xref: /third_party/ffmpeg/libavcodec/vp3.c (revision cabdff1a)
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 "config_components.h"
34
35#include <stdio.h>
36#include <stdlib.h>
37#include <string.h>
38
39#include "libavutil/imgutils.h"
40#include "libavutil/mem_internal.h"
41
42#include "avcodec.h"
43#include "codec_internal.h"
44#include "get_bits.h"
45#include "hpeldsp.h"
46#include "internal.h"
47#include "mathops.h"
48#include "thread.h"
49#include "threadframe.h"
50#include "videodsp.h"
51#include "vp3data.h"
52#include "vp4data.h"
53#include "vp3dsp.h"
54#include "xiph.h"
55
56#define VP3_MV_VLC_BITS     6
57#define VP4_MV_VLC_BITS     6
58#define SUPERBLOCK_VLC_BITS 6
59
60#define FRAGMENT_PIXELS 8
61
62// FIXME split things out into their own arrays
63typedef struct Vp3Fragment {
64    int16_t dc;
65    uint8_t coding_method;
66    uint8_t qpi;
67} Vp3Fragment;
68
69#define SB_NOT_CODED        0
70#define SB_PARTIALLY_CODED  1
71#define SB_FULLY_CODED      2
72
73// This is the maximum length of a single long bit run that can be encoded
74// for superblock coding or block qps. Theora special-cases this to read a
75// bit instead of flipping the current bit to allow for runs longer than 4129.
76#define MAXIMUM_LONG_BIT_RUN 4129
77
78#define MODE_INTER_NO_MV      0
79#define MODE_INTRA            1
80#define MODE_INTER_PLUS_MV    2
81#define MODE_INTER_LAST_MV    3
82#define MODE_INTER_PRIOR_LAST 4
83#define MODE_USING_GOLDEN     5
84#define MODE_GOLDEN_MV        6
85#define MODE_INTER_FOURMV     7
86#define CODING_MODE_COUNT     8
87
88/* special internal mode */
89#define MODE_COPY             8
90
91static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb);
92static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
93
94
95/* There are 6 preset schemes, plus a free-form scheme */
96static const int ModeAlphabet[6][CODING_MODE_COUNT] = {
97    /* scheme 1: Last motion vector dominates */
98    { MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
99      MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
100      MODE_INTRA,            MODE_USING_GOLDEN,
101      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
102
103    /* scheme 2 */
104    { MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
105      MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
106      MODE_INTRA,            MODE_USING_GOLDEN,
107      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
108
109    /* scheme 3 */
110    { MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
111      MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
112      MODE_INTRA,            MODE_USING_GOLDEN,
113      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
114
115    /* scheme 4 */
116    { MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
117      MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
118      MODE_INTRA,            MODE_USING_GOLDEN,
119      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
120
121    /* scheme 5: No motion vector dominates */
122    { MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
123      MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
124      MODE_INTRA,            MODE_USING_GOLDEN,
125      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
126
127    /* scheme 6 */
128    { MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
129      MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
130      MODE_INTER_PLUS_MV,    MODE_INTRA,
131      MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
132};
133
134static const uint8_t hilbert_offset[16][2] = {
135    { 0, 0 }, { 1, 0 }, { 1, 1 }, { 0, 1 },
136    { 0, 2 }, { 0, 3 }, { 1, 3 }, { 1, 2 },
137    { 2, 2 }, { 2, 3 }, { 3, 3 }, { 3, 2 },
138    { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 }
139};
140
141enum {
142    VP4_DC_INTRA  = 0,
143    VP4_DC_INTER  = 1,
144    VP4_DC_GOLDEN = 2,
145    NB_VP4_DC_TYPES,
146    VP4_DC_UNDEFINED = NB_VP4_DC_TYPES
147};
148
149static const uint8_t vp4_pred_block_type_map[8] = {
150    [MODE_INTER_NO_MV]      = VP4_DC_INTER,
151    [MODE_INTRA]            = VP4_DC_INTRA,
152    [MODE_INTER_PLUS_MV]    = VP4_DC_INTER,
153    [MODE_INTER_LAST_MV]    = VP4_DC_INTER,
154    [MODE_INTER_PRIOR_LAST] = VP4_DC_INTER,
155    [MODE_USING_GOLDEN]     = VP4_DC_GOLDEN,
156    [MODE_GOLDEN_MV]        = VP4_DC_GOLDEN,
157    [MODE_INTER_FOURMV]     = VP4_DC_INTER,
158};
159
160typedef struct {
161    int dc;
162    int type;
163} VP4Predictor;
164
165#define MIN_DEQUANT_VAL 2
166
167typedef struct HuffEntry {
168    uint8_t len, sym;
169} HuffEntry;
170
171typedef struct HuffTable {
172    HuffEntry entries[32];
173    uint8_t   nb_entries;
174} HuffTable;
175
176typedef struct Vp3DecodeContext {
177    AVCodecContext *avctx;
178    int theora, theora_tables, theora_header;
179    int version;
180    int width, height;
181    int chroma_x_shift, chroma_y_shift;
182    ThreadFrame golden_frame;
183    ThreadFrame last_frame;
184    ThreadFrame current_frame;
185    int keyframe;
186    uint8_t idct_permutation[64];
187    uint8_t idct_scantable[64];
188    HpelDSPContext hdsp;
189    VideoDSPContext vdsp;
190    VP3DSPContext vp3dsp;
191    DECLARE_ALIGNED(16, int16_t, block)[64];
192    int flipped_image;
193    int last_slice_end;
194    int skip_loop_filter;
195
196    int qps[3];
197    int nqps;
198    int last_qps[3];
199
200    int superblock_count;
201    int y_superblock_width;
202    int y_superblock_height;
203    int y_superblock_count;
204    int c_superblock_width;
205    int c_superblock_height;
206    int c_superblock_count;
207    int u_superblock_start;
208    int v_superblock_start;
209    unsigned char *superblock_coding;
210
211    int macroblock_count; /* y macroblock count */
212    int macroblock_width;
213    int macroblock_height;
214    int c_macroblock_count;
215    int c_macroblock_width;
216    int c_macroblock_height;
217    int yuv_macroblock_count; /* y+u+v macroblock count */
218
219    int fragment_count;
220    int fragment_width[2];
221    int fragment_height[2];
222
223    Vp3Fragment *all_fragments;
224    int fragment_start[3];
225    int data_offset[3];
226    uint8_t offset_x;
227    uint8_t offset_y;
228    int offset_x_warned;
229
230    int8_t (*motion_val[2])[2];
231
232    /* tables */
233    uint16_t coded_dc_scale_factor[2][64];
234    uint32_t coded_ac_scale_factor[64];
235    uint8_t base_matrix[384][64];
236    uint8_t qr_count[2][3];
237    uint8_t qr_size[2][3][64];
238    uint16_t qr_base[2][3][64];
239
240    /**
241     * This is a list of all tokens in bitstream order. Reordering takes place
242     * by pulling from each level during IDCT. As a consequence, IDCT must be
243     * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
244     * otherwise. The 32 different tokens with up to 12 bits of extradata are
245     * collapsed into 3 types, packed as follows:
246     *   (from the low to high bits)
247     *
248     * 2 bits: type (0,1,2)
249     *   0: EOB run, 14 bits for run length (12 needed)
250     *   1: zero run, 7 bits for run length
251     *                7 bits for the next coefficient (3 needed)
252     *   2: coefficient, 14 bits (11 needed)
253     *
254     * Coefficients are signed, so are packed in the highest bits for automatic
255     * sign extension.
256     */
257    int16_t *dct_tokens[3][64];
258    int16_t *dct_tokens_base;
259#define TOKEN_EOB(eob_run)              ((eob_run) << 2)
260#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) * 512) + ((zero_run) << 2) + 1)
261#define TOKEN_COEFF(coeff)              (((coeff) * 4) + 2)
262
263    /**
264     * number of blocks that contain DCT coefficients at
265     * the given level or higher
266     */
267    int num_coded_frags[3][64];
268    int total_num_coded_frags;
269
270    /* this is a list of indexes into the all_fragments array indicating
271     * which of the fragments are coded */
272    int *coded_fragment_list[3];
273
274    int *kf_coded_fragment_list;
275    int *nkf_coded_fragment_list;
276    int num_kf_coded_fragment[3];
277
278    /* The first 16 of the following VLCs are for the dc coefficients;
279       the others are four groups of 16 VLCs each for ac coefficients. */
280    VLC coeff_vlc[5 * 16];
281
282    VLC superblock_run_length_vlc; /* version < 2 */
283    VLC fragment_run_length_vlc; /* version < 2 */
284    VLC block_pattern_vlc[2]; /* version >= 2*/
285    VLC mode_code_vlc;
286    VLC motion_vector_vlc; /* version < 2 */
287    VLC vp4_mv_vlc[2][7]; /* version >=2 */
288
289    /* these arrays need to be on 16-byte boundaries since SSE2 operations
290     * index into them */
291    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     ///< qmat[qpi][is_inter][plane]
292
293    /* This table contains superblock_count * 16 entries. Each set of 16
294     * numbers corresponds to the fragment indexes 0..15 of the superblock.
295     * An entry will be -1 to indicate that no entry corresponds to that
296     * index. */
297    int *superblock_fragments;
298
299    /* This is an array that indicates how a particular macroblock
300     * is coded. */
301    unsigned char *macroblock_coding;
302
303    uint8_t *edge_emu_buffer;
304
305    /* Huffman decode */
306    HuffTable huffman_table[5 * 16];
307
308    uint8_t filter_limit_values[64];
309    DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
310
311    VP4Predictor * dc_pred_row; /* dc_pred_row[y_superblock_width * 4] */
312} Vp3DecodeContext;
313
314/************************************************************************
315 * VP3 specific functions
316 ************************************************************************/
317
318static av_cold void free_tables(AVCodecContext *avctx)
319{
320    Vp3DecodeContext *s = avctx->priv_data;
321
322    av_freep(&s->superblock_coding);
323    av_freep(&s->all_fragments);
324    av_freep(&s->nkf_coded_fragment_list);
325    av_freep(&s->kf_coded_fragment_list);
326    av_freep(&s->dct_tokens_base);
327    av_freep(&s->superblock_fragments);
328    av_freep(&s->macroblock_coding);
329    av_freep(&s->dc_pred_row);
330    av_freep(&s->motion_val[0]);
331    av_freep(&s->motion_val[1]);
332}
333
334static void vp3_decode_flush(AVCodecContext *avctx)
335{
336    Vp3DecodeContext *s = avctx->priv_data;
337
338    if (s->golden_frame.f)
339        ff_thread_release_ext_buffer(avctx, &s->golden_frame);
340    if (s->last_frame.f)
341        ff_thread_release_ext_buffer(avctx, &s->last_frame);
342    if (s->current_frame.f)
343        ff_thread_release_ext_buffer(avctx, &s->current_frame);
344}
345
346static av_cold int vp3_decode_end(AVCodecContext *avctx)
347{
348    Vp3DecodeContext *s = avctx->priv_data;
349    int i, j;
350
351    free_tables(avctx);
352    av_freep(&s->edge_emu_buffer);
353
354    s->theora_tables = 0;
355
356    /* release all frames */
357    vp3_decode_flush(avctx);
358    av_frame_free(&s->current_frame.f);
359    av_frame_free(&s->last_frame.f);
360    av_frame_free(&s->golden_frame.f);
361
362    for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++)
363        ff_free_vlc(&s->coeff_vlc[i]);
364
365    ff_free_vlc(&s->superblock_run_length_vlc);
366    ff_free_vlc(&s->fragment_run_length_vlc);
367    ff_free_vlc(&s->mode_code_vlc);
368    ff_free_vlc(&s->motion_vector_vlc);
369
370    for (j = 0; j < 2; j++)
371        for (i = 0; i < 7; i++)
372            ff_free_vlc(&s->vp4_mv_vlc[j][i]);
373
374    for (i = 0; i < 2; i++)
375        ff_free_vlc(&s->block_pattern_vlc[i]);
376    return 0;
377}
378
379/**
380 * This function sets up all of the various blocks mappings:
381 * superblocks <-> fragments, macroblocks <-> fragments,
382 * superblocks <-> macroblocks
383 *
384 * @return 0 is successful; returns 1 if *anything* went wrong.
385 */
386static int init_block_mapping(Vp3DecodeContext *s)
387{
388    int sb_x, sb_y, plane;
389    int x, y, i, j = 0;
390
391    for (plane = 0; plane < 3; plane++) {
392        int sb_width    = plane ? s->c_superblock_width
393                                : s->y_superblock_width;
394        int sb_height   = plane ? s->c_superblock_height
395                                : s->y_superblock_height;
396        int frag_width  = s->fragment_width[!!plane];
397        int frag_height = s->fragment_height[!!plane];
398
399        for (sb_y = 0; sb_y < sb_height; sb_y++)
400            for (sb_x = 0; sb_x < sb_width; sb_x++)
401                for (i = 0; i < 16; i++) {
402                    x = 4 * sb_x + hilbert_offset[i][0];
403                    y = 4 * sb_y + hilbert_offset[i][1];
404
405                    if (x < frag_width && y < frag_height)
406                        s->superblock_fragments[j++] = s->fragment_start[plane] +
407                                                       y * frag_width + x;
408                    else
409                        s->superblock_fragments[j++] = -1;
410                }
411    }
412
413    return 0;  /* successful path out */
414}
415
416/*
417 * This function sets up the dequantization tables used for a particular
418 * frame.
419 */
420static void init_dequantizer(Vp3DecodeContext *s, int qpi)
421{
422    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
423    int i, plane, inter, qri, bmi, bmj, qistart;
424
425    for (inter = 0; inter < 2; inter++) {
426        for (plane = 0; plane < 3; plane++) {
427            int dc_scale_factor = s->coded_dc_scale_factor[!!plane][s->qps[qpi]];
428            int sum = 0;
429            for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
430                sum += s->qr_size[inter][plane][qri];
431                if (s->qps[qpi] <= sum)
432                    break;
433            }
434            qistart = sum - s->qr_size[inter][plane][qri];
435            bmi     = s->qr_base[inter][plane][qri];
436            bmj     = s->qr_base[inter][plane][qri + 1];
437            for (i = 0; i < 64; i++) {
438                int coeff = (2 * (sum     - s->qps[qpi]) * s->base_matrix[bmi][i] -
439                             2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
440                             s->qr_size[inter][plane][qri]) /
441                            (2 * s->qr_size[inter][plane][qri]);
442
443                int qmin   = 8 << (inter + !i);
444                int qscale = i ? ac_scale_factor : dc_scale_factor;
445                int qbias = (1 + inter) * 3;
446                s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
447                    (i == 0 || s->version < 2) ? av_clip((qscale * coeff) / 100 * 4, qmin, 4096)
448                                               : (qscale * (coeff - qbias) / 100 + qbias) * 4;
449            }
450            /* all DC coefficients use the same quant so as not to interfere
451             * with DC prediction */
452            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
453        }
454    }
455}
456
457/*
458 * This function initializes the loop filter boundary limits if the frame's
459 * quality index is different from the previous frame's.
460 *
461 * The filter_limit_values may not be larger than 127.
462 */
463static void init_loop_filter(Vp3DecodeContext *s)
464{
465    ff_vp3dsp_set_bounding_values(s->bounding_values_array, s->filter_limit_values[s->qps[0]]);
466}
467
468/*
469 * This function unpacks all of the superblock/macroblock/fragment coding
470 * information from the bitstream.
471 */
472static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
473{
474    int superblock_starts[3] = {
475        0, s->u_superblock_start, s->v_superblock_start
476    };
477    int bit = 0;
478    int current_superblock = 0;
479    int current_run = 0;
480    int num_partial_superblocks = 0;
481
482    int i, j;
483    int current_fragment;
484    int plane;
485    int plane0_num_coded_frags = 0;
486
487    if (s->keyframe) {
488        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
489    } else {
490        /* unpack the list of partially-coded superblocks */
491        bit         = get_bits1(gb) ^ 1;
492        current_run = 0;
493
494        while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
495            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
496                bit = get_bits1(gb);
497            else
498                bit ^= 1;
499
500            current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
501                                   SUPERBLOCK_VLC_BITS, 2);
502            if (current_run == 34)
503                current_run += get_bits(gb, 12);
504
505            if (current_run > s->superblock_count - current_superblock) {
506                av_log(s->avctx, AV_LOG_ERROR,
507                       "Invalid partially coded superblock run length\n");
508                return -1;
509            }
510
511            memset(s->superblock_coding + current_superblock, bit, current_run);
512
513            current_superblock += current_run;
514            if (bit)
515                num_partial_superblocks += current_run;
516        }
517
518        /* unpack the list of fully coded superblocks if any of the blocks were
519         * not marked as partially coded in the previous step */
520        if (num_partial_superblocks < s->superblock_count) {
521            int superblocks_decoded = 0;
522
523            current_superblock = 0;
524            bit                = get_bits1(gb) ^ 1;
525            current_run        = 0;
526
527            while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
528                   get_bits_left(gb) > 0) {
529                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
530                    bit = get_bits1(gb);
531                else
532                    bit ^= 1;
533
534                current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
535                                       SUPERBLOCK_VLC_BITS, 2);
536                if (current_run == 34)
537                    current_run += get_bits(gb, 12);
538
539                for (j = 0; j < current_run; current_superblock++) {
540                    if (current_superblock >= s->superblock_count) {
541                        av_log(s->avctx, AV_LOG_ERROR,
542                               "Invalid fully coded superblock run length\n");
543                        return -1;
544                    }
545
546                    /* skip any superblocks already marked as partially coded */
547                    if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
548                        s->superblock_coding[current_superblock] = 2 * bit;
549                        j++;
550                    }
551                }
552                superblocks_decoded += current_run;
553            }
554        }
555
556        /* if there were partial blocks, initialize bitstream for
557         * unpacking fragment codings */
558        if (num_partial_superblocks) {
559            current_run = 0;
560            bit         = get_bits1(gb);
561            /* toggle the bit because as soon as the first run length is
562             * fetched the bit will be toggled again */
563            bit ^= 1;
564        }
565    }
566
567    /* figure out which fragments are coded; iterate through each
568     * superblock (all planes) */
569    s->total_num_coded_frags = 0;
570    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
571
572    s->coded_fragment_list[0] = s->keyframe ? s->kf_coded_fragment_list
573                                            : s->nkf_coded_fragment_list;
574
575    for (plane = 0; plane < 3; plane++) {
576        int sb_start = superblock_starts[plane];
577        int sb_end   = sb_start + (plane ? s->c_superblock_count
578                                         : s->y_superblock_count);
579        int num_coded_frags = 0;
580
581        if (s->keyframe) {
582            if (s->num_kf_coded_fragment[plane] == -1) {
583                for (i = sb_start; i < sb_end; i++) {
584                    /* iterate through all 16 fragments in a superblock */
585                    for (j = 0; j < 16; j++) {
586                        /* if the fragment is in bounds, check its coding status */
587                        current_fragment = s->superblock_fragments[i * 16 + j];
588                        if (current_fragment != -1) {
589                            s->coded_fragment_list[plane][num_coded_frags++] =
590                                current_fragment;
591                        }
592                    }
593                }
594                s->num_kf_coded_fragment[plane] = num_coded_frags;
595            } else
596                num_coded_frags = s->num_kf_coded_fragment[plane];
597        } else {
598            for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
599                if (get_bits_left(gb) < plane0_num_coded_frags >> 2) {
600                    return AVERROR_INVALIDDATA;
601                }
602                /* iterate through all 16 fragments in a superblock */
603                for (j = 0; j < 16; j++) {
604                    /* if the fragment is in bounds, check its coding status */
605                    current_fragment = s->superblock_fragments[i * 16 + j];
606                    if (current_fragment != -1) {
607                        int coded = s->superblock_coding[i];
608
609                        if (coded == SB_PARTIALLY_CODED) {
610                            /* fragment may or may not be coded; this is the case
611                             * that cares about the fragment coding runs */
612                            if (current_run-- == 0) {
613                                bit        ^= 1;
614                                current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2);
615                            }
616                            coded = bit;
617                        }
618
619                        if (coded) {
620                            /* default mode; actual mode will be decoded in
621                             * the next phase */
622                            s->all_fragments[current_fragment].coding_method =
623                                MODE_INTER_NO_MV;
624                            s->coded_fragment_list[plane][num_coded_frags++] =
625                                current_fragment;
626                        } else {
627                            /* not coded; copy this fragment from the prior frame */
628                            s->all_fragments[current_fragment].coding_method =
629                                MODE_COPY;
630                        }
631                    }
632                }
633            }
634        }
635        if (!plane)
636            plane0_num_coded_frags = num_coded_frags;
637        s->total_num_coded_frags += num_coded_frags;
638        for (i = 0; i < 64; i++)
639            s->num_coded_frags[plane][i] = num_coded_frags;
640        if (plane < 2)
641            s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] +
642                                                num_coded_frags;
643    }
644    return 0;
645}
646
647#define BLOCK_X (2 * mb_x + (k & 1))
648#define BLOCK_Y (2 * mb_y + (k >> 1))
649
650#if CONFIG_VP4_DECODER
651/**
652 * @return number of blocks, or > yuv_macroblock_count on error.
653 *         return value is always >= 1.
654 */
655static int vp4_get_mb_count(Vp3DecodeContext *s, GetBitContext *gb)
656{
657    int v = 1;
658    int bits;
659    while ((bits = show_bits(gb, 9)) == 0x1ff) {
660        skip_bits(gb, 9);
661        v += 256;
662        if (v > s->yuv_macroblock_count) {
663            av_log(s->avctx, AV_LOG_ERROR, "Invalid run length\n");
664            return v;
665        }
666    }
667#define body(n) { \
668    skip_bits(gb, 2 + n); \
669    v += (1 << n) + get_bits(gb, n); }
670#define thresh(n) (0x200 - (0x80 >> n))
671#define else_if(n) else if (bits < thresh(n)) body(n)
672    if (bits < 0x100) {
673        skip_bits(gb, 1);
674    } else if (bits < thresh(0)) {
675        skip_bits(gb, 2);
676        v += 1;
677    }
678    else_if(1)
679    else_if(2)
680    else_if(3)
681    else_if(4)
682    else_if(5)
683    else_if(6)
684    else body(7)
685#undef body
686#undef thresh
687#undef else_if
688    return v;
689}
690
691static int vp4_get_block_pattern(Vp3DecodeContext *s, GetBitContext *gb, int *next_block_pattern_table)
692{
693    int v = get_vlc2(gb, s->block_pattern_vlc[*next_block_pattern_table].table, 3, 2);
694    *next_block_pattern_table = vp4_block_pattern_table_selector[v];
695    return v + 1;
696}
697
698static int vp4_unpack_macroblocks(Vp3DecodeContext *s, GetBitContext *gb)
699{
700    int plane, i, j, k, fragment;
701    int next_block_pattern_table;
702    int bit, current_run, has_partial;
703
704    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
705
706    if (s->keyframe)
707        return 0;
708
709    has_partial = 0;
710    bit         = get_bits1(gb);
711    for (i = 0; i < s->yuv_macroblock_count; i += current_run) {
712        if (get_bits_left(gb) <= 0)
713            return AVERROR_INVALIDDATA;
714        current_run = vp4_get_mb_count(s, gb);
715        if (current_run > s->yuv_macroblock_count - i)
716            return -1;
717        memset(s->superblock_coding + i, 2 * bit, current_run);
718        bit ^= 1;
719        has_partial |= bit;
720    }
721
722    if (has_partial) {
723        if (get_bits_left(gb) <= 0)
724            return AVERROR_INVALIDDATA;
725        bit  = get_bits1(gb);
726        current_run = vp4_get_mb_count(s, gb);
727        for (i = 0; i < s->yuv_macroblock_count; i++) {
728            if (!s->superblock_coding[i]) {
729                if (!current_run) {
730                    bit ^= 1;
731                    current_run = vp4_get_mb_count(s, gb);
732                }
733                s->superblock_coding[i] = bit;
734                current_run--;
735            }
736        }
737        if (current_run) /* handle situation when vp4_get_mb_count() fails */
738            return -1;
739    }
740
741    next_block_pattern_table = 0;
742    i = 0;
743    for (plane = 0; plane < 3; plane++) {
744        int sb_x, sb_y;
745        int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
746        int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
747        int mb_width = plane ? s->c_macroblock_width : s->macroblock_width;
748        int mb_height = plane ? s->c_macroblock_height : s->macroblock_height;
749        int fragment_width = s->fragment_width[!!plane];
750        int fragment_height = s->fragment_height[!!plane];
751
752        for (sb_y = 0; sb_y < sb_height; sb_y++) {
753            for (sb_x = 0; sb_x < sb_width; sb_x++) {
754                for (j = 0; j < 4; j++) {
755                    int mb_x = 2 * sb_x + (j >> 1);
756                    int mb_y = 2 * sb_y + (j >> 1) ^ (j & 1);
757                    int mb_coded, pattern, coded;
758
759                    if (mb_x >= mb_width || mb_y >= mb_height)
760                        continue;
761
762                    mb_coded = s->superblock_coding[i++];
763
764                    if (mb_coded == SB_FULLY_CODED)
765                        pattern = 0xF;
766                    else if (mb_coded == SB_PARTIALLY_CODED)
767                        pattern = vp4_get_block_pattern(s, gb, &next_block_pattern_table);
768                    else
769                        pattern = 0;
770
771                    for (k = 0; k < 4; k++) {
772                        if (BLOCK_X >= fragment_width || BLOCK_Y >= fragment_height)
773                            continue;
774                        fragment = s->fragment_start[plane] + BLOCK_Y * fragment_width + BLOCK_X;
775                        coded = pattern & (8 >> k);
776                        /* MODE_INTER_NO_MV is the default for coded fragments.
777                           the actual method is decoded in the next phase. */
778                        s->all_fragments[fragment].coding_method = coded ? MODE_INTER_NO_MV : MODE_COPY;
779                    }
780                }
781            }
782        }
783    }
784    return 0;
785}
786#endif
787
788/*
789 * This function unpacks all the coding mode data for individual macroblocks
790 * from the bitstream.
791 */
792static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
793{
794    int i, j, k, sb_x, sb_y;
795    int scheme;
796    int current_macroblock;
797    int current_fragment;
798    int coding_mode;
799    int custom_mode_alphabet[CODING_MODE_COUNT];
800    const int *alphabet;
801    Vp3Fragment *frag;
802
803    if (s->keyframe) {
804        for (i = 0; i < s->fragment_count; i++)
805            s->all_fragments[i].coding_method = MODE_INTRA;
806    } else {
807        /* fetch the mode coding scheme for this frame */
808        scheme = get_bits(gb, 3);
809
810        /* is it a custom coding scheme? */
811        if (scheme == 0) {
812            for (i = 0; i < 8; i++)
813                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
814            for (i = 0; i < 8; i++)
815                custom_mode_alphabet[get_bits(gb, 3)] = i;
816            alphabet = custom_mode_alphabet;
817        } else
818            alphabet = ModeAlphabet[scheme - 1];
819
820        /* iterate through all of the macroblocks that contain 1 or more
821         * coded fragments */
822        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
823            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
824                if (get_bits_left(gb) <= 0)
825                    return -1;
826
827                for (j = 0; j < 4; j++) {
828                    int mb_x = 2 * sb_x + (j >> 1);
829                    int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
830                    current_macroblock = mb_y * s->macroblock_width + mb_x;
831
832                    if (mb_x >= s->macroblock_width ||
833                        mb_y >= s->macroblock_height)
834                        continue;
835
836                    /* coding modes are only stored if the macroblock has
837                     * at least one luma block coded, otherwise it must be
838                     * INTER_NO_MV */
839                    for (k = 0; k < 4; k++) {
840                        current_fragment = BLOCK_Y *
841                                           s->fragment_width[0] + BLOCK_X;
842                        if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
843                            break;
844                    }
845                    if (k == 4) {
846                        s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
847                        continue;
848                    }
849
850                    /* mode 7 means get 3 bits for each coding mode */
851                    if (scheme == 7)
852                        coding_mode = get_bits(gb, 3);
853                    else
854                        coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
855
856                    s->macroblock_coding[current_macroblock] = coding_mode;
857                    for (k = 0; k < 4; k++) {
858                        frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
859                        if (frag->coding_method != MODE_COPY)
860                            frag->coding_method = coding_mode;
861                    }
862
863#define SET_CHROMA_MODES                                                      \
864    if (frag[s->fragment_start[1]].coding_method != MODE_COPY)                \
865        frag[s->fragment_start[1]].coding_method = coding_mode;               \
866    if (frag[s->fragment_start[2]].coding_method != MODE_COPY)                \
867        frag[s->fragment_start[2]].coding_method = coding_mode;
868
869                    if (s->chroma_y_shift) {
870                        frag = s->all_fragments + mb_y *
871                               s->fragment_width[1] + mb_x;
872                        SET_CHROMA_MODES
873                    } else if (s->chroma_x_shift) {
874                        frag = s->all_fragments +
875                               2 * mb_y * s->fragment_width[1] + mb_x;
876                        for (k = 0; k < 2; k++) {
877                            SET_CHROMA_MODES
878                            frag += s->fragment_width[1];
879                        }
880                    } else {
881                        for (k = 0; k < 4; k++) {
882                            frag = s->all_fragments +
883                                   BLOCK_Y * s->fragment_width[1] + BLOCK_X;
884                            SET_CHROMA_MODES
885                        }
886                    }
887                }
888            }
889        }
890    }
891
892    return 0;
893}
894
895static int vp4_get_mv(Vp3DecodeContext *s, GetBitContext *gb, int axis, int last_motion)
896{
897    int v = get_vlc2(gb, s->vp4_mv_vlc[axis][vp4_mv_table_selector[FFABS(last_motion)]].table,
898                     VP4_MV_VLC_BITS, 2);
899    return last_motion < 0 ? -v : v;
900}
901
902/*
903 * This function unpacks all the motion vectors for the individual
904 * macroblocks from the bitstream.
905 */
906static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
907{
908    int j, k, sb_x, sb_y;
909    int coding_mode;
910    int motion_x[4];
911    int motion_y[4];
912    int last_motion_x = 0;
913    int last_motion_y = 0;
914    int prior_last_motion_x = 0;
915    int prior_last_motion_y = 0;
916    int last_gold_motion_x = 0;
917    int last_gold_motion_y = 0;
918    int current_macroblock;
919    int current_fragment;
920    int frag;
921
922    if (s->keyframe)
923        return 0;
924
925    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme; 2 is VP4 code scheme */
926    coding_mode = s->version < 2 ? get_bits1(gb) : 2;
927
928    /* iterate through all of the macroblocks that contain 1 or more
929     * coded fragments */
930    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
931        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
932            if (get_bits_left(gb) <= 0)
933                return -1;
934
935            for (j = 0; j < 4; j++) {
936                int mb_x = 2 * sb_x + (j >> 1);
937                int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
938                current_macroblock = mb_y * s->macroblock_width + mb_x;
939
940                if (mb_x >= s->macroblock_width  ||
941                    mb_y >= s->macroblock_height ||
942                    s->macroblock_coding[current_macroblock] == MODE_COPY)
943                    continue;
944
945                switch (s->macroblock_coding[current_macroblock]) {
946                case MODE_GOLDEN_MV:
947                    if (coding_mode == 2) { /* VP4 */
948                        last_gold_motion_x = motion_x[0] = vp4_get_mv(s, gb, 0, last_gold_motion_x);
949                        last_gold_motion_y = motion_y[0] = vp4_get_mv(s, gb, 1, last_gold_motion_y);
950                        break;
951                    } /* otherwise fall through */
952                case MODE_INTER_PLUS_MV:
953                    /* all 6 fragments use the same motion vector */
954                    if (coding_mode == 0) {
955                        motion_x[0] = get_vlc2(gb, s->motion_vector_vlc.table,
956                                               VP3_MV_VLC_BITS, 2);
957                        motion_y[0] = get_vlc2(gb, s->motion_vector_vlc.table,
958                                               VP3_MV_VLC_BITS, 2);
959                    } else if (coding_mode == 1) {
960                        motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
961                        motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
962                    } else { /* VP4 */
963                        motion_x[0] = vp4_get_mv(s, gb, 0, last_motion_x);
964                        motion_y[0] = vp4_get_mv(s, gb, 1, last_motion_y);
965                    }
966
967                    /* vector maintenance, only on MODE_INTER_PLUS_MV */
968                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
969                        prior_last_motion_x = last_motion_x;
970                        prior_last_motion_y = last_motion_y;
971                        last_motion_x       = motion_x[0];
972                        last_motion_y       = motion_y[0];
973                    }
974                    break;
975
976                case MODE_INTER_FOURMV:
977                    /* vector maintenance */
978                    prior_last_motion_x = last_motion_x;
979                    prior_last_motion_y = last_motion_y;
980
981                    /* fetch 4 vectors from the bitstream, one for each
982                     * Y fragment, then average for the C fragment vectors */
983                    for (k = 0; k < 4; k++) {
984                        current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
985                        if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
986                            if (coding_mode == 0) {
987                                motion_x[k] = get_vlc2(gb, s->motion_vector_vlc.table,
988                                                       VP3_MV_VLC_BITS, 2);
989                                motion_y[k] = get_vlc2(gb, s->motion_vector_vlc.table,
990                                                       VP3_MV_VLC_BITS, 2);
991                            } else if (coding_mode == 1) {
992                                motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
993                                motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
994                            } else { /* VP4 */
995                                motion_x[k] = vp4_get_mv(s, gb, 0, prior_last_motion_x);
996                                motion_y[k] = vp4_get_mv(s, gb, 1, prior_last_motion_y);
997                            }
998                            last_motion_x = motion_x[k];
999                            last_motion_y = motion_y[k];
1000                        } else {
1001                            motion_x[k] = 0;
1002                            motion_y[k] = 0;
1003                        }
1004                    }
1005                    break;
1006
1007                case MODE_INTER_LAST_MV:
1008                    /* all 6 fragments use the last motion vector */
1009                    motion_x[0] = last_motion_x;
1010                    motion_y[0] = last_motion_y;
1011
1012                    /* no vector maintenance (last vector remains the
1013                     * last vector) */
1014                    break;
1015
1016                case MODE_INTER_PRIOR_LAST:
1017                    /* all 6 fragments use the motion vector prior to the
1018                     * last motion vector */
1019                    motion_x[0] = prior_last_motion_x;
1020                    motion_y[0] = prior_last_motion_y;
1021
1022                    /* vector maintenance */
1023                    prior_last_motion_x = last_motion_x;
1024                    prior_last_motion_y = last_motion_y;
1025                    last_motion_x       = motion_x[0];
1026                    last_motion_y       = motion_y[0];
1027                    break;
1028
1029                default:
1030                    /* covers intra, inter without MV, golden without MV */
1031                    motion_x[0] = 0;
1032                    motion_y[0] = 0;
1033
1034                    /* no vector maintenance */
1035                    break;
1036                }
1037
1038                /* assign the motion vectors to the correct fragments */
1039                for (k = 0; k < 4; k++) {
1040                    current_fragment =
1041                        BLOCK_Y * s->fragment_width[0] + BLOCK_X;
1042                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1043                        s->motion_val[0][current_fragment][0] = motion_x[k];
1044                        s->motion_val[0][current_fragment][1] = motion_y[k];
1045                    } else {
1046                        s->motion_val[0][current_fragment][0] = motion_x[0];
1047                        s->motion_val[0][current_fragment][1] = motion_y[0];
1048                    }
1049                }
1050
1051                if (s->chroma_y_shift) {
1052                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1053                        motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
1054                                             motion_x[2] + motion_x[3], 2);
1055                        motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
1056                                             motion_y[2] + motion_y[3], 2);
1057                    }
1058                    if (s->version <= 2) {
1059                        motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1060                        motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
1061                    }
1062                    frag = mb_y * s->fragment_width[1] + mb_x;
1063                    s->motion_val[1][frag][0] = motion_x[0];
1064                    s->motion_val[1][frag][1] = motion_y[0];
1065                } else if (s->chroma_x_shift) {
1066                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1067                        motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
1068                        motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
1069                        motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
1070                        motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
1071                    } else {
1072                        motion_x[1] = motion_x[0];
1073                        motion_y[1] = motion_y[0];
1074                    }
1075                    if (s->version <= 2) {
1076                        motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1077                        motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
1078                    }
1079                    frag = 2 * mb_y * s->fragment_width[1] + mb_x;
1080                    for (k = 0; k < 2; k++) {
1081                        s->motion_val[1][frag][0] = motion_x[k];
1082                        s->motion_val[1][frag][1] = motion_y[k];
1083                        frag += s->fragment_width[1];
1084                    }
1085                } else {
1086                    for (k = 0; k < 4; k++) {
1087                        frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
1088                        if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1089                            s->motion_val[1][frag][0] = motion_x[k];
1090                            s->motion_val[1][frag][1] = motion_y[k];
1091                        } else {
1092                            s->motion_val[1][frag][0] = motion_x[0];
1093                            s->motion_val[1][frag][1] = motion_y[0];
1094                        }
1095                    }
1096                }
1097            }
1098        }
1099    }
1100
1101    return 0;
1102}
1103
1104static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
1105{
1106    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
1107    int num_blocks = s->total_num_coded_frags;
1108
1109    for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
1110        i = blocks_decoded = num_blocks_at_qpi = 0;
1111
1112        bit        = get_bits1(gb) ^ 1;
1113        run_length = 0;
1114
1115        do {
1116            if (run_length == MAXIMUM_LONG_BIT_RUN)
1117                bit = get_bits1(gb);
1118            else
1119                bit ^= 1;
1120
1121            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table,
1122                                  SUPERBLOCK_VLC_BITS, 2);
1123            if (run_length == 34)
1124                run_length += get_bits(gb, 12);
1125            blocks_decoded += run_length;
1126
1127            if (!bit)
1128                num_blocks_at_qpi += run_length;
1129
1130            for (j = 0; j < run_length; i++) {
1131                if (i >= s->total_num_coded_frags)
1132                    return -1;
1133
1134                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
1135                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
1136                    j++;
1137                }
1138            }
1139        } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
1140
1141        num_blocks -= num_blocks_at_qpi;
1142    }
1143
1144    return 0;
1145}
1146
1147static inline int get_eob_run(GetBitContext *gb, int token)
1148{
1149    int v = eob_run_table[token].base;
1150    if (eob_run_table[token].bits)
1151        v += get_bits(gb, eob_run_table[token].bits);
1152    return v;
1153}
1154
1155static inline int get_coeff(GetBitContext *gb, int token, int16_t *coeff)
1156{
1157    int bits_to_get, zero_run;
1158
1159    bits_to_get = coeff_get_bits[token];
1160    if (bits_to_get)
1161        bits_to_get = get_bits(gb, bits_to_get);
1162    *coeff = coeff_tables[token][bits_to_get];
1163
1164    zero_run = zero_run_base[token];
1165    if (zero_run_get_bits[token])
1166        zero_run += get_bits(gb, zero_run_get_bits[token]);
1167
1168    return zero_run;
1169}
1170
1171/*
1172 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1173 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1174 * data. This function unpacks all the VLCs for either the Y plane or both
1175 * C planes, and is called for DC coefficients or different AC coefficient
1176 * levels (since different coefficient types require different VLC tables.
1177 *
1178 * This function returns a residual eob run. E.g, if a particular token gave
1179 * instructions to EOB the next 5 fragments and there were only 2 fragments
1180 * left in the current fragment range, 3 would be returned so that it could
1181 * be passed into the next call to this same function.
1182 */
1183static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1184                       VLC *table, int coeff_index,
1185                       int plane,
1186                       int eob_run)
1187{
1188    int i, j = 0;
1189    int token;
1190    int zero_run  = 0;
1191    int16_t coeff = 0;
1192    int blocks_ended;
1193    int coeff_i = 0;
1194    int num_coeffs      = s->num_coded_frags[plane][coeff_index];
1195    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
1196
1197    /* local references to structure members to avoid repeated dereferences */
1198    int *coded_fragment_list   = s->coded_fragment_list[plane];
1199    Vp3Fragment *all_fragments = s->all_fragments;
1200    const VLCElem *vlc_table = table->table;
1201
1202    if (num_coeffs < 0) {
1203        av_log(s->avctx, AV_LOG_ERROR,
1204               "Invalid number of coefficients at level %d\n", coeff_index);
1205        return AVERROR_INVALIDDATA;
1206    }
1207
1208    if (eob_run > num_coeffs) {
1209        coeff_i      =
1210        blocks_ended = num_coeffs;
1211        eob_run     -= num_coeffs;
1212    } else {
1213        coeff_i      =
1214        blocks_ended = eob_run;
1215        eob_run      = 0;
1216    }
1217
1218    // insert fake EOB token to cover the split between planes or zzi
1219    if (blocks_ended)
1220        dct_tokens[j++] = blocks_ended << 2;
1221
1222    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
1223        /* decode a VLC into a token */
1224        token = get_vlc2(gb, vlc_table, 11, 3);
1225        /* use the token to get a zero run, a coefficient, and an eob run */
1226        if ((unsigned) token <= 6U) {
1227            eob_run = get_eob_run(gb, token);
1228            if (!eob_run)
1229                eob_run = INT_MAX;
1230
1231            // record only the number of blocks ended in this plane,
1232            // any spill will be recorded in the next plane.
1233            if (eob_run > num_coeffs - coeff_i) {
1234                dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
1235                blocks_ended   += num_coeffs - coeff_i;
1236                eob_run        -= num_coeffs - coeff_i;
1237                coeff_i         = num_coeffs;
1238            } else {
1239                dct_tokens[j++] = TOKEN_EOB(eob_run);
1240                blocks_ended   += eob_run;
1241                coeff_i        += eob_run;
1242                eob_run         = 0;
1243            }
1244        } else if (token >= 0) {
1245            zero_run = get_coeff(gb, token, &coeff);
1246
1247            if (zero_run) {
1248                dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
1249            } else {
1250                // Save DC into the fragment structure. DC prediction is
1251                // done in raster order, so the actual DC can't be in with
1252                // other tokens. We still need the token in dct_tokens[]
1253                // however, or else the structure collapses on itself.
1254                if (!coeff_index)
1255                    all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
1256
1257                dct_tokens[j++] = TOKEN_COEFF(coeff);
1258            }
1259
1260            if (coeff_index + zero_run > 64) {
1261                av_log(s->avctx, AV_LOG_DEBUG,
1262                       "Invalid zero run of %d with %d coeffs left\n",
1263                       zero_run, 64 - coeff_index);
1264                zero_run = 64 - coeff_index;
1265            }
1266
1267            // zero runs code multiple coefficients,
1268            // so don't try to decode coeffs for those higher levels
1269            for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1270                s->num_coded_frags[plane][i]--;
1271            coeff_i++;
1272        } else {
1273            av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1274            return -1;
1275        }
1276    }
1277
1278    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1279        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1280
1281    // decrement the number of blocks that have higher coefficients for each
1282    // EOB run at this level
1283    if (blocks_ended)
1284        for (i = coeff_index + 1; i < 64; i++)
1285            s->num_coded_frags[plane][i] -= blocks_ended;
1286
1287    // setup the next buffer
1288    if (plane < 2)
1289        s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1290    else if (coeff_index < 63)
1291        s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1292
1293    return eob_run;
1294}
1295
1296static void reverse_dc_prediction(Vp3DecodeContext *s,
1297                                  int first_fragment,
1298                                  int fragment_width,
1299                                  int fragment_height);
1300/*
1301 * This function unpacks all of the DCT coefficient data from the
1302 * bitstream.
1303 */
1304static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1305{
1306    int i;
1307    int dc_y_table;
1308    int dc_c_table;
1309    int ac_y_table;
1310    int ac_c_table;
1311    int residual_eob_run = 0;
1312    VLC *y_tables[64];
1313    VLC *c_tables[64];
1314
1315    s->dct_tokens[0][0] = s->dct_tokens_base;
1316
1317    if (get_bits_left(gb) < 16)
1318        return AVERROR_INVALIDDATA;
1319
1320    /* fetch the DC table indexes */
1321    dc_y_table = get_bits(gb, 4);
1322    dc_c_table = get_bits(gb, 4);
1323
1324    /* unpack the Y plane DC coefficients */
1325    residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_y_table], 0,
1326                                   0, residual_eob_run);
1327    if (residual_eob_run < 0)
1328        return residual_eob_run;
1329    if (get_bits_left(gb) < 8)
1330        return AVERROR_INVALIDDATA;
1331
1332    /* reverse prediction of the Y-plane DC coefficients */
1333    reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1334
1335    /* unpack the C plane DC coefficients */
1336    residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_c_table], 0,
1337                                   1, residual_eob_run);
1338    if (residual_eob_run < 0)
1339        return residual_eob_run;
1340    residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_c_table], 0,
1341                                   2, residual_eob_run);
1342    if (residual_eob_run < 0)
1343        return residual_eob_run;
1344
1345    /* reverse prediction of the C-plane DC coefficients */
1346    if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1347        reverse_dc_prediction(s, s->fragment_start[1],
1348                              s->fragment_width[1], s->fragment_height[1]);
1349        reverse_dc_prediction(s, s->fragment_start[2],
1350                              s->fragment_width[1], s->fragment_height[1]);
1351    }
1352
1353    if (get_bits_left(gb) < 8)
1354        return AVERROR_INVALIDDATA;
1355    /* fetch the AC table indexes */
1356    ac_y_table = get_bits(gb, 4);
1357    ac_c_table = get_bits(gb, 4);
1358
1359    /* build tables of AC VLC tables */
1360    for (i = 1; i <= 5; i++) {
1361        /* AC VLC table group 1 */
1362        y_tables[i] = &s->coeff_vlc[ac_y_table + 16];
1363        c_tables[i] = &s->coeff_vlc[ac_c_table + 16];
1364    }
1365    for (i = 6; i <= 14; i++) {
1366        /* AC VLC table group 2 */
1367        y_tables[i] = &s->coeff_vlc[ac_y_table + 32];
1368        c_tables[i] = &s->coeff_vlc[ac_c_table + 32];
1369    }
1370    for (i = 15; i <= 27; i++) {
1371        /* AC VLC table group 3 */
1372        y_tables[i] = &s->coeff_vlc[ac_y_table + 48];
1373        c_tables[i] = &s->coeff_vlc[ac_c_table + 48];
1374    }
1375    for (i = 28; i <= 63; i++) {
1376        /* AC VLC table group 4 */
1377        y_tables[i] = &s->coeff_vlc[ac_y_table + 64];
1378        c_tables[i] = &s->coeff_vlc[ac_c_table + 64];
1379    }
1380
1381    /* decode all AC coefficients */
1382    for (i = 1; i <= 63; i++) {
1383        residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1384                                       0, residual_eob_run);
1385        if (residual_eob_run < 0)
1386            return residual_eob_run;
1387
1388        residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1389                                       1, residual_eob_run);
1390        if (residual_eob_run < 0)
1391            return residual_eob_run;
1392        residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1393                                       2, residual_eob_run);
1394        if (residual_eob_run < 0)
1395            return residual_eob_run;
1396    }
1397
1398    return 0;
1399}
1400
1401#if CONFIG_VP4_DECODER
1402/**
1403 * eob_tracker[] is instead of TOKEN_EOB(value)
1404 * a dummy TOKEN_EOB(0) value is used to make vp3_dequant work
1405 *
1406 * @return < 0 on error
1407 */
1408static int vp4_unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1409                       VLC *vlc_tables[64],
1410                       int plane, int eob_tracker[64], int fragment)
1411{
1412    int token;
1413    int zero_run  = 0;
1414    int16_t coeff = 0;
1415    int coeff_i = 0;
1416    int eob_run;
1417
1418    while (!eob_tracker[coeff_i]) {
1419        if (get_bits_left(gb) < 1)
1420            return AVERROR_INVALIDDATA;
1421
1422        token = get_vlc2(gb, vlc_tables[coeff_i]->table, 11, 3);
1423
1424        /* use the token to get a zero run, a coefficient, and an eob run */
1425        if ((unsigned) token <= 6U) {
1426            eob_run = get_eob_run(gb, token);
1427            *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1428            eob_tracker[coeff_i] = eob_run - 1;
1429            return 0;
1430        } else if (token >= 0) {
1431            zero_run = get_coeff(gb, token, &coeff);
1432
1433            if (zero_run) {
1434                if (coeff_i + zero_run > 64) {
1435                    av_log(s->avctx, AV_LOG_DEBUG,
1436                        "Invalid zero run of %d with %d coeffs left\n",
1437                        zero_run, 64 - coeff_i);
1438                    zero_run = 64 - coeff_i;
1439                }
1440                *s->dct_tokens[plane][coeff_i]++ = TOKEN_ZERO_RUN(coeff, zero_run);
1441                coeff_i += zero_run;
1442            } else {
1443                if (!coeff_i)
1444                    s->all_fragments[fragment].dc = coeff;
1445
1446                *s->dct_tokens[plane][coeff_i]++ = TOKEN_COEFF(coeff);
1447            }
1448            coeff_i++;
1449            if (coeff_i >= 64) /* > 64 occurs when there is a zero_run overflow */
1450                return 0; /* stop */
1451        } else {
1452            av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1453            return -1;
1454        }
1455    }
1456    *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1457    eob_tracker[coeff_i]--;
1458    return 0;
1459}
1460
1461static void vp4_dc_predictor_reset(VP4Predictor *p)
1462{
1463    p->dc = 0;
1464    p->type = VP4_DC_UNDEFINED;
1465}
1466
1467static void vp4_dc_pred_before(const Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1468{
1469    int i, j;
1470
1471    for (i = 0; i < 4; i++)
1472        dc_pred[0][i + 1] = s->dc_pred_row[sb_x * 4 + i];
1473
1474    for (j = 1; j < 5; j++)
1475        for (i = 0; i < 4; i++)
1476            vp4_dc_predictor_reset(&dc_pred[j][i + 1]);
1477}
1478
1479static void vp4_dc_pred_after(Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1480{
1481    int i;
1482
1483    for (i = 0; i < 4; i++)
1484        s->dc_pred_row[sb_x * 4 + i] = dc_pred[4][i + 1];
1485
1486    for (i = 1; i < 5; i++)
1487        dc_pred[i][0] = dc_pred[i][4];
1488}
1489
1490/* note: dc_pred points to the current block */
1491static int vp4_dc_pred(const Vp3DecodeContext *s, const VP4Predictor * dc_pred, const int * last_dc, int type, int plane)
1492{
1493    int count = 0;
1494    int dc = 0;
1495
1496    if (dc_pred[-6].type == type) {
1497        dc += dc_pred[-6].dc;
1498        count++;
1499    }
1500
1501    if (dc_pred[6].type == type) {
1502        dc += dc_pred[6].dc;
1503        count++;
1504    }
1505
1506    if (count != 2 && dc_pred[-1].type == type) {
1507        dc += dc_pred[-1].dc;
1508        count++;
1509    }
1510
1511    if (count != 2 && dc_pred[1].type == type) {
1512        dc += dc_pred[1].dc;
1513        count++;
1514    }
1515
1516    /* using division instead of shift to correctly handle negative values */
1517    return count == 2 ? dc / 2 : last_dc[type];
1518}
1519
1520static void vp4_set_tokens_base(Vp3DecodeContext *s)
1521{
1522    int plane, i;
1523    int16_t *base = s->dct_tokens_base;
1524    for (plane = 0; plane < 3; plane++) {
1525        for (i = 0; i < 64; i++) {
1526            s->dct_tokens[plane][i] = base;
1527            base += s->fragment_width[!!plane] * s->fragment_height[!!plane];
1528        }
1529    }
1530}
1531
1532static int vp4_unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1533{
1534    int i, j;
1535    int dc_y_table;
1536    int dc_c_table;
1537    int ac_y_table;
1538    int ac_c_table;
1539    VLC *tables[2][64];
1540    int plane, sb_y, sb_x;
1541    int eob_tracker[64];
1542    VP4Predictor dc_pred[6][6];
1543    int last_dc[NB_VP4_DC_TYPES];
1544
1545    if (get_bits_left(gb) < 16)
1546        return AVERROR_INVALIDDATA;
1547
1548    /* fetch the DC table indexes */
1549    dc_y_table = get_bits(gb, 4);
1550    dc_c_table = get_bits(gb, 4);
1551
1552    ac_y_table = get_bits(gb, 4);
1553    ac_c_table = get_bits(gb, 4);
1554
1555    /* build tables of DC/AC VLC tables */
1556
1557    /* DC table group */
1558    tables[0][0] = &s->coeff_vlc[dc_y_table];
1559    tables[1][0] = &s->coeff_vlc[dc_c_table];
1560    for (i = 1; i <= 5; i++) {
1561        /* AC VLC table group 1 */
1562        tables[0][i] = &s->coeff_vlc[ac_y_table + 16];
1563        tables[1][i] = &s->coeff_vlc[ac_c_table + 16];
1564    }
1565    for (i = 6; i <= 14; i++) {
1566        /* AC VLC table group 2 */
1567        tables[0][i] = &s->coeff_vlc[ac_y_table + 32];
1568        tables[1][i] = &s->coeff_vlc[ac_c_table + 32];
1569    }
1570    for (i = 15; i <= 27; i++) {
1571        /* AC VLC table group 3 */
1572        tables[0][i] = &s->coeff_vlc[ac_y_table + 48];
1573        tables[1][i] = &s->coeff_vlc[ac_c_table + 48];
1574    }
1575    for (i = 28; i <= 63; i++) {
1576        /* AC VLC table group 4 */
1577        tables[0][i] = &s->coeff_vlc[ac_y_table + 64];
1578        tables[1][i] = &s->coeff_vlc[ac_c_table + 64];
1579    }
1580
1581    vp4_set_tokens_base(s);
1582
1583    memset(last_dc, 0, sizeof(last_dc));
1584
1585    for (plane = 0; plane < ((s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 1 : 3); plane++) {
1586        memset(eob_tracker, 0, sizeof(eob_tracker));
1587
1588        /* initialise dc prediction */
1589        for (i = 0; i < s->fragment_width[!!plane]; i++)
1590            vp4_dc_predictor_reset(&s->dc_pred_row[i]);
1591
1592        for (j = 0; j < 6; j++)
1593            for (i = 0; i < 6; i++)
1594                vp4_dc_predictor_reset(&dc_pred[j][i]);
1595
1596        for (sb_y = 0; sb_y * 4 < s->fragment_height[!!plane]; sb_y++) {
1597            for (sb_x = 0; sb_x *4 < s->fragment_width[!!plane]; sb_x++) {
1598                vp4_dc_pred_before(s, dc_pred, sb_x);
1599                for (j = 0; j < 16; j++) {
1600                        int hx = hilbert_offset[j][0];
1601                        int hy = hilbert_offset[j][1];
1602                        int x  = 4 * sb_x + hx;
1603                        int y  = 4 * sb_y + hy;
1604                        VP4Predictor *this_dc_pred = &dc_pred[hy + 1][hx + 1];
1605                        int fragment, dc_block_type;
1606
1607                        if (x >= s->fragment_width[!!plane] || y >= s->fragment_height[!!plane])
1608                            continue;
1609
1610                        fragment = s->fragment_start[plane] + y * s->fragment_width[!!plane] + x;
1611
1612                        if (s->all_fragments[fragment].coding_method == MODE_COPY)
1613                            continue;
1614
1615                        if (vp4_unpack_vlcs(s, gb, tables[!!plane], plane, eob_tracker, fragment) < 0)
1616                            return -1;
1617
1618                        dc_block_type = vp4_pred_block_type_map[s->all_fragments[fragment].coding_method];
1619
1620                        s->all_fragments[fragment].dc +=
1621                            vp4_dc_pred(s, this_dc_pred, last_dc, dc_block_type, plane);
1622
1623                        this_dc_pred->type = dc_block_type,
1624                        this_dc_pred->dc   = last_dc[dc_block_type] = s->all_fragments[fragment].dc;
1625                }
1626                vp4_dc_pred_after(s, dc_pred, sb_x);
1627            }
1628        }
1629    }
1630
1631    vp4_set_tokens_base(s);
1632
1633    return 0;
1634}
1635#endif
1636
1637/*
1638 * This function reverses the DC prediction for each coded fragment in
1639 * the frame. Much of this function is adapted directly from the original
1640 * VP3 source code.
1641 */
1642#define COMPATIBLE_FRAME(x)                                                   \
1643    (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1644#define DC_COEFF(u) s->all_fragments[u].dc
1645
1646static void reverse_dc_prediction(Vp3DecodeContext *s,
1647                                  int first_fragment,
1648                                  int fragment_width,
1649                                  int fragment_height)
1650{
1651#define PUL 8
1652#define PU 4
1653#define PUR 2
1654#define PL 1
1655
1656    int x, y;
1657    int i = first_fragment;
1658
1659    int predicted_dc;
1660
1661    /* DC values for the left, up-left, up, and up-right fragments */
1662    int vl, vul, vu, vur;
1663
1664    /* indexes for the left, up-left, up, and up-right fragments */
1665    int l, ul, u, ur;
1666
1667    /*
1668     * The 6 fields mean:
1669     *   0: up-left multiplier
1670     *   1: up multiplier
1671     *   2: up-right multiplier
1672     *   3: left multiplier
1673     */
1674    static const int predictor_transform[16][4] = {
1675        {    0,   0,   0,   0 },
1676        {    0,   0,   0, 128 }, // PL
1677        {    0,   0, 128,   0 }, // PUR
1678        {    0,   0,  53,  75 }, // PUR|PL
1679        {    0, 128,   0,   0 }, // PU
1680        {    0,  64,   0,  64 }, // PU |PL
1681        {    0, 128,   0,   0 }, // PU |PUR
1682        {    0,   0,  53,  75 }, // PU |PUR|PL
1683        {  128,   0,   0,   0 }, // PUL
1684        {    0,   0,   0, 128 }, // PUL|PL
1685        {   64,   0,  64,   0 }, // PUL|PUR
1686        {    0,   0,  53,  75 }, // PUL|PUR|PL
1687        {    0, 128,   0,   0 }, // PUL|PU
1688        { -104, 116,   0, 116 }, // PUL|PU |PL
1689        {   24,  80,  24,   0 }, // PUL|PU |PUR
1690        { -104, 116,   0, 116 }  // PUL|PU |PUR|PL
1691    };
1692
1693    /* This table shows which types of blocks can use other blocks for
1694     * prediction. For example, INTRA is the only mode in this table to
1695     * have a frame number of 0. That means INTRA blocks can only predict
1696     * from other INTRA blocks. There are 2 golden frame coding types;
1697     * blocks encoding in these modes can only predict from other blocks
1698     * that were encoded with these 1 of these 2 modes. */
1699    static const unsigned char compatible_frame[9] = {
1700        1,    /* MODE_INTER_NO_MV */
1701        0,    /* MODE_INTRA */
1702        1,    /* MODE_INTER_PLUS_MV */
1703        1,    /* MODE_INTER_LAST_MV */
1704        1,    /* MODE_INTER_PRIOR_MV */
1705        2,    /* MODE_USING_GOLDEN */
1706        2,    /* MODE_GOLDEN_MV */
1707        1,    /* MODE_INTER_FOUR_MV */
1708        3     /* MODE_COPY */
1709    };
1710    int current_frame_type;
1711
1712    /* there is a last DC predictor for each of the 3 frame types */
1713    short last_dc[3];
1714
1715    int transform = 0;
1716
1717    vul =
1718    vu  =
1719    vur =
1720    vl  = 0;
1721    last_dc[0] =
1722    last_dc[1] =
1723    last_dc[2] = 0;
1724
1725    /* for each fragment row... */
1726    for (y = 0; y < fragment_height; y++) {
1727        /* for each fragment in a row... */
1728        for (x = 0; x < fragment_width; x++, i++) {
1729
1730            /* reverse prediction if this block was coded */
1731            if (s->all_fragments[i].coding_method != MODE_COPY) {
1732                current_frame_type =
1733                    compatible_frame[s->all_fragments[i].coding_method];
1734
1735                transform = 0;
1736                if (x) {
1737                    l  = i - 1;
1738                    vl = DC_COEFF(l);
1739                    if (COMPATIBLE_FRAME(l))
1740                        transform |= PL;
1741                }
1742                if (y) {
1743                    u  = i - fragment_width;
1744                    vu = DC_COEFF(u);
1745                    if (COMPATIBLE_FRAME(u))
1746                        transform |= PU;
1747                    if (x) {
1748                        ul  = i - fragment_width - 1;
1749                        vul = DC_COEFF(ul);
1750                        if (COMPATIBLE_FRAME(ul))
1751                            transform |= PUL;
1752                    }
1753                    if (x + 1 < fragment_width) {
1754                        ur  = i - fragment_width + 1;
1755                        vur = DC_COEFF(ur);
1756                        if (COMPATIBLE_FRAME(ur))
1757                            transform |= PUR;
1758                    }
1759                }
1760
1761                if (transform == 0) {
1762                    /* if there were no fragments to predict from, use last
1763                     * DC saved */
1764                    predicted_dc = last_dc[current_frame_type];
1765                } else {
1766                    /* apply the appropriate predictor transform */
1767                    predicted_dc =
1768                        (predictor_transform[transform][0] * vul) +
1769                        (predictor_transform[transform][1] * vu) +
1770                        (predictor_transform[transform][2] * vur) +
1771                        (predictor_transform[transform][3] * vl);
1772
1773                    predicted_dc /= 128;
1774
1775                    /* check for outranging on the [ul u l] and
1776                     * [ul u ur l] predictors */
1777                    if ((transform == 15) || (transform == 13)) {
1778                        if (FFABS(predicted_dc - vu) > 128)
1779                            predicted_dc = vu;
1780                        else if (FFABS(predicted_dc - vl) > 128)
1781                            predicted_dc = vl;
1782                        else if (FFABS(predicted_dc - vul) > 128)
1783                            predicted_dc = vul;
1784                    }
1785                }
1786
1787                /* at long last, apply the predictor */
1788                DC_COEFF(i) += predicted_dc;
1789                /* save the DC */
1790                last_dc[current_frame_type] = DC_COEFF(i);
1791            }
1792        }
1793    }
1794}
1795
1796static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1797                              int ystart, int yend)
1798{
1799    int x, y;
1800    int *bounding_values = s->bounding_values_array + 127;
1801
1802    int width           = s->fragment_width[!!plane];
1803    int height          = s->fragment_height[!!plane];
1804    int fragment        = s->fragment_start[plane] + ystart * width;
1805    ptrdiff_t stride    = s->current_frame.f->linesize[plane];
1806    uint8_t *plane_data = s->current_frame.f->data[plane];
1807    if (!s->flipped_image)
1808        stride = -stride;
1809    plane_data += s->data_offset[plane] + 8 * ystart * stride;
1810
1811    for (y = ystart; y < yend; y++) {
1812        for (x = 0; x < width; x++) {
1813            /* This code basically just deblocks on the edges of coded blocks.
1814             * However, it has to be much more complicated because of the
1815             * brain damaged deblock ordering used in VP3/Theora. Order matters
1816             * because some pixels get filtered twice. */
1817            if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1818                /* do not perform left edge filter for left columns frags */
1819                if (x > 0) {
1820                    s->vp3dsp.h_loop_filter(
1821                        plane_data + 8 * x,
1822                        stride, bounding_values);
1823                }
1824
1825                /* do not perform top edge filter for top row fragments */
1826                if (y > 0) {
1827                    s->vp3dsp.v_loop_filter(
1828                        plane_data + 8 * x,
1829                        stride, bounding_values);
1830                }
1831
1832                /* do not perform right edge filter for right column
1833                 * fragments or if right fragment neighbor is also coded
1834                 * in this frame (it will be filtered in next iteration) */
1835                if ((x < width - 1) &&
1836                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1837                    s->vp3dsp.h_loop_filter(
1838                        plane_data + 8 * x + 8,
1839                        stride, bounding_values);
1840                }
1841
1842                /* do not perform bottom edge filter for bottom row
1843                 * fragments or if bottom fragment neighbor is also coded
1844                 * in this frame (it will be filtered in the next row) */
1845                if ((y < height - 1) &&
1846                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1847                    s->vp3dsp.v_loop_filter(
1848                        plane_data + 8 * x + 8 * stride,
1849                        stride, bounding_values);
1850                }
1851            }
1852
1853            fragment++;
1854        }
1855        plane_data += 8 * stride;
1856    }
1857}
1858
1859/**
1860 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1861 * for the next block in coding order
1862 */
1863static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1864                              int plane, int inter, int16_t block[64])
1865{
1866    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1867    uint8_t *perm = s->idct_scantable;
1868    int i = 0;
1869
1870    do {
1871        int token = *s->dct_tokens[plane][i];
1872        switch (token & 3) {
1873        case 0: // EOB
1874            if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1875                s->dct_tokens[plane][i]++;
1876            else
1877                *s->dct_tokens[plane][i] = token & ~3;
1878            goto end;
1879        case 1: // zero run
1880            s->dct_tokens[plane][i]++;
1881            i += (token >> 2) & 0x7f;
1882            if (i > 63) {
1883                av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1884                return i;
1885            }
1886            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1887            i++;
1888            break;
1889        case 2: // coeff
1890            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1891            s->dct_tokens[plane][i++]++;
1892            break;
1893        default: // shouldn't happen
1894            return i;
1895        }
1896    } while (i < 64);
1897    // return value is expected to be a valid level
1898    i--;
1899end:
1900    // the actual DC+prediction is in the fragment structure
1901    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1902    return i;
1903}
1904
1905/**
1906 * called when all pixels up to row y are complete
1907 */
1908static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1909{
1910    int h, cy, i;
1911    int offset[AV_NUM_DATA_POINTERS];
1912
1913    if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1914        int y_flipped = s->flipped_image ? s->height - y : y;
1915
1916        /* At the end of the frame, report INT_MAX instead of the height of
1917         * the frame. This makes the other threads' ff_thread_await_progress()
1918         * calls cheaper, because they don't have to clip their values. */
1919        ff_thread_report_progress(&s->current_frame,
1920                                  y_flipped == s->height ? INT_MAX
1921                                                         : y_flipped - 1,
1922                                  0);
1923    }
1924
1925    if (!s->avctx->draw_horiz_band)
1926        return;
1927
1928    h = y - s->last_slice_end;
1929    s->last_slice_end = y;
1930    y -= h;
1931
1932    if (!s->flipped_image)
1933        y = s->height - y - h;
1934
1935    cy        = y >> s->chroma_y_shift;
1936    offset[0] = s->current_frame.f->linesize[0] * y;
1937    offset[1] = s->current_frame.f->linesize[1] * cy;
1938    offset[2] = s->current_frame.f->linesize[2] * cy;
1939    for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1940        offset[i] = 0;
1941
1942    emms_c();
1943    s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1944}
1945
1946/**
1947 * Wait for the reference frame of the current fragment.
1948 * The progress value is in luma pixel rows.
1949 */
1950static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1951                                int motion_y, int y)
1952{
1953    ThreadFrame *ref_frame;
1954    int ref_row;
1955    int border = motion_y & 1;
1956
1957    if (fragment->coding_method == MODE_USING_GOLDEN ||
1958        fragment->coding_method == MODE_GOLDEN_MV)
1959        ref_frame = &s->golden_frame;
1960    else
1961        ref_frame = &s->last_frame;
1962
1963    ref_row = y + (motion_y >> 1);
1964    ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1965
1966    ff_thread_await_progress(ref_frame, ref_row, 0);
1967}
1968
1969#if CONFIG_VP4_DECODER
1970/**
1971 * @return non-zero if temp (edge_emu_buffer) was populated
1972 */
1973static int vp4_mc_loop_filter(Vp3DecodeContext *s, int plane, int motion_x, int motion_y, int bx, int by,
1974       uint8_t * motion_source, int stride, int src_x, int src_y, uint8_t *temp)
1975{
1976    int motion_shift = plane ? 4 : 2;
1977    int subpel_mask = plane ? 3 : 1;
1978    int *bounding_values = s->bounding_values_array + 127;
1979
1980    int i;
1981    int x, y;
1982    int x2, y2;
1983    int x_subpel, y_subpel;
1984    int x_offset, y_offset;
1985
1986    int block_width = plane ? 8 : 16;
1987    int plane_width  = s->width  >> (plane && s->chroma_x_shift);
1988    int plane_height = s->height >> (plane && s->chroma_y_shift);
1989
1990#define loop_stride 12
1991    uint8_t loop[12 * loop_stride];
1992
1993    /* using division instead of shift to correctly handle negative values */
1994    x = 8 * bx + motion_x / motion_shift;
1995    y = 8 * by + motion_y / motion_shift;
1996
1997    x_subpel = motion_x & subpel_mask;
1998    y_subpel = motion_y & subpel_mask;
1999
2000    if (x_subpel || y_subpel) {
2001        x--;
2002        y--;
2003
2004        if (x_subpel)
2005            x = FFMIN(x, x + FFSIGN(motion_x));
2006
2007        if (y_subpel)
2008            y = FFMIN(y, y + FFSIGN(motion_y));
2009
2010        x2 = x + block_width;
2011        y2 = y + block_width;
2012
2013        if (x2 < 0 || x2 >= plane_width || y2 < 0 || y2 >= plane_height)
2014            return 0;
2015
2016        x_offset = (-(x + 2) & 7) + 2;
2017        y_offset = (-(y + 2) & 7) + 2;
2018
2019        if (x_offset > 8 + x_subpel && y_offset > 8 + y_subpel)
2020            return 0;
2021
2022        s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2023             loop_stride, stride,
2024             12, 12, src_x - 1, src_y - 1,
2025             plane_width,
2026             plane_height);
2027
2028        if (x_offset <= 8 + x_subpel)
2029            ff_vp3dsp_h_loop_filter_12(loop + x_offset, loop_stride, bounding_values);
2030
2031        if (y_offset <= 8 + y_subpel)
2032            ff_vp3dsp_v_loop_filter_12(loop + y_offset*loop_stride, loop_stride, bounding_values);
2033
2034    } else {
2035
2036        x_offset = -x & 7;
2037        y_offset = -y & 7;
2038
2039        if (!x_offset && !y_offset)
2040            return 0;
2041
2042        s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2043             loop_stride, stride,
2044             12, 12, src_x - 1, src_y - 1,
2045             plane_width,
2046             plane_height);
2047
2048#define safe_loop_filter(name, ptr, stride, bounding_values) \
2049    if ((uintptr_t)(ptr) & 7) \
2050        s->vp3dsp.name##_unaligned(ptr, stride, bounding_values); \
2051    else \
2052        s->vp3dsp.name(ptr, stride, bounding_values);
2053
2054        if (x_offset)
2055            safe_loop_filter(h_loop_filter, loop + loop_stride + x_offset + 1, loop_stride, bounding_values);
2056
2057        if (y_offset)
2058            safe_loop_filter(v_loop_filter, loop + (y_offset + 1)*loop_stride + 1, loop_stride, bounding_values);
2059    }
2060
2061    for (i = 0; i < 9; i++)
2062        memcpy(temp + i*stride, loop + (i + 1) * loop_stride + 1, 9);
2063
2064    return 1;
2065}
2066#endif
2067
2068/*
2069 * Perform the final rendering for a particular slice of data.
2070 * The slice number ranges from 0..(c_superblock_height - 1).
2071 */
2072static void render_slice(Vp3DecodeContext *s, int slice)
2073{
2074    int x, y, i, j, fragment;
2075    int16_t *block = s->block;
2076    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
2077    int motion_halfpel_index;
2078    uint8_t *motion_source;
2079    int plane, first_pixel;
2080
2081    if (slice >= s->c_superblock_height)
2082        return;
2083
2084    for (plane = 0; plane < 3; plane++) {
2085        uint8_t *output_plane = s->current_frame.f->data[plane] +
2086                                s->data_offset[plane];
2087        uint8_t *last_plane = s->last_frame.f->data[plane] +
2088                              s->data_offset[plane];
2089        uint8_t *golden_plane = s->golden_frame.f->data[plane] +
2090                                s->data_offset[plane];
2091        ptrdiff_t stride = s->current_frame.f->linesize[plane];
2092        int plane_width  = s->width  >> (plane && s->chroma_x_shift);
2093        int plane_height = s->height >> (plane && s->chroma_y_shift);
2094        int8_t(*motion_val)[2] = s->motion_val[!!plane];
2095
2096        int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
2097        int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
2098        int slice_width  = plane ? s->c_superblock_width
2099                                 : s->y_superblock_width;
2100
2101        int fragment_width  = s->fragment_width[!!plane];
2102        int fragment_height = s->fragment_height[!!plane];
2103        int fragment_start  = s->fragment_start[plane];
2104
2105        int do_await = !plane && HAVE_THREADS &&
2106                       (s->avctx->active_thread_type & FF_THREAD_FRAME);
2107
2108        if (!s->flipped_image)
2109            stride = -stride;
2110        if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY))
2111            continue;
2112
2113        /* for each superblock row in the slice (both of them)... */
2114        for (; sb_y < slice_height; sb_y++) {
2115            /* for each superblock in a row... */
2116            for (sb_x = 0; sb_x < slice_width; sb_x++) {
2117                /* for each block in a superblock... */
2118                for (j = 0; j < 16; j++) {
2119                    x        = 4 * sb_x + hilbert_offset[j][0];
2120                    y        = 4 * sb_y + hilbert_offset[j][1];
2121                    fragment = y * fragment_width + x;
2122
2123                    i = fragment_start + fragment;
2124
2125                    // bounds check
2126                    if (x >= fragment_width || y >= fragment_height)
2127                        continue;
2128
2129                    first_pixel = 8 * y * stride + 8 * x;
2130
2131                    if (do_await &&
2132                        s->all_fragments[i].coding_method != MODE_INTRA)
2133                        await_reference_row(s, &s->all_fragments[i],
2134                                            motion_val[fragment][1],
2135                                            (16 * y) >> s->chroma_y_shift);
2136
2137                    /* transform if this block was coded */
2138                    if (s->all_fragments[i].coding_method != MODE_COPY) {
2139                        if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
2140                            (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
2141                            motion_source = golden_plane;
2142                        else
2143                            motion_source = last_plane;
2144
2145                        motion_source       += first_pixel;
2146                        motion_halfpel_index = 0;
2147
2148                        /* sort out the motion vector if this fragment is coded
2149                         * using a motion vector method */
2150                        if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
2151                            (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
2152                            int src_x, src_y;
2153                            int standard_mc = 1;
2154                            motion_x = motion_val[fragment][0];
2155                            motion_y = motion_val[fragment][1];
2156#if CONFIG_VP4_DECODER
2157                            if (plane && s->version >= 2) {
2158                                motion_x = (motion_x >> 1) | (motion_x & 1);
2159                                motion_y = (motion_y >> 1) | (motion_y & 1);
2160                            }
2161#endif
2162
2163                            src_x = (motion_x >> 1) + 8 * x;
2164                            src_y = (motion_y >> 1) + 8 * y;
2165
2166                            motion_halfpel_index = motion_x & 0x01;
2167                            motion_source       += (motion_x >> 1);
2168
2169                            motion_halfpel_index |= (motion_y & 0x01) << 1;
2170                            motion_source        += ((motion_y >> 1) * stride);
2171
2172#if CONFIG_VP4_DECODER
2173                            if (s->version >= 2) {
2174                                uint8_t *temp = s->edge_emu_buffer;
2175                                if (stride < 0)
2176                                    temp -= 8 * stride;
2177                                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)) {
2178                                    motion_source = temp;
2179                                    standard_mc = 0;
2180                                }
2181                            }
2182#endif
2183
2184                            if (standard_mc && (
2185                                src_x < 0 || src_y < 0 ||
2186                                src_x + 9 >= plane_width ||
2187                                src_y + 9 >= plane_height)) {
2188                                uint8_t *temp = s->edge_emu_buffer;
2189                                if (stride < 0)
2190                                    temp -= 8 * stride;
2191
2192                                s->vdsp.emulated_edge_mc(temp, motion_source,
2193                                                         stride, stride,
2194                                                         9, 9, src_x, src_y,
2195                                                         plane_width,
2196                                                         plane_height);
2197                                motion_source = temp;
2198                            }
2199                        }
2200
2201                        /* first, take care of copying a block from either the
2202                         * previous or the golden frame */
2203                        if (s->all_fragments[i].coding_method != MODE_INTRA) {
2204                            /* Note, it is possible to implement all MC cases
2205                             * with put_no_rnd_pixels_l2 which would look more
2206                             * like the VP3 source but this would be slower as
2207                             * put_no_rnd_pixels_tab is better optimized */
2208                            if (motion_halfpel_index != 3) {
2209                                s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
2210                                    output_plane + first_pixel,
2211                                    motion_source, stride, 8);
2212                            } else {
2213                                /* d is 0 if motion_x and _y have the same sign,
2214                                 * else -1 */
2215                                int d = (motion_x ^ motion_y) >> 31;
2216                                s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
2217                                                               motion_source - d,
2218                                                               motion_source + stride + 1 + d,
2219                                                               stride, 8);
2220                            }
2221                        }
2222
2223                        /* invert DCT and place (or add) in final output */
2224
2225                        if (s->all_fragments[i].coding_method == MODE_INTRA) {
2226                            vp3_dequant(s, s->all_fragments + i,
2227                                        plane, 0, block);
2228                            s->vp3dsp.idct_put(output_plane + first_pixel,
2229                                               stride,
2230                                               block);
2231                        } else {
2232                            if (vp3_dequant(s, s->all_fragments + i,
2233                                            plane, 1, block)) {
2234                                s->vp3dsp.idct_add(output_plane + first_pixel,
2235                                                   stride,
2236                                                   block);
2237                            } else {
2238                                s->vp3dsp.idct_dc_add(output_plane + first_pixel,
2239                                                      stride, block);
2240                            }
2241                        }
2242                    } else {
2243                        /* copy directly from the previous frame */
2244                        s->hdsp.put_pixels_tab[1][0](
2245                            output_plane + first_pixel,
2246                            last_plane + first_pixel,
2247                            stride, 8);
2248                    }
2249                }
2250            }
2251
2252            // Filter up to the last row in the superblock row
2253            if (s->version < 2 && !s->skip_loop_filter)
2254                apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
2255                                  FFMIN(4 * sb_y + 3, fragment_height - 1));
2256        }
2257    }
2258
2259    /* this looks like a good place for slice dispatch... */
2260    /* algorithm:
2261     *   if (slice == s->macroblock_height - 1)
2262     *     dispatch (both last slice & 2nd-to-last slice);
2263     *   else if (slice > 0)
2264     *     dispatch (slice - 1);
2265     */
2266
2267    vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
2268                                 s->height - 16));
2269}
2270
2271/// Allocate tables for per-frame data in Vp3DecodeContext
2272static av_cold int allocate_tables(AVCodecContext *avctx)
2273{
2274    Vp3DecodeContext *s = avctx->priv_data;
2275    int y_fragment_count, c_fragment_count;
2276
2277    free_tables(avctx);
2278
2279    y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
2280    c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
2281
2282    /* superblock_coding is used by unpack_superblocks (VP3/Theora) and vp4_unpack_macroblocks (VP4) */
2283    s->superblock_coding = av_mallocz(FFMAX(s->superblock_count, s->yuv_macroblock_count));
2284    s->all_fragments     = av_calloc(s->fragment_count, sizeof(*s->all_fragments));
2285
2286    s-> kf_coded_fragment_list = av_calloc(s->fragment_count, sizeof(int));
2287    s->nkf_coded_fragment_list = av_calloc(s->fragment_count, sizeof(int));
2288    memset(s-> num_kf_coded_fragment, -1, sizeof(s-> num_kf_coded_fragment));
2289
2290    s->dct_tokens_base = av_calloc(s->fragment_count,
2291                                   64 * sizeof(*s->dct_tokens_base));
2292    s->motion_val[0] = av_calloc(y_fragment_count, sizeof(*s->motion_val[0]));
2293    s->motion_val[1] = av_calloc(c_fragment_count, sizeof(*s->motion_val[1]));
2294
2295    /* work out the block mapping tables */
2296    s->superblock_fragments = av_calloc(s->superblock_count, 16 * sizeof(int));
2297    s->macroblock_coding    = av_mallocz(s->macroblock_count + 1);
2298
2299    s->dc_pred_row = av_malloc_array(s->y_superblock_width * 4, sizeof(*s->dc_pred_row));
2300
2301    if (!s->superblock_coding    || !s->all_fragments          ||
2302        !s->dct_tokens_base      || !s->kf_coded_fragment_list ||
2303        !s->nkf_coded_fragment_list ||
2304        !s->superblock_fragments || !s->macroblock_coding      ||
2305        !s->dc_pred_row ||
2306        !s->motion_val[0]        || !s->motion_val[1]) {
2307        return -1;
2308    }
2309
2310    init_block_mapping(s);
2311
2312    return 0;
2313}
2314
2315static av_cold int init_frames(Vp3DecodeContext *s)
2316{
2317    s->current_frame.f = av_frame_alloc();
2318    s->last_frame.f    = av_frame_alloc();
2319    s->golden_frame.f  = av_frame_alloc();
2320
2321    if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f)
2322        return AVERROR(ENOMEM);
2323
2324    return 0;
2325}
2326
2327static av_cold int vp3_decode_init(AVCodecContext *avctx)
2328{
2329    Vp3DecodeContext *s = avctx->priv_data;
2330    int i, inter, plane, ret;
2331    int c_width;
2332    int c_height;
2333    int y_fragment_count, c_fragment_count;
2334#if CONFIG_VP4_DECODER
2335    int j;
2336#endif
2337
2338    ret = init_frames(s);
2339    if (ret < 0)
2340        return ret;
2341
2342    if (avctx->codec_tag == MKTAG('V', 'P', '4', '0')) {
2343        s->version = 3;
2344#if !CONFIG_VP4_DECODER
2345        av_log(avctx, AV_LOG_ERROR, "This build does not support decoding VP4.\n");
2346        return AVERROR_DECODER_NOT_FOUND;
2347#endif
2348    } else if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
2349        s->version = 0;
2350    else
2351        s->version = 1;
2352
2353    s->avctx  = avctx;
2354    s->width  = FFALIGN(avctx->coded_width, 16);
2355    s->height = FFALIGN(avctx->coded_height, 16);
2356    if (s->width < 18)
2357        return AVERROR_PATCHWELCOME;
2358    if (avctx->codec_id != AV_CODEC_ID_THEORA)
2359        avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2360    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
2361    ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT);
2362    ff_videodsp_init(&s->vdsp, 8);
2363    ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
2364
2365    for (i = 0; i < 64; i++) {
2366#define TRANSPOSE(x) (((x) >> 3) | (((x) & 7) << 3))
2367        s->idct_permutation[i] = TRANSPOSE(i);
2368        s->idct_scantable[i]   = TRANSPOSE(ff_zigzag_direct[i]);
2369#undef TRANSPOSE
2370    }
2371
2372    /* initialize to an impossible value which will force a recalculation
2373     * in the first frame decode */
2374    for (i = 0; i < 3; i++)
2375        s->qps[i] = -1;
2376
2377    ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
2378    if (ret)
2379        return ret;
2380
2381    s->y_superblock_width  = (s->width  + 31) / 32;
2382    s->y_superblock_height = (s->height + 31) / 32;
2383    s->y_superblock_count  = s->y_superblock_width * s->y_superblock_height;
2384
2385    /* work out the dimensions for the C planes */
2386    c_width                = s->width >> s->chroma_x_shift;
2387    c_height               = s->height >> s->chroma_y_shift;
2388    s->c_superblock_width  = (c_width  + 31) / 32;
2389    s->c_superblock_height = (c_height + 31) / 32;
2390    s->c_superblock_count  = s->c_superblock_width * s->c_superblock_height;
2391
2392    s->superblock_count   = s->y_superblock_count + (s->c_superblock_count * 2);
2393    s->u_superblock_start = s->y_superblock_count;
2394    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
2395
2396    s->macroblock_width  = (s->width  + 15) / 16;
2397    s->macroblock_height = (s->height + 15) / 16;
2398    s->macroblock_count  = s->macroblock_width * s->macroblock_height;
2399    s->c_macroblock_width  = (c_width  + 15) / 16;
2400    s->c_macroblock_height = (c_height + 15) / 16;
2401    s->c_macroblock_count  = s->c_macroblock_width * s->c_macroblock_height;
2402    s->yuv_macroblock_count = s->macroblock_count + 2 * s->c_macroblock_count;
2403
2404    s->fragment_width[0]  = s->width / FRAGMENT_PIXELS;
2405    s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
2406    s->fragment_width[1]  = s->fragment_width[0] >> s->chroma_x_shift;
2407    s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
2408
2409    /* fragment count covers all 8x8 blocks for all 3 planes */
2410    y_fragment_count     = s->fragment_width[0] * s->fragment_height[0];
2411    c_fragment_count     = s->fragment_width[1] * s->fragment_height[1];
2412    s->fragment_count    = y_fragment_count + 2 * c_fragment_count;
2413    s->fragment_start[1] = y_fragment_count;
2414    s->fragment_start[2] = y_fragment_count + c_fragment_count;
2415
2416    if (!s->theora_tables) {
2417        const uint8_t (*bias_tabs)[32][2];
2418
2419        for (i = 0; i < 64; i++) {
2420            s->coded_dc_scale_factor[0][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_y_dc_scale_factor[i];
2421            s->coded_dc_scale_factor[1][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_uv_dc_scale_factor[i];
2422            s->coded_ac_scale_factor[i] = s->version < 2 ? vp31_ac_scale_factor[i] : vp4_ac_scale_factor[i];
2423            s->base_matrix[0][i]        = s->version < 2 ? vp31_intra_y_dequant[i] : vp4_generic_dequant[i];
2424            s->base_matrix[1][i]        = s->version < 2 ? vp31_intra_c_dequant[i] : vp4_generic_dequant[i];
2425            s->base_matrix[2][i]        = s->version < 2 ? vp31_inter_dequant[i]   : vp4_generic_dequant[i];
2426            s->filter_limit_values[i]   = s->version < 2 ? vp31_filter_limit_values[i] : vp4_filter_limit_values[i];
2427        }
2428
2429        for (inter = 0; inter < 2; inter++) {
2430            for (plane = 0; plane < 3; plane++) {
2431                s->qr_count[inter][plane]   = 1;
2432                s->qr_size[inter][plane][0] = 63;
2433                s->qr_base[inter][plane][0] =
2434                s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
2435            }
2436        }
2437
2438        /* init VLC tables */
2439        bias_tabs = CONFIG_VP4_DECODER && s->version >= 2 ? vp4_bias : vp3_bias;
2440        for (int i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++) {
2441            ret = ff_init_vlc_from_lengths(&s->coeff_vlc[i], 11, 32,
2442                                           &bias_tabs[i][0][1], 2,
2443                                           &bias_tabs[i][0][0], 2, 1,
2444                                           0, 0, avctx);
2445            if (ret < 0)
2446                return ret;
2447        }
2448    } else {
2449        for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++) {
2450            const HuffTable *tab = &s->huffman_table[i];
2451
2452            ret = ff_init_vlc_from_lengths(&s->coeff_vlc[i], 11, tab->nb_entries,
2453                                           &tab->entries[0].len, sizeof(*tab->entries),
2454                                           &tab->entries[0].sym, sizeof(*tab->entries), 1,
2455                                           0, 0, avctx);
2456            if (ret < 0)
2457                return ret;
2458        }
2459    }
2460
2461    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    if (ret < 0)
2465        return ret;
2466
2467    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    if (ret < 0)
2471        return ret;
2472
2473    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    if (ret < 0)
2477        return ret;
2478
2479    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    if (ret < 0)
2484        return ret;
2485
2486#if CONFIG_VP4_DECODER
2487    for (j = 0; j < 2; j++)
2488        for (i = 0; i < 7; i++) {
2489            ret = ff_init_vlc_from_lengths(&s->vp4_mv_vlc[j][i], VP4_MV_VLC_BITS, 63,
2490                                           &vp4_mv_vlc[j][i][0][1], 2,
2491                                           &vp4_mv_vlc[j][i][0][0], 2, 1, -31,
2492                                           0, avctx);
2493            if (ret < 0)
2494                return ret;
2495        }
2496
2497    /* version >= 2 */
2498    for (i = 0; i < 2; i++)
2499        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    return allocate_tables(avctx);
2506}
2507
2508/// Release and shuffle frames after decode finishes
2509static int update_frames(AVCodecContext *avctx)
2510{
2511    Vp3DecodeContext *s = avctx->priv_data;
2512    int ret = 0;
2513
2514    /* shuffle frames (last = current) */
2515    ff_thread_release_ext_buffer(avctx, &s->last_frame);
2516    ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
2517    if (ret < 0)
2518        goto fail;
2519
2520    if (s->keyframe) {
2521        ff_thread_release_ext_buffer(avctx, &s->golden_frame);
2522        ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
2523    }
2524
2525fail:
2526    ff_thread_release_ext_buffer(avctx, &s->current_frame);
2527    return ret;
2528}
2529
2530#if HAVE_THREADS
2531static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
2532{
2533    ff_thread_release_ext_buffer(s->avctx, dst);
2534    if (src->f->data[0])
2535        return ff_thread_ref_frame(dst, src);
2536    return 0;
2537}
2538
2539static 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
2549static 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
2591static int vp3_decode_frame(AVCodecContext *avctx, AVFrame *frame,
2592                            int *got_frame, AVPacket *avpkt)
2593{
2594    const uint8_t *buf  = avpkt->data;
2595    int buf_size        = avpkt->size;
2596    Vp3DecodeContext *s = avctx->priv_data;
2597    GetBitContext gb;
2598    int i, ret;
2599
2600    if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0)
2601        return ret;
2602
2603#if CONFIG_THEORA_DECODER
2604    if (s->theora && get_bits1(&gb)) {
2605        int type = get_bits(&gb, 7);
2606        skip_bits_long(&gb, 6*8); /* "theora" */
2607
2608        if (s->avctx->active_thread_type&FF_THREAD_FRAME) {
2609            av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n");
2610            return AVERROR_PATCHWELCOME;
2611        }
2612        if (type == 0) {
2613            vp3_decode_end(avctx);
2614            ret = theora_decode_header(avctx, &gb);
2615
2616            if (ret >= 0)
2617                ret = vp3_decode_init(avctx);
2618            if (ret < 0) {
2619                vp3_decode_end(avctx);
2620                return ret;
2621            }
2622            return buf_size;
2623        } else if (type == 2) {
2624            vp3_decode_end(avctx);
2625            ret = theora_decode_tables(avctx, &gb);
2626            if (ret >= 0)
2627                ret = vp3_decode_init(avctx);
2628            if (ret < 0) {
2629                vp3_decode_end(avctx);
2630                return ret;
2631            }
2632            return buf_size;
2633        }
2634
2635        av_log(avctx, AV_LOG_ERROR,
2636               "Header packet passed to frame decoder, skipping\n");
2637        return -1;
2638    }
2639#endif
2640
2641    s->keyframe = !get_bits1(&gb);
2642    if (!s->all_fragments) {
2643        av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n");
2644        return -1;
2645    }
2646    if (!s->theora)
2647        skip_bits(&gb, 1);
2648    for (i = 0; i < 3; i++)
2649        s->last_qps[i] = s->qps[i];
2650
2651    s->nqps = 0;
2652    do {
2653        s->qps[s->nqps++] = get_bits(&gb, 6);
2654    } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2655    for (i = s->nqps; i < 3; i++)
2656        s->qps[i] = -1;
2657
2658    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2659        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2660               s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2661
2662    s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2663                          avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2664                                                                  : AVDISCARD_NONKEY);
2665
2666    if (s->qps[0] != s->last_qps[0])
2667        init_loop_filter(s);
2668
2669    for (i = 0; i < s->nqps; i++)
2670        // reinit all dequantizers if the first one changed, because
2671        // the DC of the first quantizer must be used for all matrices
2672        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2673            init_dequantizer(s, i);
2674
2675    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2676        return buf_size;
2677
2678    s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2679                                                : AV_PICTURE_TYPE_P;
2680    s->current_frame.f->key_frame = s->keyframe;
2681    if ((ret = ff_thread_get_ext_buffer(avctx, &s->current_frame,
2682                                        AV_GET_BUFFER_FLAG_REF)) < 0)
2683        goto error;
2684
2685    if (!s->edge_emu_buffer) {
2686        s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2687        if (!s->edge_emu_buffer) {
2688            ret = AVERROR(ENOMEM);
2689            goto error;
2690        }
2691    }
2692
2693    if (s->keyframe) {
2694        if (!s->theora) {
2695            skip_bits(&gb, 4); /* width code */
2696            skip_bits(&gb, 4); /* height code */
2697            if (s->version) {
2698                int version = get_bits(&gb, 5);
2699#if !CONFIG_VP4_DECODER
2700                if (version >= 2) {
2701                    av_log(avctx, AV_LOG_ERROR, "This build does not support decoding VP4.\n");
2702                    return AVERROR_DECODER_NOT_FOUND;
2703                }
2704#endif
2705                s->version = version;
2706                if (avctx->frame_number == 0)
2707                    av_log(s->avctx, AV_LOG_DEBUG,
2708                           "VP version: %d\n", s->version);
2709            }
2710        }
2711        if (s->version || s->theora) {
2712            if (get_bits1(&gb))
2713                av_log(s->avctx, AV_LOG_ERROR,
2714                       "Warning, unsupported keyframe coding type?!\n");
2715            skip_bits(&gb, 2); /* reserved? */
2716
2717#if CONFIG_VP4_DECODER
2718            if (s->version >= 2) {
2719                int mb_height, mb_width;
2720                int mb_width_mul, mb_width_div, mb_height_mul, mb_height_div;
2721
2722                mb_height = get_bits(&gb, 8);
2723                mb_width  = get_bits(&gb, 8);
2724                if (mb_height != s->macroblock_height ||
2725                    mb_width != s->macroblock_width)
2726                    avpriv_request_sample(s->avctx, "macroblock dimension mismatch");
2727
2728                mb_width_mul = get_bits(&gb, 5);
2729                mb_width_div = get_bits(&gb, 3);
2730                mb_height_mul = get_bits(&gb, 5);
2731                mb_height_div = get_bits(&gb, 3);
2732                if (mb_width_mul != 1 || mb_width_div != 1 || mb_height_mul != 1 || mb_height_div != 1)
2733                    avpriv_request_sample(s->avctx, "unexpected macroblock dimension multipler/divider");
2734
2735                if (get_bits(&gb, 2))
2736                    avpriv_request_sample(s->avctx, "unknown bits");
2737            }
2738#endif
2739        }
2740    } else {
2741        if (!s->golden_frame.f->data[0]) {
2742            av_log(s->avctx, AV_LOG_WARNING,
2743                   "vp3: first frame not a keyframe\n");
2744
2745            s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2746            if ((ret = ff_thread_get_ext_buffer(avctx, &s->golden_frame,
2747                                                AV_GET_BUFFER_FLAG_REF)) < 0)
2748                goto error;
2749            ff_thread_release_ext_buffer(avctx, &s->last_frame);
2750            if ((ret = ff_thread_ref_frame(&s->last_frame,
2751                                           &s->golden_frame)) < 0)
2752                goto error;
2753            ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2754        }
2755    }
2756
2757    memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2758    ff_thread_finish_setup(avctx);
2759
2760    if (s->version < 2) {
2761        if ((ret = unpack_superblocks(s, &gb)) < 0) {
2762            av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2763            goto error;
2764        }
2765#if CONFIG_VP4_DECODER
2766    } else {
2767        if ((ret = vp4_unpack_macroblocks(s, &gb)) < 0) {
2768            av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_macroblocks\n");
2769            goto error;
2770    }
2771#endif
2772    }
2773    if ((ret = unpack_modes(s, &gb)) < 0) {
2774        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2775        goto error;
2776    }
2777    if (ret = unpack_vectors(s, &gb)) {
2778        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2779        goto error;
2780    }
2781    if ((ret = unpack_block_qpis(s, &gb)) < 0) {
2782        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2783        goto error;
2784    }
2785
2786    if (s->version < 2) {
2787        if ((ret = unpack_dct_coeffs(s, &gb)) < 0) {
2788            av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2789            goto error;
2790        }
2791#if CONFIG_VP4_DECODER
2792    } else {
2793        if ((ret = vp4_unpack_dct_coeffs(s, &gb)) < 0) {
2794            av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_dct_coeffs\n");
2795            goto error;
2796        }
2797#endif
2798    }
2799
2800    for (i = 0; i < 3; i++) {
2801        int height = s->height >> (i && s->chroma_y_shift);
2802        if (s->flipped_image)
2803            s->data_offset[i] = 0;
2804        else
2805            s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2806    }
2807
2808    s->last_slice_end = 0;
2809    for (i = 0; i < s->c_superblock_height; i++)
2810        render_slice(s, i);
2811
2812    // filter the last row
2813    if (s->version < 2)
2814        for (i = 0; i < 3; i++) {
2815            int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2816            apply_loop_filter(s, i, row, row + 1);
2817        }
2818    vp3_draw_horiz_band(s, s->height);
2819
2820    /* output frame, offset as needed */
2821    if ((ret = av_frame_ref(frame, s->current_frame.f)) < 0)
2822        return ret;
2823
2824    frame->crop_left   = s->offset_x;
2825    frame->crop_right  = avctx->coded_width - avctx->width - s->offset_x;
2826    frame->crop_top    = s->offset_y;
2827    frame->crop_bottom = avctx->coded_height - avctx->height - s->offset_y;
2828
2829    *got_frame = 1;
2830
2831    if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2832        ret = update_frames(avctx);
2833        if (ret < 0)
2834            return ret;
2835    }
2836
2837    return buf_size;
2838
2839error:
2840    ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2841
2842    if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2843        av_frame_unref(s->current_frame.f);
2844
2845    return ret;
2846}
2847
2848static int read_huffman_tree(HuffTable *huff, GetBitContext *gb, int length,
2849                             AVCodecContext *avctx)
2850{
2851    if (get_bits1(gb)) {
2852        int token;
2853        if (huff->nb_entries >= 32) { /* overflow */
2854            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2855            return -1;
2856        }
2857        token = get_bits(gb, 5);
2858        ff_dlog(avctx, "code length %d, curr entry %d, token %d\n",
2859                length, huff->nb_entries, token);
2860        huff->entries[huff->nb_entries++] = (HuffEntry){ length, token };
2861    } else {
2862        /* The following bound follows from the fact that nb_entries <= 32. */
2863        if (length >= 31) { /* overflow */
2864            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2865            return -1;
2866        }
2867        length++;
2868        if (read_huffman_tree(huff, gb, length, avctx))
2869            return -1;
2870        if (read_huffman_tree(huff, gb, length, avctx))
2871            return -1;
2872    }
2873    return 0;
2874}
2875
2876#if CONFIG_THEORA_DECODER
2877static const enum AVPixelFormat theora_pix_fmts[4] = {
2878    AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2879};
2880
2881static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2882{
2883    Vp3DecodeContext *s = avctx->priv_data;
2884    int visible_width, visible_height, colorspace;
2885    uint8_t offset_x = 0, offset_y = 0;
2886    int ret;
2887    AVRational fps, aspect;
2888
2889    if (get_bits_left(gb) < 206)
2890        return AVERROR_INVALIDDATA;
2891
2892    s->theora_header = 0;
2893    s->theora = get_bits(gb, 24);
2894    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2895    if (!s->theora) {
2896        s->theora = 1;
2897        avpriv_request_sample(s->avctx, "theora 0");
2898    }
2899
2900    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2901     * but previous versions have the image flipped relative to vp3 */
2902    if (s->theora < 0x030200) {
2903        s->flipped_image = 1;
2904        av_log(avctx, AV_LOG_DEBUG,
2905               "Old (<alpha3) Theora bitstream, flipped image\n");
2906    }
2907
2908    visible_width  =
2909    s->width       = get_bits(gb, 16) << 4;
2910    visible_height =
2911    s->height      = get_bits(gb, 16) << 4;
2912
2913    if (s->theora >= 0x030200) {
2914        visible_width  = get_bits(gb, 24);
2915        visible_height = get_bits(gb, 24);
2916
2917        offset_x = get_bits(gb, 8); /* offset x */
2918        offset_y = get_bits(gb, 8); /* offset y, from bottom */
2919    }
2920
2921    /* sanity check */
2922    if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 ||
2923        visible_width  + offset_x > s->width ||
2924        visible_height + offset_y > s->height ||
2925        visible_width < 18
2926    ) {
2927        av_log(avctx, AV_LOG_ERROR,
2928               "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n",
2929               visible_width, visible_height, offset_x, offset_y,
2930               s->width, s->height);
2931        return AVERROR_INVALIDDATA;
2932    }
2933
2934    fps.num = get_bits_long(gb, 32);
2935    fps.den = get_bits_long(gb, 32);
2936    if (fps.num && fps.den) {
2937        if (fps.num < 0 || fps.den < 0) {
2938            av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2939            return AVERROR_INVALIDDATA;
2940        }
2941        av_reduce(&avctx->framerate.den, &avctx->framerate.num,
2942                  fps.den, fps.num, 1 << 30);
2943    }
2944
2945    aspect.num = get_bits(gb, 24);
2946    aspect.den = get_bits(gb, 24);
2947    if (aspect.num && aspect.den) {
2948        av_reduce(&avctx->sample_aspect_ratio.num,
2949                  &avctx->sample_aspect_ratio.den,
2950                  aspect.num, aspect.den, 1 << 30);
2951        ff_set_sar(avctx, avctx->sample_aspect_ratio);
2952    }
2953
2954    if (s->theora < 0x030200)
2955        skip_bits(gb, 5); /* keyframe frequency force */
2956    colorspace = get_bits(gb, 8);
2957    skip_bits(gb, 24); /* bitrate */
2958
2959    skip_bits(gb, 6); /* quality hint */
2960
2961    if (s->theora >= 0x030200) {
2962        skip_bits(gb, 5); /* keyframe frequency force */
2963        avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2964        if (avctx->pix_fmt == AV_PIX_FMT_NONE) {
2965            av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n");
2966            return AVERROR_INVALIDDATA;
2967        }
2968        skip_bits(gb, 3); /* reserved */
2969    } else
2970        avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2971
2972    if (s->width < 18)
2973        return AVERROR_PATCHWELCOME;
2974    ret = ff_set_dimensions(avctx, s->width, s->height);
2975    if (ret < 0)
2976        return ret;
2977    if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP)) {
2978        avctx->width  = visible_width;
2979        avctx->height = visible_height;
2980        // translate offsets from theora axis ([0,0] lower left)
2981        // to normal axis ([0,0] upper left)
2982        s->offset_x = offset_x;
2983        s->offset_y = s->height - visible_height - offset_y;
2984    }
2985
2986    if (colorspace == 1)
2987        avctx->color_primaries = AVCOL_PRI_BT470M;
2988    else if (colorspace == 2)
2989        avctx->color_primaries = AVCOL_PRI_BT470BG;
2990
2991    if (colorspace == 1 || colorspace == 2) {
2992        avctx->colorspace = AVCOL_SPC_BT470BG;
2993        avctx->color_trc  = AVCOL_TRC_BT709;
2994    }
2995
2996    s->theora_header = 1;
2997    return 0;
2998}
2999
3000static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
3001{
3002    Vp3DecodeContext *s = avctx->priv_data;
3003    int i, n, matrices, inter, plane, ret;
3004
3005    if (!s->theora_header)
3006        return AVERROR_INVALIDDATA;
3007
3008    if (s->theora >= 0x030200) {
3009        n = get_bits(gb, 3);
3010        /* loop filter limit values table */
3011        if (n)
3012            for (i = 0; i < 64; i++)
3013                s->filter_limit_values[i] = get_bits(gb, n);
3014    }
3015
3016    if (s->theora >= 0x030200)
3017        n = get_bits(gb, 4) + 1;
3018    else
3019        n = 16;
3020    /* quality threshold table */
3021    for (i = 0; i < 64; i++)
3022        s->coded_ac_scale_factor[i] = get_bits(gb, n);
3023
3024    if (s->theora >= 0x030200)
3025        n = get_bits(gb, 4) + 1;
3026    else
3027        n = 16;
3028    /* dc scale factor table */
3029    for (i = 0; i < 64; i++)
3030        s->coded_dc_scale_factor[0][i] =
3031        s->coded_dc_scale_factor[1][i] = get_bits(gb, n);
3032
3033    if (s->theora >= 0x030200)
3034        matrices = get_bits(gb, 9) + 1;
3035    else
3036        matrices = 3;
3037
3038    if (matrices > 384) {
3039        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
3040        return -1;
3041    }
3042
3043    for (n = 0; n < matrices; n++)
3044        for (i = 0; i < 64; i++)
3045            s->base_matrix[n][i] = get_bits(gb, 8);
3046
3047    for (inter = 0; inter <= 1; inter++) {
3048        for (plane = 0; plane <= 2; plane++) {
3049            int newqr = 1;
3050            if (inter || plane > 0)
3051                newqr = get_bits1(gb);
3052            if (!newqr) {
3053                int qtj, plj;
3054                if (inter && get_bits1(gb)) {
3055                    qtj = 0;
3056                    plj = plane;
3057                } else {
3058                    qtj = (3 * inter + plane - 1) / 3;
3059                    plj = (plane + 2) % 3;
3060                }
3061                s->qr_count[inter][plane] = s->qr_count[qtj][plj];
3062                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
3063                       sizeof(s->qr_size[0][0]));
3064                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
3065                       sizeof(s->qr_base[0][0]));
3066            } else {
3067                int qri = 0;
3068                int qi  = 0;
3069
3070                for (;;) {
3071                    i = get_bits(gb, av_log2(matrices - 1) + 1);
3072                    if (i >= matrices) {
3073                        av_log(avctx, AV_LOG_ERROR,
3074                               "invalid base matrix index\n");
3075                        return -1;
3076                    }
3077                    s->qr_base[inter][plane][qri] = i;
3078                    if (qi >= 63)
3079                        break;
3080                    i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
3081                    s->qr_size[inter][plane][qri++] = i;
3082                    qi += i;
3083                }
3084
3085                if (qi > 63) {
3086                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
3087                    return -1;
3088                }
3089                s->qr_count[inter][plane] = qri;
3090            }
3091        }
3092    }
3093
3094    /* Huffman tables */
3095    for (int i = 0; i < FF_ARRAY_ELEMS(s->huffman_table); i++) {
3096        s->huffman_table[i].nb_entries = 0;
3097        if ((ret = read_huffman_tree(&s->huffman_table[i], gb, 0, avctx)) < 0)
3098            return ret;
3099    }
3100
3101    s->theora_tables = 1;
3102
3103    return 0;
3104}
3105
3106static av_cold int theora_decode_init(AVCodecContext *avctx)
3107{
3108    Vp3DecodeContext *s = avctx->priv_data;
3109    GetBitContext gb;
3110    int ptype;
3111    const uint8_t *header_start[3];
3112    int header_len[3];
3113    int i;
3114    int ret;
3115
3116    avctx->pix_fmt = AV_PIX_FMT_YUV420P;
3117
3118    s->theora = 1;
3119
3120    if (!avctx->extradata_size) {
3121        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
3122        return -1;
3123    }
3124
3125    if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
3126                                  42, header_start, header_len) < 0) {
3127        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
3128        return -1;
3129    }
3130
3131    for (i = 0; i < 3; i++) {
3132        if (header_len[i] <= 0)
3133            continue;
3134        ret = init_get_bits8(&gb, header_start[i], header_len[i]);
3135        if (ret < 0)
3136            return ret;
3137
3138        ptype = get_bits(&gb, 8);
3139
3140        if (!(ptype & 0x80)) {
3141            av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
3142//          return -1;
3143        }
3144
3145        // FIXME: Check for this as well.
3146        skip_bits_long(&gb, 6 * 8); /* "theora" */
3147
3148        switch (ptype) {
3149        case 0x80:
3150            if (theora_decode_header(avctx, &gb) < 0)
3151                return -1;
3152            break;
3153        case 0x81:
3154// FIXME: is this needed? it breaks sometimes
3155//            theora_decode_comments(avctx, gb);
3156            break;
3157        case 0x82:
3158            if (theora_decode_tables(avctx, &gb))
3159                return -1;
3160            break;
3161        default:
3162            av_log(avctx, AV_LOG_ERROR,
3163                   "Unknown Theora config packet: %d\n", ptype & ~0x80);
3164            break;
3165        }
3166        if (ptype != 0x81 && get_bits_left(&gb) >= 8U)
3167            av_log(avctx, AV_LOG_WARNING,
3168                   "%d bits left in packet %X\n",
3169                   get_bits_left(&gb), ptype);
3170        if (s->theora < 0x030200)
3171            break;
3172    }
3173
3174    return vp3_decode_init(avctx);
3175}
3176
3177const FFCodec ff_theora_decoder = {
3178    .p.name                = "theora",
3179    .p.long_name           = NULL_IF_CONFIG_SMALL("Theora"),
3180    .p.type                = AVMEDIA_TYPE_VIDEO,
3181    .p.id                  = AV_CODEC_ID_THEORA,
3182    .priv_data_size        = sizeof(Vp3DecodeContext),
3183    .init                  = theora_decode_init,
3184    .close                 = vp3_decode_end,
3185    FF_CODEC_DECODE_CB(vp3_decode_frame),
3186    .p.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_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP |
3191                             FF_CODEC_CAP_EXPORTS_CROPPING | FF_CODEC_CAP_ALLOCATE_PROGRESS,
3192};
3193#endif
3194
3195const FFCodec ff_vp3_decoder = {
3196    .p.name                = "vp3",
3197    .p.long_name           = NULL_IF_CONFIG_SMALL("On2 VP3"),
3198    .p.type                = AVMEDIA_TYPE_VIDEO,
3199    .p.id                  = AV_CODEC_ID_VP3,
3200    .priv_data_size        = sizeof(Vp3DecodeContext),
3201    .init                  = vp3_decode_init,
3202    .close                 = vp3_decode_end,
3203    FF_CODEC_DECODE_CB(vp3_decode_frame),
3204    .p.capabilities        = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3205                             AV_CODEC_CAP_FRAME_THREADS,
3206    .flush                 = vp3_decode_flush,
3207    .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3208    .caps_internal         = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP |
3209                             FF_CODEC_CAP_ALLOCATE_PROGRESS,
3210};
3211
3212#if CONFIG_VP4_DECODER
3213const FFCodec ff_vp4_decoder = {
3214    .p.name                = "vp4",
3215    .p.long_name           = NULL_IF_CONFIG_SMALL("On2 VP4"),
3216    .p.type                = AVMEDIA_TYPE_VIDEO,
3217    .p.id                  = AV_CODEC_ID_VP4,
3218    .priv_data_size        = sizeof(Vp3DecodeContext),
3219    .init                  = vp3_decode_init,
3220    .close                 = vp3_decode_end,
3221    FF_CODEC_DECODE_CB(vp3_decode_frame),
3222    .p.capabilities        = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3223                             AV_CODEC_CAP_FRAME_THREADS,
3224    .flush                 = vp3_decode_flush,
3225    .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3226    .caps_internal         = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP |
3227                             FF_CODEC_CAP_ALLOCATE_PROGRESS,
3228};
3229#endif
3230