328
    /* reverse prediction of the Y-plane DC coefficients */
1329
140
    reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1330
1331
    /* unpack the C plane DC coefficients */
1332
140
    residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_c_table], 0,
1333
                                   1, residual_eob_run);
1334
140
    if (residual_eob_run < 0)
1335
        return residual_eob_run;
1336
140
    residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_c_table], 0,
1337
                                   2, residual_eob_run);
1338
140
    if (residual_eob_run < 0)
1339
        return residual_eob_run;
1340
1341
    /* reverse prediction of the C-plane DC coefficients */
1342
140
    if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1343
140
        reverse_dc_prediction(s, s->fragment_start[1],
1344
                              s->fragment_width[1], s->fragment_height[1]);
1345
140
        reverse_dc_prediction(s, s->fragment_start[2],
1346
                              s->fragment_width[1], s->fragment_height[1]);
1347
    }
1348
1349
140
    if (get_bits_left(gb) < 8)
1350
        return AVERROR_INVALIDDATA;
1351
    /* fetch the AC table indexes */
1352
140
    ac_y_table = get_bits(gb, 4);
1353
140
    ac_c_table = get_bits(gb, 4);
1354
1355
    /* build tables of AC VLC tables */
1356
840
    for (i = 1; i <= 5; i++) {
1357
        /* AC VLC table group 1 */
1358
700
        y_tables[i] = &s->coeff_vlc[ac_y_table + 16];
1359
700
        c_tables[i] = &s->coeff_vlc[ac_c_table + 16];
1360
    }
1361
1400
    for (i = 6; i <= 14; i++) {
1362
        /* AC VLC table group 2 */
1363
1260
        y_tables[i] = &s->coeff_vlc[ac_y_table + 32];
1364
1260
        c_tables[i] = &s->coeff_vlc[ac_c_table + 32];
1365
    }
1366
1960
    for (i = 15; i <= 27; i++) {
1367
        /* AC VLC table group 3 */
1368
1820
        y_tables[i] = &s->coeff_vlc[ac_y_table + 48];
1369
1820
        c_tables[i] = &s->coeff_vlc[ac_c_table + 48];
1370
    }
1371
5180
    for (i = 28; i <= 63; i++) {
1372
        /* AC VLC table group 4 */
1373
5040
        y_tables[i] = &s->coeff_vlc[ac_y_table + 64];
1374
5040
        c_tables[i] = &s->coeff_vlc[ac_c_table + 64];
1375
    }
1376
1377
    /* decode all AC coefficients */
1378
8960
    for (i = 1; i <= 63; i++) {
1379
8820
        residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1380
                                       0, residual_eob_run);
1381
8820
        if (residual_eob_run < 0)
1382
            return residual_eob_run;
1383
1384
8820
        residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1385
                                       1, residual_eob_run);
1386
8820
        if (residual_eob_run < 0)
1387
            return residual_eob_run;
1388
8820
        residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1389
                                       2, residual_eob_run);
1390
8820
        if (residual_eob_run < 0)
1391
            return residual_eob_run;
1392
    }
1393
1394
140
    return 0;
1395
}
1396
1397
#if CONFIG_VP4_DECODER
1398
/**
1399
 * eob_tracker[] is instead of TOKEN_EOB(value)
1400
 * a dummy TOKEN_EOB(0) value is used to make vp3_dequant work
1401
 *
1402
 * @return < 0 on error
1403
 */
1404
63252
static int vp4_unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1405
                       VLC *vlc_tables[64],
1406
                       int plane, int eob_tracker[64], int fragment)
1407
{
1408
    int token;
1409
63252
    int zero_run  = 0;
1410
63252
    int16_t coeff = 0;
1411
63252
    int coeff_i = 0;
1412
    int eob_run;
1413
1414
973472
    while (!eob_tracker[coeff_i]) {
1415
954301
        if (get_bits_left(gb) < 1)
1416
            return AVERROR_INVALIDDATA;
1417
1418
954301
        token = get_vlc2(gb, vlc_tables[coeff_i]->table, 11, 3);
1419
1420
        /* use the token to get a zero run, a coefficient, and an eob run */
1421
954301
        if ((unsigned) token <= 6U) {
1422
38991
            eob_run = get_eob_run(gb, token);
1423
38991
            *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1424
38991
            eob_tracker[coeff_i] = eob_run - 1;
1425
38991
            return 0;
1426
915310
        } else if (token >= 0) {
1427
915310
            zero_run = get_coeff(gb, token, &coeff);
1428
1429
915310
            if (zero_run) {
1430
311952
                if (coeff_i + zero_run > 64) {
1431
                    av_log(s->avctx, AV_LOG_DEBUG,
1432
                        "Invalid zero run of %d with %d coeffs left\n",
1433
                        zero_run, 64 - coeff_i);
1434
                    zero_run = 64 - coeff_i;
1435
                }
1436
311952
                *s->dct_tokens[plane][coeff_i]++ = TOKEN_ZERO_RUN(coeff, zero_run);
1437
311952
                coeff_i += zero_run;
1438
            } else {
1439
603358
                if (!coeff_i)
1440
40795
                    s->all_fragments[fragment].dc = coeff;
1441
1442
603358
                *s->dct_tokens[plane][coeff_i]++ = TOKEN_COEFF(coeff);
1443
            }
1444
915310
            coeff_i++;
1445
915310
            if (coeff_i >= 64) /* > 64 occurs when there is a zero_run overflow */
1446
5090
                return 0; /* stop */
1447
        } else {
1448
            av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1449
            return -1;
1450
        }
1451
    }
1452
19171
    *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1453
19171
    eob_tracker[coeff_i]--;
1454
19171
    return 0;
1455
}
1456
1457
95328
static void vp4_dc_predictor_reset(VP4Predictor *p)
1458
{
1459
95328
    p->dc = 0;
1460
95328
    p->type = VP4_DC_UNDEFINED;
1461
95328
}
1462
1463
5568
static void vp4_dc_pred_before(const Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1464
{
1465
    int i, j;
1466
1467
27840
    for (i = 0; i < 4; i++)
1468
22272
        dc_pred[0][i + 1] = s->dc_pred_row[sb_x * 4 + i];
1469
1470
27840
    for (j = 1; j < 5; j++)
1471
111360
        for (i = 0; i < 4; i++)
1472
89088
            vp4_dc_predictor_reset(&dc_pred[j][i + 1]);
1473
5568
}
1474
1475
5568
static void vp4_dc_pred_after(Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1476
{
1477
    int i;
1478
1479
27840
    for (i = 0; i < 4; i++)
1480
22272
        s->dc_pred_row[sb_x * 4 + i] = dc_pred[4][i + 1];
1481
1482
27840
    for (i = 1; i < 5; i++)
1483
22272
        dc_pred[i][0] = dc_pred[i][4];
1484
5568
}
1485
1486
/* note: dc_pred points to the current block */
1487
63252
static int vp4_dc_pred(const Vp3DecodeContext *s, const VP4Predictor * dc_pred, const int * last_dc, int type, int plane)
1488
{
1489
63252
    int count = 0;
1490
63252
    int dc = 0;
1491
1492
63252
    if (dc_pred[-6].type == type) {
1493
29726
        dc += dc_pred[-6].dc;
1494
29726
        count++;
1495
    }
1496
1497
63252
    if (dc_pred[6].type == type) {
1498
19581
        dc += dc_pred[6].dc;
1499
19581
        count++;
1500
    }
1501
1502

63252
    if (count != 2 && dc_pred[-1].type == type) {
1503
38121
        dc += dc_pred[-1].dc;
1504
38121
        count++;
1505
    }
1506
1507

63252
    if (count != 2 && dc_pred[1].type == type) {
1508
2073
        dc += dc_pred[1].dc;
1509
2073
        count++;
1510
    }
1511
1512
    /* using division instead of shift to correctly handle negative values */
1513
63252
    return count == 2 ? dc / 2 : last_dc[type];
1514
}
1515
1516
48
static void vp4_set_tokens_base(Vp3DecodeContext *s)
1517
{
1518
    int plane, i;
1519
48
    int16_t *base = s->dct_tokens_base;
1520
192
    for (plane = 0; plane < 3; plane++) {
1521
9360
        for (i = 0; i < 64; i++) {
1522
9216
            s->dct_tokens[plane][i] = base;
1523
9216
            base += s->fragment_width[!!plane] * s->fragment_height[!!plane];
1524
        }
1525
    }
1526
48
}
1527
1528
24
static int vp4_unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1529
{
1530
    int i, j;
1531
    int dc_y_table;
1532
    int dc_c_table;
1533
    int ac_y_table;
1534
    int ac_c_table;
1535
    VLC *tables[2][64];
1536
    int plane, sb_y, sb_x;
1537
    int eob_tracker[64];
1538
    VP4Predictor dc_pred[6][6];
1539
    int last_dc[NB_VP4_DC_TYPES];
1540
1541
24
    if (get_bits_left(gb) < 16)
1542
        return AVERROR_INVALIDDATA;
1543
1544
    /* fetch the DC table indexes */
1545
24
    dc_y_table = get_bits(gb, 4);
1546
24
    dc_c_table = get_bits(gb, 4);
1547
1548
24
    ac_y_table = get_bits(gb, 4);
1549
24
    ac_c_table = get_bits(gb, 4);
1550
1551
    /* build tables of DC/AC VLC tables */
1552
1553
    /* DC table group */
1554
24
    tables[0][0] = &s->coeff_vlc[dc_y_table];
1555
24
    tables[1][0] = &s->coeff_vlc[dc_c_table];
1556
144
    for (i = 1; i <= 5; i++) {
1557
        /* AC VLC table group 1 */
1558
120
        tables[0][i] = &s->coeff_vlc[ac_y_table + 16];
1559
120
        tables[1][i] = &s->coeff_vlc[ac_c_table + 16];
1560
    }
1561
240
    for (i = 6; i <= 14; i++) {
1562
        /* AC VLC table group 2 */
1563
216
        tables[0][i] = &s->coeff_vlc[ac_y_table + 32];
1564
216
        tables[1][i] = &s->coeff_vlc[ac_c_table + 32];
1565
    }
1566
336
    for (i = 15; i <= 27; i++) {
1567
        /* AC VLC table group 3 */
1568
312
        tables[0][i] = &s->coeff_vlc[ac_y_table + 48];
1569
312
        tables[1][i] = &s->coeff_vlc[ac_c_table + 48];
1570
    }
1571
888
    for (i = 28; i <= 63; i++) {
1572
        /* AC VLC table group 4 */
1573
864
        tables[0][i] = &s->coeff_vlc[ac_y_table + 64];
1574
864
        tables[1][i] = &s->coeff_vlc[ac_c_table + 64];
1575
    }
1576
1577
24
    vp4_set_tokens_base(s);
1578
1579
24
    memset(last_dc, 0, sizeof(last_dc));
1580
1581

96
    for (plane = 0; plane < ((s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 1 : 3); plane++) {
1582
72
        memset(eob_tracker, 0, sizeof(eob_tracker));
1583
1584
        /* initialise dc prediction */
1585
3720
        for (i = 0; i < s->fragment_width[!!plane]; i++)
1586
3648
            vp4_dc_predictor_reset(&s->dc_pred_row[i]);
1587
1588
504
        for (j = 0; j < 6; j++)
1589
3024
            for (i = 0; i < 6; i++)
1590
2592
                vp4_dc_predictor_reset(&dc_pred[j][i]);
1591
1592
456
        for (sb_y = 0; sb_y * 4 < s->fragment_height[!!plane]; sb_y++) {
1593
5952
            for (sb_x = 0; sb_x *4 < s->fragment_width[!!plane]; sb_x++) {
1594
5568
                vp4_dc_pred_before(s, dc_pred, sb_x);
1595
94656
                for (j = 0; j < 16; j++) {
1596
89088
                        int hx = hilbert_offset[j][0];
1597
89088
                        int hy = hilbert_offset[j][1];
1598
89088
                        int x  = 4 * sb_x + hx;
1599
89088
                        int y  = 4 * sb_y + hy;
1600
89088
                        VP4Predictor *this_dc_pred = &dc_pred[hy + 1][hx + 1];
1601
                        int fragment, dc_block_type;
1602
1603

89088
                        if (x >= s->fragment_width[!!plane] || y >= s->fragment_height[!!plane])
1604
1536
                            continue;
1605
1606
87552
                        fragment = s->fragment_start[plane] + y * s->fragment_width[!!plane] + x;
1607
1608
87552
                        if (s->all_fragments[fragment].coding_method == MODE_COPY)
1609
24300
                            continue;
1610
1611
63252
                        if (vp4_unpack_vlcs(s, gb, tables[!!plane], plane, eob_tracker, fragment) < 0)
1612
                            return -1;
1613
1614
63252
                        dc_block_type = vp4_pred_block_type_map[s->all_fragments[fragment].coding_method];
1615
1616
63252
                        s->all_fragments[fragment].dc +=
1617
63252
                            vp4_dc_pred(s, this_dc_pred, last_dc, dc_block_type, plane);
1618
1619
63252
                        this_dc_pred->type = dc_block_type,
1620
63252
                        this_dc_pred->dc   = last_dc[dc_block_type] = s->all_fragments[fragment].dc;
1621
                }
1622
5568
                vp4_dc_pred_after(s, dc_pred, sb_x);
1623
            }
1624
        }
1625
    }
1626
1627
24
    vp4_set_tokens_base(s);
1628
1629
24
    return 0;
1630
}
1631
#endif
1632
1633
/*
1634
 * This function reverses the DC prediction for each coded fragment in
1635
 * the frame. Much of this function is adapted directly from the original
1636
 * VP3 source code.
1637
 */
1638
#define COMPATIBLE_FRAME(x)                                                   \
1639
    (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1640
#define DC_COEFF(u) s->all_fragments[u].dc
1641
1642
420
static void reverse_dc_prediction(Vp3DecodeContext *s,
1643
                                  int first_fragment,
1644
                                  int fragment_width,
1645
                                  int fragment_height)
1646
{
1647
#define PUL 8
1648
#define PU 4
1649
#define PUR 2
1650
#define PL 1
1651
1652
    int x, y;
1653
420
    int i = first_fragment;
1654
1655
    int predicted_dc;
1656
1657
    /* DC values for the left, up-left, up, and up-right fragments */
1658
    int vl, vul, vu, vur;
1659
1660
    /* indexes for the left, up-left, up, and up-right fragments */
1661
    int l, ul, u, ur;
1662
1663
    /*
1664
     * The 6 fields mean:
1665
     *   0: up-left multiplier
1666
     *   1: up multiplier
1667
     *   2: up-right multiplier
1668
     *   3: left multiplier
1669
     */
1670
    static const int predictor_transform[16][4] = {
1671
        {    0,   0,   0,   0 },
1672
        {    0,   0,   0, 128 }, // PL
1673
        {    0,   0, 128,   0 }, // PUR
1674
        {    0,   0,  53,  75 }, // PUR|PL
1675
        {    0, 128,   0,   0 }, // PU
1676
        {    0,  64,   0,  64 }, // PU |PL
1677
        {    0, 128,   0,   0 }, // PU |PUR
1678
        {    0,   0,  53,  75 }, // PU |PUR|PL
1679
        {  128,   0,   0,   0 }, // PUL
1680
        {    0,   0,   0, 128 }, // PUL|PL
1681
        {   64,   0,  64,   0 }, // PUL|PUR
1682
        {    0,   0,  53,  75 }, // PUL|PUR|PL
1683
        {    0, 128,   0,   0 }, // PUL|PU
1684
        { -104, 116,   0, 116 }, // PUL|PU |PL
1685
        {   24,  80,  24,   0 }, // PUL|PU |PUR
1686
        { -104, 116,   0, 116 }  // PUL|PU |PUR|PL
1687
    };
1688
1689
    /* This table shows which types of blocks can use other blocks for
1690
     * prediction. For example, INTRA is the only mode in this table to
1691
     * have a frame number of 0. That means INTRA blocks can only predict
1692
     * from other INTRA blocks. There are 2 golden frame coding types;
1693
     * blocks encoding in these modes can only predict from other blocks
1694
     * that were encoded with these 1 of these 2 modes. */
1695
    static const unsigned char compatible_frame[9] = {
1696
        1,    /* MODE_INTER_NO_MV */
1697
        0,    /* MODE_INTRA */
1698
        1,    /* MODE_INTER_PLUS_MV */
1699
        1,    /* MODE_INTER_LAST_MV */
1700
        1,    /* MODE_INTER_PRIOR_MV */
1701
        2,    /* MODE_USING_GOLDEN */
1702
        2,    /* MODE_GOLDEN_MV */
1703
        1,    /* MODE_INTER_FOUR_MV */
1704
        3     /* MODE_COPY */
1705
    };
1706
    int current_frame_type;
1707
1708
    /* there is a last DC predictor for each of the 3 frame types */
1709
    short last_dc[3];
1710
1711
420
    int transform = 0;
1712
1713
420
    vul =
1714
420
    vu  =
1715
420
    vur =
1716
420
    vl  = 0;
1717
420
    last_dc[0] =
1718
420
    last_dc[1] =
1719
420
    last_dc[2] = 0;
1720
1721
    /* for each fragment row... */
1722
14292
    for (y = 0; y < fragment_height; y++) {
1723
        /* for each fragment in a row... */
1724
1774512
        for (x = 0; x < fragment_width; x++, i++) {
1725
1726
            /* reverse prediction if this block was coded */
1727
1760640
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1728
532576
                current_frame_type =
1729
532576
                    compatible_frame[s->all_fragments[i].coding_method];
1730
1731
532576
                transform = 0;
1732
532576
                if (x) {
1733
528445
                    l  = i - 1;
1734
528445
                    vl = DC_COEFF(l);
1735
528445
                    if (COMPATIBLE_FRAME(l))
1736
483186
                        transform |= PL;
1737
                }
1738
532576
                if (y) {
1739
513468
                    u  = i - fragment_width;
1740
513468
                    vu = DC_COEFF(u);
1741
513468
                    if (COMPATIBLE_FRAME(u))
1742
476069
                        transform |= PU;
1743
513468
                    if (x) {
1744
509661
                        ul  = i - fragment_width - 1;
1745
509661
                        vul = DC_COEFF(ul);
1746
509661
                        if (COMPATIBLE_FRAME(ul))
1747
461368
                            transform |= PUL;
1748
                    }
1749
513468
                    if (x + 1 < fragment_width) {
1750
505812
                        ur  = i - fragment_width + 1;
1751
505812
                        vur = DC_COEFF(ur);
1752
505812
                        if (COMPATIBLE_FRAME(ur))
1753
459980
                            transform |= PUR;
1754
                    }
1755
                }
1756
1757
532576
                if (transform == 0) {
1758
                    /* if there were no fragments to predict from, use last
1759
                     * DC saved */
1760
10817
                    predicted_dc = last_dc[current_frame_type];
1761
                } else {
1762
                    /* apply the appropriate predictor transform */
1763
521759
                    predicted_dc =
1764
521759
                        (predictor_transform[transform][0] * vul) +
1765
521759
                        (predictor_transform[transform][1] * vu) +
1766
521759
                        (predictor_transform[transform][2] * vur) +
1767
521759
                        (predictor_transform[transform][3] * vl);
1768
1769
521759
                    predicted_dc /= 128;
1770
1771
                    /* check for outranging on the [ul u l] and
1772
                     * [ul u ur l] predictors */
1773

521759
                    if ((transform == 15) || (transform == 13)) {
1774
434711
                        if (FFABS(predicted_dc - vu) > 128)
1775
2351
                            predicted_dc = vu;
1776
432360
                        else if (FFABS(predicted_dc - vl) > 128)
1777
1111
                            predicted_dc = vl;
1778
431249
                        else if (FFABS(predicted_dc - vul) > 128)
1779
2253
                            predicted_dc = vul;
1780
                    }
1781
                }
1782
1783
                /* at long last, apply the predictor */
1784
532576
                DC_COEFF(i) += predicted_dc;
1785
                /* save the DC */
1786
532576
                last_dc[current_frame_type] = DC_COEFF(i);
1787
            }
1788
        }
1789
    }
1790
420
}
1791
1792
2700
static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1793
                              int ystart, int yend)
1794
{
1795
    int x, y;
1796
2700
    int *bounding_values = s->bounding_values_array + 127;
1797
1798
2700
    int width           = s->fragment_width[!!plane];
1799
2700
    int height          = s->fragment_height[!!plane];
1800
2700
    int fragment        = s->fragment_start[plane] + ystart * width;
1801
2700
    ptrdiff_t stride    = s->current_frame.f->linesize[plane];
1802
2700
    uint8_t *plane_data = s->current_frame.f->data[plane];
1803
2700
    if (!s->flipped_image)
1804
2700
        stride = -stride;
1805
2700
    plane_data += s->data_offset[plane] + 8 * ystart * stride;
1806
1807
10530
    for (y = ystart; y < yend; y++) {
1808
486710
        for (x = 0; x < width; x++) {
1809
            /* This code basically just deblocks on the edges of coded blocks.
1810
             * However, it has to be much more complicated because of the
1811
             * brain damaged deblock ordering used in VP3/Theora. Order matters
1812
             * because some pixels get filtered twice. */
1813
478880
            if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1814
                /* do not perform left edge filter for left columns frags */
1815
367025
                if (x > 0) {
1816
363601
                    s->vp3dsp.h_loop_filter(
1817
363601
                        plane_data + 8 * x,
1818
                        stride, bounding_values);
1819
                }
1820
1821
                /* do not perform top edge filter for top row fragments */
1822
367025
                if (y > 0) {
1823
349625
                    s->vp3dsp.v_loop_filter(
1824
349625
                        plane_data + 8 * x,
1825
                        stride, bounding_values);
1826
                }
1827
1828
                /* do not perform right edge filter for right column
1829
                 * fragments or if right fragment neighbor is also coded
1830
                 * in this frame (it will be filtered in next iteration) */
1831
367025
                if ((x < width - 1) &&
1832
359725
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1833
32268
                    s->vp3dsp.h_loop_filter(
1834
32268
                        plane_data + 8 * x + 8,
1835
                        stride, bounding_values);
1836
                }
1837
1838
                /* do not perform bottom edge filter for bottom row
1839
                 * fragments or if bottom fragment neighbor is also coded
1840
                 * in this frame (it will be filtered in the next row) */
1841
367025
                if ((y < height - 1) &&
1842
352272
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1843
32946
                    s->vp3dsp.v_loop_filter(
1844
32946
                        plane_data + 8 * x + 8 * stride,
1845
                        stride, bounding_values);
1846
                }
1847
            }
1848
1849
478880
            fragment++;
1850
        }
1851
7830
        plane_data += 8 * stride;
1852
    }
1853
2700
}
1854
1855
/**
1856
 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1857
 * for the next block in coding order
1858
 */
1859
595828
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1860
                              int plane, int inter, int16_t block[64])
1861
{
1862
595828
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1863
595828
    uint8_t *perm = s->idct_scantable;
1864
595828
    int i = 0;
1865
1866
    do {
1867
3343039
        int token = *s->dct_tokens[plane][i];
1868

3343039
        switch (token & 3) {
1869
587550
        case 0: // EOB
1870
587550
            if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1871
280242
                s->dct_tokens[plane][i]++;
1872
            else
1873
307308
                *s->dct_tokens[plane][i] = token & ~3;
1874
587550
            goto end;
1875
913868
        case 1: // zero run
1876
913868
            s->dct_tokens[plane][i]++;
1877
913868
            i += (token >> 2) & 0x7f;
1878
913868
            if (i > 63) {
1879
                av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1880
                return i;
1881
            }
1882
913868
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1883
913868
            i++;
1884
913868
            break;
1885
1841621
        case 2: // coeff
1886
1841621
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1887
1841621
            s->dct_tokens[plane][i++]++;
1888
1841621
            break;
1889
        default: // shouldn't happen
1890
            return i;
1891
        }
1892
2755489
    } while (i < 64);
1893
    // return value is expected to be a valid level
1894
8278
    i--;
1895
595828
end:
1896
    // the actual DC+prediction is in the fragment structure
1897
595828
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1898
595828
    return i;
1899
}
1900
1901
/**
1902
 * called when all pixels up to row y are complete
1903
 */
1904
1232
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1905
{
1906
    int h, cy, i;
1907
    int offset[AV_NUM_DATA_POINTERS];
1908
1909
1232
    if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1910
        int y_flipped = s->flipped_image ? s->height - y : y;
1911
1912
        /* At the end of the frame, report INT_MAX instead of the height of
1913
         * the frame. This makes the other threads' ff_thread_await_progress()
1914
         * calls cheaper, because they don't have to clip their values. */
1915
        ff_thread_report_progress(&s->current_frame,
1916
                                  y_flipped == s->height ? INT_MAX
1917
                                                         : y_flipped - 1,
1918
                                  0);
1919
    }
1920
1921
1232
    if (!s->avctx->draw_horiz_band)
1922
1232
        return;
1923
1924
    h = y - s->last_slice_end;
1925
    s->last_slice_end = y;
1926
    y -= h;
1927
1928
    if (!s->flipped_image)
1929
        y = s->height - y - h;
1930
1931
    cy        = y >> s->chroma_y_shift;
1932
    offset[0] = s->current_frame.f->linesize[0] * y;
1933
    offset[1] = s->current_frame.f->linesize[1] * cy;
1934
    offset[2] = s->current_frame.f->linesize[2] * cy;
1935
    for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1936
        offset[i] = 0;
1937
1938
    emms_c();
1939
    s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1940
}
1941
1942
/**
1943
 * Wait for the reference frame of the current fragment.
1944
 * The progress value is in luma pixel rows.
1945
 */
1946
static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1947
                                int motion_y, int y)
1948
{
1949
    ThreadFrame *ref_frame;
1950
    int ref_row;
1951
    int border = motion_y & 1;
1952
1953
    if (fragment->coding_method == MODE_USING_GOLDEN ||
1954
        fragment->coding_method == MODE_GOLDEN_MV)
1955
        ref_frame = &s->golden_frame;
1956
    else
1957
        ref_frame = &s->last_frame;
1958
1959
    ref_row = y + (motion_y >> 1);
1960
    ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1961
1962
    ff_thread_await_progress(ref_frame, ref_row, 0);
1963
}
1964
1965
#if CONFIG_VP4_DECODER
1966
/**
1967
 * @return non-zero if temp (edge_emu_buffer) was populated
1968
 */
1969
36568
static int vp4_mc_loop_filter(Vp3DecodeContext *s, int plane, int motion_x, int motion_y, int bx, int by,
1970
       uint8_t * motion_source, int stride, int src_x, int src_y, uint8_t *temp)
1971
{
1972
36568
    int motion_shift = plane ? 4 : 2;
1973
36568
    int subpel_mask = plane ? 3 : 1;
1974
36568
    int *bounding_values = s->bounding_values_array + 127;
1975
1976
    int i;
1977
    int x, y;
1978
    int x2, y2;
1979
    int x_subpel, y_subpel;
1980
    int x_offset, y_offset;
1981
1982
36568
    int block_width = plane ? 8 : 16;
1983

36568
    int plane_width  = s->width  >> (plane && s->chroma_x_shift);
1984

36568
    int plane_height = s->height >> (plane && s->chroma_y_shift);
1985
1986
#define loop_stride 12
1987
    uint8_t loop[12 * loop_stride];
1988
1989
    /* using division instead of shift to correctly handle negative values */
1990
36568
    x = 8 * bx + motion_x / motion_shift;
1991
36568
    y = 8 * by + motion_y / motion_shift;
1992
1993
36568
    x_subpel = motion_x & subpel_mask;
1994
36568
    y_subpel = motion_y & subpel_mask;
1995
1996

36568
    if (x_subpel || y_subpel) {
1997
30063
        x--;
1998
30063
        y--;
1999
2000
30063
        if (x_subpel)
2001

22774
            x = FFMIN(x, x + FFSIGN(motion_x));
2002
2003
30063
        if (y_subpel)
2004

15343
            y = FFMIN(y, y + FFSIGN(motion_y));
2005
2006
30063
        x2 = x + block_width;
2007
30063
        y2 = y + block_width;
2008
2009


30063
        if (x2 < 0 || x2 >= plane_width || y2 < 0 || y2 >= plane_height)
2010
596
            return 0;
2011
2012
29467
        x_offset = (-(x + 2) & 7) + 2;
2013
29467
        y_offset = (-(y + 2) & 7) + 2;
2014
2015

29467
        if (x_offset > 8 + x_subpel && y_offset > 8 + y_subpel)
2016
            return 0;
2017
2018
29467
        s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2019
             loop_stride, stride,
2020
             12, 12, src_x - 1, src_y - 1,
2021
             plane_width,
2022
             plane_height);
2023
2024
29467
        if (x_offset <= 8 + x_subpel)
2025
25000
            ff_vp3dsp_h_loop_filter_12(loop + x_offset, loop_stride, bounding_values);
2026
2027
29467
        if (y_offset <= 8 + y_subpel)
2028
17032
            ff_vp3dsp_v_loop_filter_12(loop + y_offset*loop_stride, loop_stride, bounding_values);
2029
2030
    } else {
2031
2032
6505
        x_offset = -x & 7;
2033
6505
        y_offset = -y & 7;
2034
2035

6505
        if (!x_offset && !y_offset)
2036
118
            return 0;
2037
2038
6387
        s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2039
             loop_stride, stride,
2040
             12, 12, src_x - 1, src_y - 1,
2041
             plane_width,
2042
             plane_height);
2043
2044
#define safe_loop_filter(name, ptr, stride, bounding_values) \
2045
    if ((uintptr_t)(ptr) & 7) \
2046
        s->vp3dsp.name##_unaligned(ptr, stride, bounding_values); \
2047
    else \
2048
        s->vp3dsp.name(ptr, stride, bounding_values);
2049
2050
6387
        if (x_offset)
2051
5671
            safe_loop_filter(h_loop_filter, loop + loop_stride + x_offset + 1, loop_stride, bounding_values);
2052
2053
6387
        if (y_offset)
2054
3055
            safe_loop_filter(v_loop_filter, loop + (y_offset + 1)*loop_stride + 1, loop_stride, bounding_values);
2055
    }
2056
2057
358540
    for (i = 0; i < 9; i++)
2058
322686
        memcpy(temp + i*stride, loop + (i + 1) * loop_stride + 1, 9);
2059
2060
35854
    return 1;
2061
}
2062
#endif
2063
2064
/*
2065
 * Perform the final rendering for a particular slice of data.
2066
 * The slice number ranges from 0..(c_superblock_height - 1).
2067
 */
2068
1068
static void render_slice(Vp3DecodeContext *s, int slice)
2069
{
2070
    int x, y, i, j, fragment;
2071
1068
    int16_t *block = s->block;
2072
1068
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
2073
    int motion_halfpel_index;
2074
    uint8_t *motion_source;
2075
    int plane, first_pixel;
2076
2077
1068
    if (slice >= s->c_superblock_height)
2078
        return;
2079
2080
4272
    for (plane = 0; plane < 3; plane++) {
2081
3204
        uint8_t *output_plane = s->current_frame.f->data[plane] +
2082
3204
                                s->data_offset[plane];
2083
3204
        uint8_t *last_plane = s->last_frame.f->data[plane] +
2084
3204
                              s->data_offset[plane];
2085
3204
        uint8_t *golden_plane = s->golden_frame.f->data[plane] +
2086
3204
                                s->data_offset[plane];
2087
3204
        ptrdiff_t stride = s->current_frame.f->linesize[plane];
2088

3204
        int plane_width  = s->width  >> (plane && s->chroma_x_shift);
2089

3204
        int plane_height = s->height >> (plane && s->chroma_y_shift);
2090
3204
        int8_t(*motion_val)[2] = s->motion_val[!!plane];
2091
2092

3204
        int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
2093

3204
        int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
2094
3204
        int slice_width  = plane ? s->c_superblock_width
2095
3204
                                 : s->y_superblock_width;
2096
2097
3204
        int fragment_width  = s->fragment_width[!!plane];
2098
3204
        int fragment_height = s->fragment_height[!!plane];
2099
3204
        int fragment_start  = s->fragment_start[plane];
2100
2101
4272
        int do_await = !plane && HAVE_THREADS &&
2102
1068
                       (s->avctx->active_thread_type & FF_THREAD_FRAME);
2103
2104
3204
        if (!s->flipped_image)
2105
3204
            stride = -stride;
2106
        if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY))
2107
            continue;
2108
2109
        /* for each superblock row in the slice (both of them)... */
2110
7476
        for (; sb_y < slice_height; sb_y++) {
2111
            /* for each superblock in a row... */
2112
129560
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
2113
                /* for each block in a superblock... */
2114
2129896
                for (j = 0; j < 16; j++) {
2115
2004608
                    x        = 4 * sb_x + hilbert_offset[j][0];
2116
2004608
                    y        = 4 * sb_y + hilbert_offset[j][1];
2117
2004608
                    fragment = y * fragment_width + x;
2118
2119
2004608
                    i = fragment_start + fragment;
2120
2121
                    // bounds check
2122

2004608
                    if (x >= fragment_width || y >= fragment_height)
2123
156416
                        continue;
2124
2125
1848192
                    first_pixel = 8 * y * stride + 8 * x;
2126
2127
1848192
                    if (do_await &&
2128
                        s->all_fragments[i].coding_method != MODE_INTRA)
2129
                        await_reference_row(s, &s->all_fragments[i],
2130
                                            motion_val[fragment][1],
2131
                                            (16 * y) >> s->chroma_y_shift);
2132
2133
                    /* transform if this block was coded */
2134
1848192
                    if (s->all_fragments[i].coding_method != MODE_COPY) {
2135
595828
                        if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
2136
592182
                            (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
2137
7685
                            motion_source = golden_plane;
2138
                        else
2139
588143
                            motion_source = last_plane;
2140
2141
595828
                        motion_source       += first_pixel;
2142
595828
                        motion_halfpel_index = 0;
2143
2144
                        /* sort out the motion vector if this fragment is coded
2145
                         * using a motion vector method */
2146
595828
                        if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
2147
377836
                            (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
2148
                            int src_x, src_y;
2149
374190
                            int standard_mc = 1;
2150
374190
                            motion_x = motion_val[fragment][0];
2151
374190
                            motion_y = motion_val[fragment][1];
2152
#if CONFIG_VP4_DECODER
2153

374190
                            if (plane && s->version >= 2) {
2154
11931
                                motion_x = (motion_x >> 1) | (motion_x & 1);
2155
11931
                                motion_y = (motion_y >> 1) | (motion_y & 1);
2156
                            }
2157
#endif
2158
2159
374190
                            src_x = (motion_x >> 1) + 8 * x;
2160
374190
                            src_y = (motion_y >> 1) + 8 * y;
2161
2162
374190
                            motion_halfpel_index = motion_x & 0x01;
2163
374190
                            motion_source       += (motion_x >> 1);
2164
2165
374190
                            motion_halfpel_index |= (motion_y & 0x01) << 1;
2166
374190
                            motion_source        += ((motion_y >> 1) * stride);
2167
2168
#if CONFIG_VP4_DECODER
2169
374190
                            if (s->version >= 2) {
2170
36568
                                uint8_t *temp = s->edge_emu_buffer;
2171
36568
                                if (stride < 0)
2172
36568
                                    temp -= 8 * stride;
2173
36568
                                if (vp4_mc_loop_filter(s, plane, motion_val[fragment][0], motion_val[fragment][1], x, y, motion_source, stride, src_x, src_y, temp)) {
2174
35854
                                    motion_source = temp;
2175
35854
                                    standard_mc = 0;
2176
                                }
2177
                            }
2178
#endif
2179
2180

374190
                            if (standard_mc && (
2181
336975
                                src_x < 0 || src_y < 0 ||
2182
322220
                                src_x + 9 >= plane_width ||
2183
315161
                                src_y + 9 >= plane_height)) {
2184
34786
                                uint8_t *temp = s->edge_emu_buffer;
2185
34786
                                if (stride < 0)
2186
34786
                                    temp -= 8 * stride;
2187
2188
34786
                                s->vdsp.emulated_edge_mc(temp, motion_source,
2189
                                                         stride, stride,
2190
                                                         9, 9, src_x, src_y,
2191
                                                         plane_width,
2192
                                                         plane_height);
2193
34786
                                motion_source = temp;
2194
                            }
2195
                        }
2196
2197
                        /* first, take care of copying a block from either the
2198
                         * previous or the golden frame */
2199
595828
                        if (s->all_fragments[i].coding_method != MODE_INTRA) {
2200
                            /* Note, it is possible to implement all MC cases
2201
                             * with put_no_rnd_pixels_l2 which would look more
2202
                             * like the VP3 source but this would be slower as
2203
                             * put_no_rnd_pixels_tab is better optimized */
2204
411494
                            if (motion_halfpel_index != 3) {
2205
293916
                                s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
2206
                                    output_plane + first_pixel,
2207
                                    motion_source, stride, 8);
2208
                            } else {
2209
                                /* d is 0 if motion_x and _y have the same sign,
2210
                                 * else -1 */
2211
117578
                                int d = (motion_x ^ motion_y) >> 31;
2212
117578
                                s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
2213
117578
                                                               motion_source - d,
2214
117578
                                                               motion_source + stride + 1 + d,
2215
                                                               stride, 8);
2216
                            }
2217
                        }
2218
2219
                        /* invert DCT and place (or add) in final output */
2220
2221
595828
                        if (s->all_fragments[i].coding_method == MODE_INTRA) {
2222
184334
                            vp3_dequant(s, s->all_fragments + i,
2223
                                        plane, 0, block);
2224
184334
                            s->vp3dsp.idct_put(output_plane + first_pixel,
2225
                                               stride,
2226
                                               block);
2227
                        } else {
2228
411494
                            if (vp3_dequant(s, s->all_fragments + i,
2229
                                            plane, 1, block)) {
2230
284346
                                s->vp3dsp.idct_add(output_plane + first_pixel,
2231
                                                   stride,
2232
                                                   block);
2233
                            } else {
2234
127148
                                s->vp3dsp.idct_dc_add(output_plane + first_pixel,
2235
                                                      stride, block);
2236
                            }
2237
                        }
2238
                    } else {
2239
                        /* copy directly from the previous frame */
2240
1252364
                        s->hdsp.put_pixels_tab[1][0](
2241
                            output_plane + first_pixel,
2242
1252364
                            last_plane + first_pixel,
2243
                            stride, 8);
2244
                    }
2245
                }
2246
            }
2247
2248
            // Filter up to the last row in the superblock row
2249

4272
            if (s->version < 2 && !s->skip_loop_filter)
2250
2280
                apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
2251
2280
                                  FFMIN(4 * sb_y + 3, fragment_height - 1));
2252
        }
2253
    }
2254
2255
    /* this looks like a good place for slice dispatch... */
2256
    /* algorithm:
2257
     *   if (slice == s->macroblock_height - 1)
2258
     *     dispatch (both last slice & 2nd-to-last slice);
2259
     *   else if (slice > 0)
2260
     *     dispatch (slice - 1);
2261
     */
2262
2263
1068
    vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
2264
                                 s->height - 16));
2265
}
2266
2267
/// Allocate tables for per-frame data in Vp3DecodeContext
2268
12
static av_cold int allocate_tables(AVCodecContext *avctx)
2269
{
2270
12
    Vp3DecodeContext *s = avctx->priv_data;
2271
    int y_fragment_count, c_fragment_count;
2272
2273
12
    free_tables(avctx);
2274
2275
12
    y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
2276
12
    c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
2277
2278
    /* superblock_coding is used by unpack_superblocks (VP3/Theora) and vp4_unpack_macroblocks (VP4) */
2279
12
    s->superblock_coding = av_mallocz(FFMAX(s->superblock_count, s->yuv_macroblock_count));
2280
12
    s->all_fragments     = av_mallocz_array(s->fragment_count, sizeof(Vp3Fragment));
2281
2282
12
    s-> kf_coded_fragment_list = av_mallocz_array(s->fragment_count, sizeof(int));
2283
12
    s->nkf_coded_fragment_list = av_mallocz_array(s->fragment_count, sizeof(int));
2284
12
    memset(s-> num_kf_coded_fragment, -1, sizeof(s-> num_kf_coded_fragment));
2285
2286
12
    s->dct_tokens_base = av_mallocz_array(s->fragment_count,
2287
                                          64 * sizeof(*s->dct_tokens_base));
2288
12
    s->motion_val[0] = av_mallocz_array(y_fragment_count, sizeof(*s->motion_val[0]));
2289
12
    s->motion_val[1] = av_mallocz_array(c_fragment_count, sizeof(*s->motion_val[1]));
2290
2291
    /* work out the block mapping tables */
2292
12
    s->superblock_fragments = av_mallocz_array(s->superblock_count, 16 * sizeof(int));
2293
12
    s->macroblock_coding    = av_mallocz(s->macroblock_count + 1);
2294
2295
12
    s->dc_pred_row = av_malloc_array(s->y_superblock_width * 4, sizeof(*s->dc_pred_row));
2296
2297

12
    if (!s->superblock_coding    || !s->all_fragments          ||
2298

12
        !s->dct_tokens_base      || !s->kf_coded_fragment_list ||
2299
12
        !s->nkf_coded_fragment_list ||
2300

12
        !s->superblock_fragments || !s->macroblock_coding      ||
2301
12
        !s->dc_pred_row ||
2302

12
        !s->motion_val[0]        || !s->motion_val[1]) {
2303
        return -1;
2304
    }
2305
2306
12
    init_block_mapping(s);
2307
2308
12
    return 0;
2309
}
2310
2311
12
static av_cold int init_frames(Vp3DecodeContext *s)
2312
{
2313
12
    s->current_frame.f = av_frame_alloc();
2314
12
    s->last_frame.f    = av_frame_alloc();
2315
12
    s->golden_frame.f  = av_frame_alloc();
2316
2317

12
    if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f)
2318
        return AVERROR(ENOMEM);
2319
2320
12
    return 0;
2321
}
2322
2323
12
static av_cold int vp3_decode_init(AVCodecContext *avctx)
2324
{
2325
12
    Vp3DecodeContext *s = avctx->priv_data;
2326
    int i, inter, plane, ret;
2327
    int c_width;
2328
    int c_height;
2329
    int y_fragment_count, c_fragment_count;
2330
#if CONFIG_VP4_DECODER
2331
    int j;
2332
#endif
2333
2334
12
    ret = init_frames(s);
2335
12
    if (ret < 0)
2336
        return ret;
2337
2338
12
    if (avctx->codec_tag == MKTAG('V', 'P', '4', '0'))
2339
2
        s->version = 3;
2340
10
    else if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
2341
        s->version = 0;
2342
    else
2343
10
        s->version = 1;
2344
2345
12
    s->avctx  = avctx;
2346
12
    s->width  = FFALIGN(avctx->coded_width, 16);
2347
12
    s->height = FFALIGN(avctx->coded_height, 16);
2348
12
    if (avctx->codec_id != AV_CODEC_ID_THEORA)
2349
5
        avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2350
12
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
2351
12
    ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT);
2352
12
    ff_videodsp_init(&s->vdsp, 8);
2353
12
    ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
2354
2355
780
    for (i = 0; i < 64; i++) {
2356
#define TRANSPOSE(x) (((x) >> 3) | (((x) & 7) << 3))
2357
768
        s->idct_permutation[i] = TRANSPOSE(i);
2358
768
        s->idct_scantable[i]   = TRANSPOSE(ff_zigzag_direct[i]);
2359
#undef TRANSPOSE
2360
    }
2361
2362
    /* initialize to an impossible value which will force a recalculation
2363
     * in the first frame decode */
2364
48
    for (i = 0; i < 3; i++)
2365
36
        s->qps[i] = -1;
2366
2367
12
    ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
2368
12
    if (ret)
2369
        return ret;
2370
2371
12
    s->y_superblock_width  = (s->width  + 31) / 32;
2372
12
    s->y_superblock_height = (s->height + 31) / 32;
2373
12
    s->y_superblock_count  = s->y_superblock_width * s->y_superblock_height;
2374
2375
    /* work out the dimensions for the C planes */
2376
12
    c_width                = s->width >> s->chroma_x_shift;
2377
12
    c_height               = s->height >> s->chroma_y_shift;
2378
12
    s->c_superblock_width  = (c_width  + 31) / 32;
2379
12
    s->c_superblock_height = (c_height + 31) / 32;
2380
12
    s->c_superblock_count  = s->c_superblock_width * s->c_superblock_height;
2381
2382
12
    s->superblock_count   = s->y_superblock_count + (s->c_superblock_count * 2);
2383
12
    s->u_superblock_start = s->y_superblock_count;
2384
12
    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
2385
2386
12
    s->macroblock_width  = (s->width  + 15) / 16;
2387
12
    s->macroblock_height = (s->height + 15) / 16;
2388
12
    s->macroblock_count  = s->macroblock_width * s->macroblock_height;
2389
12
    s->c_macroblock_width  = (c_width  + 15) / 16;
2390
12
    s->c_macroblock_height = (c_height + 15) / 16;
2391
12
    s->c_macroblock_count  = s->c_macroblock_width * s->c_macroblock_height;
2392
12
    s->yuv_macroblock_count = s->macroblock_count + 2 * s->c_macroblock_count;
2393
2394
12
    s->fragment_width[0]  = s->width / FRAGMENT_PIXELS;
2395
12
    s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
2396
12
    s->fragment_width[1]  = s->fragment_width[0] >> s->chroma_x_shift;
2397
12
    s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
2398
2399
    /* fragment count covers all 8x8 blocks for all 3 planes */
2400
12
    y_fragment_count     = s->fragment_width[0] * s->fragment_height[0];
2401
12
    c_fragment_count     = s->fragment_width[1] * s->fragment_height[1];
2402
12
    s->fragment_count    = y_fragment_count + 2 * c_fragment_count;
2403
12
    s->fragment_start[1] = y_fragment_count;
2404
12
    s->fragment_start[2] = y_fragment_count + c_fragment_count;
2405
2406
12
    if (!s->theora_tables) {
2407
325
        for (i = 0; i < 64; i++) {
2408
320
            s->coded_dc_scale_factor[0][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_y_dc_scale_factor[i];
2409
320
            s->coded_dc_scale_factor[1][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_uv_dc_scale_factor[i];
2410
320
            s->coded_ac_scale_factor[i] = s->version < 2 ? vp31_ac_scale_factor[i] : vp4_ac_scale_factor[i];
2411
320
            s->base_matrix[0][i]        = s->version < 2 ? vp31_intra_y_dequant[i] : vp4_generic_dequant[i];
2412
320
            s->base_matrix[1][i]        = s->version < 2 ? vp31_intra_c_dequant[i] : vp4_generic_dequant[i];
2413
320
            s->base_matrix[2][i]        = s->version < 2 ? vp31_inter_dequant[i]   : vp4_generic_dequant[i];
2414
320
            s->filter_limit_values[i]   = s->version < 2 ? vp31_filter_limit_values[i] : vp4_filter_limit_values[i];
2415
        }
2416
2417
15
        for (inter = 0; inter < 2; inter++) {
2418
40
            for (plane = 0; plane < 3; plane++) {
2419
30
                s->qr_count[inter][plane]   = 1;
2420
30
                s->qr_size[inter][plane][0] = 63;
2421
30
                s->qr_base[inter][plane][0] =
2422
30
                s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
2423
            }
2424
        }
2425
2426
        /* init VLC tables */
2427
5
        if (s->version < 2) {
2428
243
            for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++) {
2429
240
                ret = ff_init_vlc_from_lengths(&s->coeff_vlc[i], 11, 32,
2430
240
                                               &vp3_bias[i][0][1], 2,
2431
240
                                               &vp3_bias[i][0][0], 2, 1,
2432
                                               0, 0, avctx);
2433
240
                if (ret < 0)
2434
                    return ret;
2435
            }
2436
#if CONFIG_VP4_DECODER
2437
        } else { /* version >= 2 */
2438
162
            for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++) {
2439
160
                ret = ff_init_vlc_from_lengths(&s->coeff_vlc[i], 11, 32,
2440
160
                                               &vp4_bias[i][0][1], 2,
2441
160
                                               &vp4_bias[i][0][0], 2, 1,
2442
                                               0, 0, avctx);
2443
160
                if (ret < 0)
2444
                    return ret;
2445
            }
2446
#endif
2447
        }
2448
    } else {
2449
567
        for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++) {
2450
560
            const HuffTable *tab = &s->huffman_table[i];
2451
2452
560
            ret = ff_init_vlc_from_lengths(&s->coeff_vlc[i], 11, tab->nb_entries,
2453
560
                                           &tab->entries[0].len, sizeof(*tab->entries),
2454
560
                                           &tab->entries[0].sym, sizeof(*tab->entries), 1,
2455
                                           0, 0, avctx);
2456
560
            if (ret < 0)
2457
                return ret;
2458
        }
2459
    }
2460
2461
12
    ret = ff_init_vlc_from_lengths(&s->superblock_run_length_vlc, SUPERBLOCK_VLC_BITS, 34,
2462
                                   superblock_run_length_vlc_lens, 1,
2463
                                   NULL, 0, 0, 1, 0, avctx);
2464
12
    if (ret < 0)
2465
        return ret;
2466
2467
12
    ret = ff_init_vlc_from_lengths(&s->fragment_run_length_vlc, 5, 30,
2468
                                   fragment_run_length_vlc_len, 1,
2469
                                   NULL, 0, 0, 0, 0, avctx);
2470
12
    if (ret < 0)
2471
        return ret;
2472
2473
12
    ret = ff_init_vlc_from_lengths(&s->mode_code_vlc, 3, 8,
2474
                                   mode_code_vlc_len, 1,
2475
                                   NULL, 0, 0, 0, 0, avctx);
2476
12
    if (ret < 0)
2477
        return ret;
2478
2479
12
    ret = ff_init_vlc_from_lengths(&s->motion_vector_vlc, VP3_MV_VLC_BITS, 63,
2480
                                   &motion_vector_vlc_table[0][1], 2,
2481
                                   &motion_vector_vlc_table[0][0], 2, 1,
2482
                                   -31, 0, avctx);
2483
12
    if (ret < 0)
2484
        return ret;
2485
2486
#if CONFIG_VP4_DECODER
2487
36
    for (j = 0; j < 2; j++)
2488
192
        for (i = 0; i < 7; i++) {
2489
168
            ret = ff_init_vlc_from_lengths(&s->vp4_mv_vlc[j][i], VP4_MV_VLC_BITS, 63,
2490
168
                                           &vp4_mv_vlc[j][i][0][1], 2,
2491
168
                                           &vp4_mv_vlc[j][i][0][0], 2, 1, -31,
2492
                                           0, avctx);
2493
168
            if (ret < 0)
2494
                return ret;
2495
        }
2496
2497
    /* version >= 2 */
2498
36
    for (i = 0; i < 2; i++)
2499
24
        if ((ret = init_vlc(&s->block_pattern_vlc[i], 3, 14,
2500
                            &vp4_block_pattern_vlc[i][0][1], 2, 1,
2501
                            &vp4_block_pattern_vlc[i][0][0], 2, 1, 0)) < 0)
2502
            return ret;
2503
#endif
2504
2505
12
    return allocate_tables(avctx);
2506
}
2507
2508
/// Release and shuffle frames after decode finishes
2509
164
static int update_frames(AVCodecContext *avctx)
2510
{
2511
164
    Vp3DecodeContext *s = avctx->priv_data;
2512
164
    int ret = 0;
2513
2514
    /* shuffle frames (last = current) */
2515
164
    ff_thread_release_buffer(avctx, &s->last_frame);
2516
164
    ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
2517
164
    if (ret < 0)
2518
        goto fail;
2519
2520
164
    if (s->keyframe) {
2521
6
        ff_thread_release_buffer(avctx, &s->golden_frame);
2522
6
        ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
2523
    }
2524
2525
158
fail:
2526
164
    ff_thread_release_buffer(avctx, &s->current_frame);
2527
164
    return ret;
2528
}
2529
2530
#if HAVE_THREADS
2531
static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
2532
{
2533
    ff_thread_release_buffer(s->avctx, dst);
2534
    if (src->f->data[0])
2535
        return ff_thread_ref_frame(dst, src);
2536
    return 0;
2537
}
2538
2539
static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
2540
{
2541
    int ret;
2542
    if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
2543
        (ret = ref_frame(dst, &dst->golden_frame,  &src->golden_frame)) < 0  ||
2544
        (ret = ref_frame(dst, &dst->last_frame,    &src->last_frame)) < 0)
2545
        return ret;
2546
    return 0;
2547
}
2548
2549
static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
2550
{
2551
    Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
2552
    int qps_changed = 0, i, err;
2553
2554
    if (!s1->current_frame.f->data[0] ||
2555
        s->width != s1->width || s->height != s1->height) {
2556
        if (s != s1)
2557
            ref_frames(s, s1);
2558
        return -1;
2559
    }
2560
2561
    if (s != s1) {
2562
        // copy previous frame data
2563
        if ((err = ref_frames(s, s1)) < 0)
2564
            return err;
2565
2566
        s->keyframe = s1->keyframe;
2567
2568
        // copy qscale data if necessary
2569
        for (i = 0; i < 3; i++) {
2570
            if (s->qps[i] != s1->qps[1]) {
2571
                qps_changed = 1;
2572
                memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
2573
            }
2574
        }
2575
2576
        if (s->qps[0] != s1->qps[0])
2577
            memcpy(&s->bounding_values_array, &s1->bounding_values_array,
2578
                   sizeof(s->bounding_values_array));
2579
2580
        if (qps_changed) {
2581
            memcpy(s->qps,      s1->qps,      sizeof(s->qps));
2582
            memcpy(s->last_qps, s1->last_qps, sizeof(s->last_qps));
2583
            s->nqps = s1->nqps;
2584
        }
2585
    }
2586
2587
    return update_frames(dst);
2588
}
2589
#endif
2590
2591
164
static int vp3_decode_frame(AVCodecContext *avctx,
2592
                            void *data, int *got_frame,
2593
                            AVPacket *avpkt)
2594
{
2595
164
    AVFrame     *frame  = data;
2596
164
    const uint8_t *buf  = avpkt->data;
2597
164
    int buf_size        = avpkt->size;
2598
164
    Vp3DecodeContext *s = avctx->priv_data;
2599
    GetBitContext gb;
2600
    int i, ret;
2601
2602
164
    if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0)
2603
        return ret;
2604
2605
#if CONFIG_THEORA_DECODER
2606

164
    if (s->theora && get_bits1(&gb)) {
2607
        int type = get_bits(&gb, 7);
2608
        skip_bits_long(&gb, 6*8); /* "theora" */
2609
2610
        if (s->avctx->active_thread_type&FF_THREAD_FRAME) {
2611
            av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n");
2612
            return AVERROR_PATCHWELCOME;
2613
        }
2614
        if (type == 0) {
2615
            vp3_decode_end(avctx);
2616
            ret = theora_decode_header(avctx, &gb);
2617
2618
            if (ret >= 0)
2619
                ret = vp3_decode_init(avctx);
2620
            if (ret < 0) {
2621
                vp3_decode_end(avctx);
2622
                return ret;
2623
            }
2624
            return buf_size;
2625
        } else if (type == 2) {
2626
            vp3_decode_end(avctx);
2627
            ret = theora_decode_tables(avctx, &gb);
2628
            if (ret >= 0)
2629
                ret = vp3_decode_init(avctx);
2630
            if (ret < 0) {
2631
                vp3_decode_end(avctx);
2632
                return ret;
2633
            }
2634
            return buf_size;
2635
        }
2636
2637
        av_log(avctx, AV_LOG_ERROR,
2638
               "Header packet passed to frame decoder, skipping\n");
2639
        return -1;
2640
    }
2641
#endif
2642
2643
164
    s->keyframe = !get_bits1(&gb);
2644
164
    if (!s->all_fragments) {
2645
        av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n");
2646
        return -1;
2647
    }
2648
164
    if (!s->theora)
2649
138
        skip_bits(&gb, 1);
2650
656
    for (i = 0; i < 3; i++)
2651
492
        s->last_qps[i] = s->qps[i];
2652
2653
164
    s->nqps = 0;
2654
    do {
2655
164
        s->qps[s->nqps++] = get_bits(&gb, 6);
2656

164
    } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2657
492
    for (i = s->nqps; i < 3; i++)
2658
328
        s->qps[i] = -1;
2659
2660
164
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2661
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2662
               s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2663
2664

302
    s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2665
138
                          avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2666
138
                                                                  : AVDISCARD_NONKEY);
2667
2668
164
    if (s->qps[0] != s->last_qps[0])
2669
62
        init_loop_filter(s);
2670
2671
328
    for (i = 0; i < s->nqps; i++)
2672
        // reinit all dequantizers if the first one changed, because
2673
        // the DC of the first quantizer must be used for all matrices
2674

164
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2675
62
            init_dequantizer(s, i);
2676
2677

164
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2678
        return buf_size;
2679
2680
328
    s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2681
164
                                                : AV_PICTURE_TYPE_P;
2682
164
    s->current_frame.f->key_frame = s->keyframe;
2683
164
    if ((ret = ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF)) < 0)
2684
        goto error;
2685
2686
164
    if (!s->edge_emu_buffer)
2687
5
        s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2688
2689
164
    if (s->keyframe) {
2690
6
        if (!s->theora) {
2691
3
            skip_bits(&gb, 4); /* width code */
2692
3
            skip_bits(&gb, 4); /* height code */
2693
3
            if (s->version) {
2694
3
                s->version = get_bits(&gb, 5);
2695
3
                if (avctx->frame_number == 0)
2696
2
                    av_log(s->avctx, AV_LOG_DEBUG,
2697
                           "VP version: %d\n", s->version);
2698
            }
2699
        }
2700

6
        if (s->version || s->theora) {
2701
6
            if (get_bits1(&gb))
2702
                av_log(s->avctx, AV_LOG_ERROR,
2703
                       "Warning, unsupported keyframe coding type?!\n");
2704
6
            skip_bits(&gb, 2); /* reserved? */
2705
2706
#if CONFIG_VP4_DECODER
2707
6
            if (s->version >= 2) {
2708
                int mb_height, mb_width;
2709
                int mb_width_mul, mb_width_div, mb_height_mul, mb_height_div;
2710
2711
2
                mb_height = get_bits(&gb, 8);
2712
2
                mb_width  = get_bits(&gb, 8);
2713
2
                if (mb_height != s->macroblock_height ||
2714
2
                    mb_width != s->macroblock_width)
2715
                    avpriv_request_sample(s->avctx, "macroblock dimension mismatch");
2716
2717
2
                mb_width_mul = get_bits(&gb, 5);
2718
2
                mb_width_div = get_bits(&gb, 3);
2719
2
                mb_height_mul = get_bits(&gb, 5);
2720
2
                mb_height_div = get_bits(&gb, 3);
2721


2
                if (mb_width_mul != 1 || mb_width_div != 1 || mb_height_mul != 1 || mb_height_div != 1)
2722
                    avpriv_request_sample(s->avctx, "unexpected macroblock dimension multipler/divider");
2723
2724
2
                if (get_bits(&gb, 2))
2725
                    avpriv_request_sample(s->avctx, "unknown bits");
2726
            }
2727
#endif
2728
        }
2729
    } else {
2730
158
        if (!s->golden_frame.f->data[0]) {
2731
            av_log(s->avctx, AV_LOG_WARNING,
2732
                   "vp3: first frame not a keyframe\n");
2733
2734
            s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2735
            if ((ret = ff_thread_get_buffer(avctx, &s->golden_frame,
2736
                                     AV_GET_BUFFER_FLAG_REF)) < 0)
2737
                goto error;
2738
            ff_thread_release_buffer(avctx, &s->last_frame);
2739
            if ((ret = ff_thread_ref_frame(&s->last_frame,
2740
                                           &s->golden_frame)) < 0)
2741
                goto error;
2742
            ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2743
        }
2744
    }
2745
2746
164
    memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2747
164
    ff_thread_finish_setup(avctx);
2748
2749
164
    if (s->version < 2) {
2750
140
        if ((ret = unpack_superblocks(s, &gb)) < 0) {
2751
            av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2752
            goto error;
2753
        }
2754
#if CONFIG_VP4_DECODER
2755
    } else {
2756
24
        if ((ret = vp4_unpack_macroblocks(s, &gb)) < 0) {
2757
            av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_macroblocks\n");
2758
            goto error;
2759
    }
2760
#endif
2761
    }
2762
164
    if ((ret = unpack_modes(s, &gb)) < 0) {
2763
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2764
        goto error;
2765
    }
2766
164
    if (ret = unpack_vectors(s, &gb)) {
2767
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2768
        goto error;
2769
    }
2770
164
    if ((ret = unpack_block_qpis(s, &gb)) < 0) {
2771
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2772
        goto error;
2773
    }
2774
2775
164
    if (s->version < 2) {
2776
140
        if ((ret = unpack_dct_coeffs(s, &gb)) < 0) {
2777
            av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2778
            goto error;
2779
        }
2780
#if CONFIG_VP4_DECODER
2781
    } else {
2782
24
        if ((ret = vp4_unpack_dct_coeffs(s, &gb)) < 0) {
2783
            av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_dct_coeffs\n");
2784
            goto error;
2785
        }
2786
#endif
2787
    }
2788
2789
656
    for (i = 0; i < 3; i++) {
2790

492
        int height = s->height >> (i && s->chroma_y_shift);
2791
492
        if (s->flipped_image)
2792
            s->data_offset[i] = 0;
2793
        else
2794
492
            s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2795
    }
2796
2797
164
    s->last_slice_end = 0;
2798
1232
    for (i = 0; i < s->c_superblock_height; i++)
2799
1068
        render_slice(s, i);
2800
2801
    // filter the last row
2802
164
    if (s->version < 2)
2803
560
        for (i = 0; i < 3; i++) {
2804

420
            int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2805
420
            apply_loop_filter(s, i, row, row + 1);
2806
        }
2807
164
    vp3_draw_horiz_band(s, s->height);
2808
2809
    /* output frame, offset as needed */
2810
164
    if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2811
        return ret;
2812
2813
164
    frame->crop_left   = s->offset_x;
2814
164
    frame->crop_right  = avctx->coded_width - avctx->width - s->offset_x;
2815
164
    frame->crop_top    = s->offset_y;
2816
164
    frame->crop_bottom = avctx->coded_height - avctx->height - s->offset_y;
2817
2818
164
    *got_frame = 1;
2819
2820
164
    if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2821
164
        ret = update_frames(avctx);
2822
164
        if (ret < 0)
2823
            return ret;
2824
    }
2825
2826
164
    return buf_size;
2827
2828
error:
2829
    ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2830
2831
    if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2832
        av_frame_unref(s->current_frame.f);
2833
2834
    return ret;
2835
}
2836
2837
35280
static int read_huffman_tree(HuffTable *huff, GetBitContext *gb, int length,
2838
                             AVCodecContext *avctx)
2839
{
2840
35280
    if (get_bits1(gb)) {
2841
        int token;
2842
17920
        if (huff->nb_entries >= 32) { /* overflow */
2843
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2844
            return -1;
2845
        }
2846
17920
        token = get_bits(gb, 5);
2847
        ff_dlog(avctx, "code length %d, curr entry %d, token %d\n",
2848
                length, huff->nb_entries, token);
2849
17920
        huff->entries[huff->nb_entries++] = (HuffEntry){ length, token };
2850
    } else {
2851
        /* The following bound follows from the fact that nb_entries <= 32. */
2852
17360
        if (length >= 31) { /* overflow */
2853
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2854
            return -1;
2855
        }
2856
17360
        length++;
2857
17360
        if (read_huffman_tree(huff, gb, length, avctx))
2858
            return -1;
2859
17360
        if (read_huffman_tree(huff, gb, length, avctx))
2860
            return -1;
2861
    }
2862
35280
    return 0;
2863
}
2864
2865
#if CONFIG_THEORA_DECODER
2866
static const enum AVPixelFormat theora_pix_fmts[4] = {
2867
    AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2868
};
2869
2870
7
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2871
{
2872
7
    Vp3DecodeContext *s = avctx->priv_data;
2873
    int visible_width, visible_height, colorspace;
2874
7
    uint8_t offset_x = 0, offset_y = 0;
2875
    int ret;
2876
    AVRational fps, aspect;
2877
2878
7
    s->theora_header = 0;
2879
7
    s->theora = get_bits(gb, 24);
2880
7
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2881
7
    if (!s->theora) {
2882
        s->theora = 1;
2883
        avpriv_request_sample(s->avctx, "theora 0");
2884
    }
2885
2886
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2887
     * but previous versions have the image flipped relative to vp3 */
2888
7
    if (s->theora < 0x030200) {
2889
        s->flipped_image = 1;
2890
        av_log(avctx, AV_LOG_DEBUG,
2891
               "Old (<alpha3) Theora bitstream, flipped image\n");
2892
    }
2893
2894
7
    visible_width  =
2895
7
    s->width       = get_bits(gb, 16) << 4;
2896
7
    visible_height =
2897
7
    s->height      = get_bits(gb, 16) << 4;
2898
2899
7
    if (s->theora >= 0x030200) {
2900
7
        visible_width  = get_bits(gb, 24);
2901
7
        visible_height = get_bits(gb, 24);
2902
2903
7
        offset_x = get_bits(gb, 8); /* offset x */
2904
7
        offset_y = get_bits(gb, 8); /* offset y, from bottom */
2905
    }
2906
2907
    /* sanity check */
2908
7
    if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 ||
2909
7
        visible_width  + offset_x > s->width ||
2910
7
        visible_height + offset_y > s->height) {
2911
        av_log(avctx, AV_LOG_ERROR,
2912
               "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n",
2913
               visible_width, visible_height, offset_x, offset_y,
2914
               s->width, s->height);
2915
        return AVERROR_INVALIDDATA;
2916
    }
2917
2918
7
    fps.num = get_bits_long(gb, 32);
2919
7
    fps.den = get_bits_long(gb, 32);
2920

7
    if (fps.num && fps.den) {
2921

7
        if (fps.num < 0 || fps.den < 0) {
2922
            av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2923
            return AVERROR_INVALIDDATA;
2924
        }
2925
7
        av_reduce(&avctx->framerate.den, &avctx->framerate.num,
2926
7
                  fps.den, fps.num, 1 << 30);
2927
    }
2928
2929
7
    aspect.num = get_bits(gb, 24);
2930
7
    aspect.den = get_bits(gb, 24);
2931

7
    if (aspect.num && aspect.den) {
2932
5
        av_reduce(&avctx->sample_aspect_ratio.num,
2933
                  &avctx->sample_aspect_ratio.den,
2934
5
                  aspect.num, aspect.den, 1 << 30);
2935
5
        ff_set_sar(avctx, avctx->sample_aspect_ratio);
2936
    }
2937
2938
7
    if (s->theora < 0x030200)
2939
        skip_bits(gb, 5); /* keyframe frequency force */
2940
7
    colorspace = get_bits(gb, 8);
2941
7
    skip_bits(gb, 24); /* bitrate */
2942
2943
7
    skip_bits(gb, 6); /* quality hint */
2944
2945
7
    if (s->theora >= 0x030200) {
2946
7
        skip_bits(gb, 5); /* keyframe frequency force */
2947
7
        avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2948
7
        if (avctx->pix_fmt == AV_PIX_FMT_NONE) {
2949
            av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n");
2950
            return AVERROR_INVALIDDATA;
2951
        }
2952
7
        skip_bits(gb, 3); /* reserved */
2953
    } else
2954
        avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2955
2956
7
    ret = ff_set_dimensions(avctx, s->width, s->height);
2957
7
    if (ret < 0)
2958
        return ret;
2959
7
    if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP)) {
2960
7
        avctx->width  = visible_width;
2961
7
        avctx->height = visible_height;
2962
        // translate offsets from theora axis ([0,0] lower left)
2963
        // to normal axis ([0,0] upper left)
2964
7
        s->offset_x = offset_x;
2965
7
        s->offset_y = s->height - visible_height - offset_y;
2966
    }
2967
2968
7
    if (colorspace == 1)
2969
        avctx->color_primaries = AVCOL_PRI_BT470M;
2970
7
    else if (colorspace == 2)
2971
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2972
2973

7
    if (colorspace == 1 || colorspace == 2) {
2974
        avctx->colorspace = AVCOL_SPC_BT470BG;
2975
        avctx->color_trc  = AVCOL_TRC_BT709;
2976
    }
2977
2978
7
    s->theora_header = 1;
2979
7
    return 0;
2980
}
2981
2982
7
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2983
{
2984
7
    Vp3DecodeContext *s = avctx->priv_data;
2985
    int i, n, matrices, inter, plane, ret;
2986
2987
7
    if (!s->theora_header)
2988
        return AVERROR_INVALIDDATA;
2989
2990
7
    if (s->theora >= 0x030200) {
2991
7
        n = get_bits(gb, 3);
2992
        /* loop filter limit values table */
2993
7
        if (n)
2994
455
            for (i = 0; i < 64; i++)
2995
448
                s->filter_limit_values[i] = get_bits(gb, n);
2996
    }
2997
2998
7
    if (s->theora >= 0x030200)
2999
7
        n = get_bits(gb, 4) + 1;
3000
    else
3001
        n = 16;
3002
    /* quality threshold table */
3003
455
    for (i = 0; i < 64; i++)
3004
448
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
3005
3006
7
    if (s->theora >= 0x030200)
3007
7
        n = get_bits(gb, 4) + 1;
3008
    else
3009
        n = 16;
3010
    /* dc scale factor table */
3011
455
    for (i = 0; i < 64; i++)
3012
448
        s->coded_dc_scale_factor[0][i] =
3013
448
        s->coded_dc_scale_factor[1][i] = get_bits(gb, n);
3014
3015
7
    if (s->theora >= 0x030200)
3016
7
        matrices = get_bits(gb, 9) + 1;
3017
    else
3018
        matrices = 3;
3019
3020
7
    if (matrices > 384) {
3021
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
3022
        return -1;
3023
    }
3024
3025
52
    for (n = 0; n < matrices; n++)
3026
2925
        for (i = 0; i < 64; i++)
3027
2880
            s->base_matrix[n][i] = get_bits(gb, 8);
3028
3029
21
    for (inter = 0; inter <= 1; inter++) {
3030
56
        for (plane = 0; plane <= 2; plane++) {
3031
42
            int newqr = 1;
3032

42
            if (inter || plane > 0)
3033
35
                newqr = get_bits1(gb);
3034
42
            if (!newqr) {
3035
                int qtj, plj;
3036

23
                if (inter && get_bits1(gb)) {
3037
6
                    qtj = 0;
3038
6
                    plj = plane;
3039
                } else {
3040
17
                    qtj = (3 * inter + plane - 1) / 3;
3041
17
                    plj = (plane + 2) % 3;
3042
                }
3043
23
                s->qr_count[inter][plane] = s->qr_count[qtj][plj];
3044
23
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
3045
                       sizeof(s->qr_size[0][0]));
3046
23
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
3047
                       sizeof(s->qr_base[0][0]));
3048
            } else {
3049
19
                int qri = 0;
3050
19
                int qi  = 0;
3051
3052
                for (;;) {
3053
62
                    i = get_bits(gb, av_log2(matrices - 1) + 1);
3054
62
                    if (i >= matrices) {
3055
                        av_log(avctx, AV_LOG_ERROR,
3056
                               "invalid base matrix index\n");
3057
                        return -1;
3058
                    }
3059
62
                    s->qr_base[inter][plane][qri] = i;
3060
62
                    if (qi >= 63)
3061
19
                        break;
3062
43
                    i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
3063
43
                    s->qr_size[inter][plane][qri++] = i;
3064
43
                    qi += i;
3065
                }
3066
3067
19
                if (qi > 63) {
3068
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
3069
                    return -1;
3070
                }
3071
19
                s->qr_count[inter][plane] = qri;
3072
            }
3073
        }
3074
    }
3075
3076
    /* Huffman tables */
3077
567
    for (int i = 0; i < FF_ARRAY_ELEMS(s->huffman_table); i++) {
3078
560
        s->huffman_table[i].nb_entries = 0;
3079
560
        if ((ret = read_huffman_tree(&s->huffman_table[i], gb, 0, avctx)) < 0)
3080
            return ret;
3081
    }
3082
3083
7
    s->theora_tables = 1;
3084
3085
7
    return 0;
3086
}
3087
3088
7
static av_cold int theora_decode_init(AVCodecContext *avctx)
3089
{
3090
7
    Vp3DecodeContext *s = avctx->priv_data;
3091
    GetBitContext gb;
3092
    int ptype;
3093
    const uint8_t *header_start[3];
3094
    int header_len[3];
3095
    int i;
3096
    int ret;
3097
3098
7
    avctx->pix_fmt = AV_PIX_FMT_YUV420P;
3099
3100
7
    s->theora = 1;
3101
3102
7
    if (!avctx->extradata_size) {
3103
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
3104
        return -1;
3105
    }
3106
3107
7
    if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
3108
                                  42, header_start, header_len) < 0) {
3109
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
3110
        return -1;
3111
    }
3112
3113
28
    for (i = 0; i < 3; i++) {
3114
21
        if (header_len[i] <= 0)
3115
            continue;
3116
21
        ret = init_get_bits8(&gb, header_start[i], header_len[i]);
3117
21
        if (ret < 0)
3118
            return ret;
3119
3120
21
        ptype = get_bits(&gb, 8);
3121
3122
21
        if (!(ptype & 0x80)) {
3123
            av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
3124
//          return -1;
3125
        }
3126
3127
        // FIXME: Check for this as well.
3128
21
        skip_bits_long(&gb, 6 * 8); /* "theora" */
3129
3130

21
        switch (ptype) {
3131
7
        case 0x80:
3132
7
            if (theora_decode_header(avctx, &gb) < 0)
3133
                return -1;
3134
7
            break;
3135
7
        case 0x81:
3136
// FIXME: is this needed? it breaks sometimes
3137
//            theora_decode_comments(avctx, gb);
3138
7
            break;
3139
7
        case 0x82:
3140
7
            if (theora_decode_tables(avctx, &gb))
3141
                return -1;
3142
7
            break;
3143
        default:
3144
            av_log(avctx, AV_LOG_ERROR,
3145
                   "Unknown Theora config packet: %d\n", ptype & ~0x80);
3146
            break;
3147
        }
3148

21
        if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb))
3149
7
            av_log(avctx, AV_LOG_WARNING,
3150
                   "%d bits left in packet %X\n",
3151
7
                   8 * header_len[i] - get_bits_count(&gb), ptype);
3152
21
        if (s->theora < 0x030200)
3153
            break;
3154
    }
3155
3156
7
    return vp3_decode_init(avctx);
3157
}
3158
3159
AVCodec ff_theora_decoder = {
3160
    .name                  = "theora",
3161
    .long_name             = NULL_IF_CONFIG_SMALL("Theora"),
3162
    .type                  = AVMEDIA_TYPE_VIDEO,
3163
    .id                    = AV_CODEC_ID_THEORA,
3164
    .priv_data_size        = sizeof(Vp3DecodeContext),
3165
    .init                  = theora_decode_init,
3166
    .close                 = vp3_decode_end,
3167
    .decode                = vp3_decode_frame,
3168
    .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3169
                             AV_CODEC_CAP_FRAME_THREADS,
3170
    .flush                 = vp3_decode_flush,
3171
    .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3172
    .caps_internal         = FF_CODEC_CAP_EXPORTS_CROPPING | FF_CODEC_CAP_ALLOCATE_PROGRESS |
3173
                             FF_CODEC_CAP_INIT_CLEANUP,
3174
};
3175
#endif
3176
3177
AVCodec ff_vp3_decoder = {
3178
    .name                  = "vp3",
3179
    .long_name             = NULL_IF_CONFIG_SMALL("On2 VP3"),
3180
    .type                  = AVMEDIA_TYPE_VIDEO,
3181
    .id                    = AV_CODEC_ID_VP3,
3182
    .priv_data_size        = sizeof(Vp3DecodeContext),
3183
    .init                  = vp3_decode_init,
3184
    .close                 = vp3_decode_end,
3185
    .decode                = vp3_decode_frame,
3186
    .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3187
                             AV_CODEC_CAP_FRAME_THREADS,
3188
    .flush                 = vp3_decode_flush,
3189
    .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3190
    .caps_internal         = FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
3191
};
3192
3193
#if CONFIG_VP4_DECODER
3194
AVCodec ff_vp4_decoder = {
3195
    .name                  = "vp4",
3196
    .long_name             = NULL_IF_CONFIG_SMALL("On2 VP4"),
3197
    .type                  = AVMEDIA_TYPE_VIDEO,
3198
    .id                    = AV_CODEC_ID_VP4,
3199
    .priv_data_size        = sizeof(Vp3DecodeContext),
3200
    .init                  = vp3_decode_init,
3201
    .close                 = vp3_decode_end,
3202
    .decode                = vp3_decode_frame,
3203
    .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3204
                             AV_CODEC_CAP_FRAME_THREADS,
3205
    .flush                 = vp3_decode_flush,
3206
    .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3207
    .caps_internal         = FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
3208
};
3209
#endif