• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2003-2004 The FFmpeg project
3  * Copyright (C) 2019 Peter Ross
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * On2 VP3/VP4 Video Decoder
25  *
26  * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
27  * For more information about the VP3 coding process, visit:
28  *   http://wiki.multimedia.cx/index.php?title=On2_VP3
29  *
30  * Theora decoder by Alex Beregszaszi
31  */
32 
33 #include <stdio.h>
34 #include <stdlib.h>
35 #include <string.h>
36 
37 #include "libavutil/imgutils.h"
38 #include "libavutil/mem_internal.h"
39 
40 #include "avcodec.h"
41 #include "get_bits.h"
42 #include "hpeldsp.h"
43 #include "internal.h"
44 #include "mathops.h"
45 #include "thread.h"
46 #include "videodsp.h"
47 #include "vp3data.h"
48 #include "vp4data.h"
49 #include "vp3dsp.h"
50 #include "xiph.h"
51 
52 #define VP3_MV_VLC_BITS     6
53 #define VP4_MV_VLC_BITS     6
54 #define SUPERBLOCK_VLC_BITS 6
55 
56 #define FRAGMENT_PIXELS 8
57 
58 // FIXME split things out into their own arrays
59 typedef struct Vp3Fragment {
60     int16_t dc;
61     uint8_t coding_method;
62     uint8_t qpi;
63 } Vp3Fragment;
64 
65 #define SB_NOT_CODED        0
66 #define SB_PARTIALLY_CODED  1
67 #define SB_FULLY_CODED      2
68 
69 // This is the maximum length of a single long bit run that can be encoded
70 // for superblock coding or block qps. Theora special-cases this to read a
71 // bit instead of flipping the current bit to allow for runs longer than 4129.
72 #define MAXIMUM_LONG_BIT_RUN 4129
73 
74 #define MODE_INTER_NO_MV      0
75 #define MODE_INTRA            1
76 #define MODE_INTER_PLUS_MV    2
77 #define MODE_INTER_LAST_MV    3
78 #define MODE_INTER_PRIOR_LAST 4
79 #define MODE_USING_GOLDEN     5
80 #define MODE_GOLDEN_MV        6
81 #define MODE_INTER_FOURMV     7
82 #define CODING_MODE_COUNT     8
83 
84 /* special internal mode */
85 #define MODE_COPY             8
86 
87 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb);
88 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
89 
90 
91 /* There are 6 preset schemes, plus a free-form scheme */
92 static const int ModeAlphabet[6][CODING_MODE_COUNT] = {
93     /* scheme 1: Last motion vector dominates */
94     { MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
95       MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
96       MODE_INTRA,            MODE_USING_GOLDEN,
97       MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
98 
99     /* scheme 2 */
100     { MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
101       MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
102       MODE_INTRA,            MODE_USING_GOLDEN,
103       MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
104 
105     /* scheme 3 */
106     { MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
107       MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
108       MODE_INTRA,            MODE_USING_GOLDEN,
109       MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
110 
111     /* scheme 4 */
112     { MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
113       MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
114       MODE_INTRA,            MODE_USING_GOLDEN,
115       MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
116 
117     /* scheme 5: No motion vector dominates */
118     { MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
119       MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
120       MODE_INTRA,            MODE_USING_GOLDEN,
121       MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
122 
123     /* scheme 6 */
124     { MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
125       MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
126       MODE_INTER_PLUS_MV,    MODE_INTRA,
127       MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
128 };
129 
130 static const uint8_t hilbert_offset[16][2] = {
131     { 0, 0 }, { 1, 0 }, { 1, 1 }, { 0, 1 },
132     { 0, 2 }, { 0, 3 }, { 1, 3 }, { 1, 2 },
133     { 2, 2 }, { 2, 3 }, { 3, 3 }, { 3, 2 },
134     { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 }
135 };
136 
137 enum {
138     VP4_DC_INTRA  = 0,
139     VP4_DC_INTER  = 1,
140     VP4_DC_GOLDEN = 2,
141     NB_VP4_DC_TYPES,
142     VP4_DC_UNDEFINED = NB_VP4_DC_TYPES
143 };
144 
145 static const uint8_t vp4_pred_block_type_map[8] = {
146     [MODE_INTER_NO_MV]      = VP4_DC_INTER,
147     [MODE_INTRA]            = VP4_DC_INTRA,
148     [MODE_INTER_PLUS_MV]    = VP4_DC_INTER,
149     [MODE_INTER_LAST_MV]    = VP4_DC_INTER,
150     [MODE_INTER_PRIOR_LAST] = VP4_DC_INTER,
151     [MODE_USING_GOLDEN]     = VP4_DC_GOLDEN,
152     [MODE_GOLDEN_MV]        = VP4_DC_GOLDEN,
153     [MODE_INTER_FOURMV]     = VP4_DC_INTER,
154 };
155 
156 typedef struct {
157     int dc;
158     int type;
159 } VP4Predictor;
160 
161 #define MIN_DEQUANT_VAL 2
162 
163 typedef struct HuffEntry {
164     uint8_t len, sym;
165 } HuffEntry;
166 
167 typedef struct HuffTable {
168     HuffEntry entries[32];
169     uint8_t   nb_entries;
170 } HuffTable;
171 
172 typedef struct Vp3DecodeContext {
173     AVCodecContext *avctx;
174     int theora, theora_tables, theora_header;
175     int version;
176     int width, height;
177     int chroma_x_shift, chroma_y_shift;
178     ThreadFrame golden_frame;
179     ThreadFrame last_frame;
180     ThreadFrame current_frame;
181     int keyframe;
182     uint8_t idct_permutation[64];
183     uint8_t idct_scantable[64];
184     HpelDSPContext hdsp;
185     VideoDSPContext vdsp;
186     VP3DSPContext vp3dsp;
187     DECLARE_ALIGNED(16, int16_t, block)[64];
188     int flipped_image;
189     int last_slice_end;
190     int skip_loop_filter;
191 
192     int qps[3];
193     int nqps;
194     int last_qps[3];
195 
196     int superblock_count;
197     int y_superblock_width;
198     int y_superblock_height;
199     int y_superblock_count;
200     int c_superblock_width;
201     int c_superblock_height;
202     int c_superblock_count;
203     int u_superblock_start;
204     int v_superblock_start;
205     unsigned char *superblock_coding;
206 
207     int macroblock_count; /* y macroblock count */
208     int macroblock_width;
209     int macroblock_height;
210     int c_macroblock_count;
211     int c_macroblock_width;
212     int c_macroblock_height;
213     int yuv_macroblock_count; /* y+u+v macroblock count */
214 
215     int fragment_count;
216     int fragment_width[2];
217     int fragment_height[2];
218 
219     Vp3Fragment *all_fragments;
220     int fragment_start[3];
221     int data_offset[3];
222     uint8_t offset_x;
223     uint8_t offset_y;
224     int offset_x_warned;
225 
226     int8_t (*motion_val[2])[2];
227 
228     /* tables */
229     uint16_t coded_dc_scale_factor[2][64];
230     uint32_t coded_ac_scale_factor[64];
231     uint8_t base_matrix[384][64];
232     uint8_t qr_count[2][3];
233     uint8_t qr_size[2][3][64];
234     uint16_t qr_base[2][3][64];
235 
236     /**
237      * This is a list of all tokens in bitstream order. Reordering takes place
238      * by pulling from each level during IDCT. As a consequence, IDCT must be
239      * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
240      * otherwise. The 32 different tokens with up to 12 bits of extradata are
241      * collapsed into 3 types, packed as follows:
242      *   (from the low to high bits)
243      *
244      * 2 bits: type (0,1,2)
245      *   0: EOB run, 14 bits for run length (12 needed)
246      *   1: zero run, 7 bits for run length
247      *                7 bits for the next coefficient (3 needed)
248      *   2: coefficient, 14 bits (11 needed)
249      *
250      * Coefficients are signed, so are packed in the highest bits for automatic
251      * sign extension.
252      */
253     int16_t *dct_tokens[3][64];
254     int16_t *dct_tokens_base;
255 #define TOKEN_EOB(eob_run)              ((eob_run) << 2)
256 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) * 512) + ((zero_run) << 2) + 1)
257 #define TOKEN_COEFF(coeff)              (((coeff) * 4) + 2)
258 
259     /**
260      * number of blocks that contain DCT coefficients at
261      * the given level or higher
262      */
263     int num_coded_frags[3][64];
264     int total_num_coded_frags;
265 
266     /* this is a list of indexes into the all_fragments array indicating
267      * which of the fragments are coded */
268     int *coded_fragment_list[3];
269 
270     int *kf_coded_fragment_list;
271     int *nkf_coded_fragment_list;
272     int num_kf_coded_fragment[3];
273 
274     /* The first 16 of the following VLCs are for the dc coefficients;
275        the others are four groups of 16 VLCs each for ac coefficients. */
276     VLC coeff_vlc[5 * 16];
277 
278     VLC superblock_run_length_vlc; /* version < 2 */
279     VLC fragment_run_length_vlc; /* version < 2 */
280     VLC block_pattern_vlc[2]; /* version >= 2*/
281     VLC mode_code_vlc;
282     VLC motion_vector_vlc; /* version < 2 */
283     VLC vp4_mv_vlc[2][7]; /* version >=2 */
284 
285     /* these arrays need to be on 16-byte boundaries since SSE2 operations
286      * index into them */
287     DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     ///< qmat[qpi][is_inter][plane]
288 
289     /* This table contains superblock_count * 16 entries. Each set of 16
290      * numbers corresponds to the fragment indexes 0..15 of the superblock.
291      * An entry will be -1 to indicate that no entry corresponds to that
292      * index. */
293     int *superblock_fragments;
294 
295     /* This is an array that indicates how a particular macroblock
296      * is coded. */
297     unsigned char *macroblock_coding;
298 
299     uint8_t *edge_emu_buffer;
300 
301     /* Huffman decode */
302     HuffTable huffman_table[5 * 16];
303 
304     uint8_t filter_limit_values[64];
305     DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
306 
307     VP4Predictor * dc_pred_row; /* dc_pred_row[y_superblock_width * 4] */
308 } Vp3DecodeContext;
309 
310 /************************************************************************
311  * VP3 specific functions
312  ************************************************************************/
313 
free_tables(AVCodecContext * avctx)314 static av_cold void free_tables(AVCodecContext *avctx)
315 {
316     Vp3DecodeContext *s = avctx->priv_data;
317 
318     av_freep(&s->superblock_coding);
319     av_freep(&s->all_fragments);
320     av_freep(&s->nkf_coded_fragment_list);
321     av_freep(&s->kf_coded_fragment_list);
322     av_freep(&s->dct_tokens_base);
323     av_freep(&s->superblock_fragments);
324     av_freep(&s->macroblock_coding);
325     av_freep(&s->dc_pred_row);
326     av_freep(&s->motion_val[0]);
327     av_freep(&s->motion_val[1]);
328 }
329 
vp3_decode_flush(AVCodecContext * avctx)330 static void vp3_decode_flush(AVCodecContext *avctx)
331 {
332     Vp3DecodeContext *s = avctx->priv_data;
333 
334     if (s->golden_frame.f)
335         ff_thread_release_buffer(avctx, &s->golden_frame);
336     if (s->last_frame.f)
337         ff_thread_release_buffer(avctx, &s->last_frame);
338     if (s->current_frame.f)
339         ff_thread_release_buffer(avctx, &s->current_frame);
340 }
341 
vp3_decode_end(AVCodecContext * avctx)342 static av_cold int vp3_decode_end(AVCodecContext *avctx)
343 {
344     Vp3DecodeContext *s = avctx->priv_data;
345     int i, j;
346 
347     free_tables(avctx);
348     av_freep(&s->edge_emu_buffer);
349 
350     s->theora_tables = 0;
351 
352     /* release all frames */
353     vp3_decode_flush(avctx);
354     av_frame_free(&s->current_frame.f);
355     av_frame_free(&s->last_frame.f);
356     av_frame_free(&s->golden_frame.f);
357 
358     for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++)
359         ff_free_vlc(&s->coeff_vlc[i]);
360 
361     ff_free_vlc(&s->superblock_run_length_vlc);
362     ff_free_vlc(&s->fragment_run_length_vlc);
363     ff_free_vlc(&s->mode_code_vlc);
364     ff_free_vlc(&s->motion_vector_vlc);
365 
366     for (j = 0; j < 2; j++)
367         for (i = 0; i < 7; i++)
368             ff_free_vlc(&s->vp4_mv_vlc[j][i]);
369 
370     for (i = 0; i < 2; i++)
371         ff_free_vlc(&s->block_pattern_vlc[i]);
372     return 0;
373 }
374 
375 /**
376  * This function sets up all of the various blocks mappings:
377  * superblocks <-> fragments, macroblocks <-> fragments,
378  * superblocks <-> macroblocks
379  *
380  * @return 0 is successful; returns 1 if *anything* went wrong.
381  */
init_block_mapping(Vp3DecodeContext * s)382 static int init_block_mapping(Vp3DecodeContext *s)
383 {
384     int sb_x, sb_y, plane;
385     int x, y, i, j = 0;
386 
387     for (plane = 0; plane < 3; plane++) {
388         int sb_width    = plane ? s->c_superblock_width
389                                 : s->y_superblock_width;
390         int sb_height   = plane ? s->c_superblock_height
391                                 : s->y_superblock_height;
392         int frag_width  = s->fragment_width[!!plane];
393         int frag_height = s->fragment_height[!!plane];
394 
395         for (sb_y = 0; sb_y < sb_height; sb_y++)
396             for (sb_x = 0; sb_x < sb_width; sb_x++)
397                 for (i = 0; i < 16; i++) {
398                     x = 4 * sb_x + hilbert_offset[i][0];
399                     y = 4 * sb_y + hilbert_offset[i][1];
400 
401                     if (x < frag_width && y < frag_height)
402                         s->superblock_fragments[j++] = s->fragment_start[plane] +
403                                                        y * frag_width + x;
404                     else
405                         s->superblock_fragments[j++] = -1;
406                 }
407     }
408 
409     return 0;  /* successful path out */
410 }
411 
412 /*
413  * This function sets up the dequantization tables used for a particular
414  * frame.
415  */
init_dequantizer(Vp3DecodeContext * s,int qpi)416 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
417 {
418     int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
419     int i, plane, inter, qri, bmi, bmj, qistart;
420 
421     for (inter = 0; inter < 2; inter++) {
422         for (plane = 0; plane < 3; plane++) {
423             int dc_scale_factor = s->coded_dc_scale_factor[!!plane][s->qps[qpi]];
424             int sum = 0;
425             for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
426                 sum += s->qr_size[inter][plane][qri];
427                 if (s->qps[qpi] <= sum)
428                     break;
429             }
430             qistart = sum - s->qr_size[inter][plane][qri];
431             bmi     = s->qr_base[inter][plane][qri];
432             bmj     = s->qr_base[inter][plane][qri + 1];
433             for (i = 0; i < 64; i++) {
434                 int coeff = (2 * (sum     - s->qps[qpi]) * s->base_matrix[bmi][i] -
435                              2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
436                              s->qr_size[inter][plane][qri]) /
437                             (2 * s->qr_size[inter][plane][qri]);
438 
439                 int qmin   = 8 << (inter + !i);
440                 int qscale = i ? ac_scale_factor : dc_scale_factor;
441                 int qbias = (1 + inter) * 3;
442                 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
443                     (i == 0 || s->version < 2) ? av_clip((qscale * coeff) / 100 * 4, qmin, 4096)
444                                                : (qscale * (coeff - qbias) / 100 + qbias) * 4;
445             }
446             /* all DC coefficients use the same quant so as not to interfere
447              * with DC prediction */
448             s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
449         }
450     }
451 }
452 
453 /*
454  * This function initializes the loop filter boundary limits if the frame's
455  * quality index is different from the previous frame's.
456  *
457  * The filter_limit_values may not be larger than 127.
458  */
init_loop_filter(Vp3DecodeContext * s)459 static void init_loop_filter(Vp3DecodeContext *s)
460 {
461     ff_vp3dsp_set_bounding_values(s->bounding_values_array, s->filter_limit_values[s->qps[0]]);
462 }
463 
464 /*
465  * This function unpacks all of the superblock/macroblock/fragment coding
466  * information from the bitstream.
467  */
unpack_superblocks(Vp3DecodeContext * s,GetBitContext * gb)468 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
469 {
470     int superblock_starts[3] = {
471         0, s->u_superblock_start, s->v_superblock_start
472     };
473     int bit = 0;
474     int current_superblock = 0;
475     int current_run = 0;
476     int num_partial_superblocks = 0;
477 
478     int i, j;
479     int current_fragment;
480     int plane;
481     int plane0_num_coded_frags = 0;
482 
483     if (s->keyframe) {
484         memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
485     } else {
486         /* unpack the list of partially-coded superblocks */
487         bit         = get_bits1(gb) ^ 1;
488         current_run = 0;
489 
490         while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
491             if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
492                 bit = get_bits1(gb);
493             else
494                 bit ^= 1;
495 
496             current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
497                                    SUPERBLOCK_VLC_BITS, 2);
498             if (current_run == 34)
499                 current_run += get_bits(gb, 12);
500 
501             if (current_run > s->superblock_count - current_superblock) {
502                 av_log(s->avctx, AV_LOG_ERROR,
503                        "Invalid partially coded superblock run length\n");
504                 return -1;
505             }
506 
507             memset(s->superblock_coding + current_superblock, bit, current_run);
508 
509             current_superblock += current_run;
510             if (bit)
511                 num_partial_superblocks += current_run;
512         }
513 
514         /* unpack the list of fully coded superblocks if any of the blocks were
515          * not marked as partially coded in the previous step */
516         if (num_partial_superblocks < s->superblock_count) {
517             int superblocks_decoded = 0;
518 
519             current_superblock = 0;
520             bit                = get_bits1(gb) ^ 1;
521             current_run        = 0;
522 
523             while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
524                    get_bits_left(gb) > 0) {
525                 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
526                     bit = get_bits1(gb);
527                 else
528                     bit ^= 1;
529 
530                 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
531                                        SUPERBLOCK_VLC_BITS, 2);
532                 if (current_run == 34)
533                     current_run += get_bits(gb, 12);
534 
535                 for (j = 0; j < current_run; current_superblock++) {
536                     if (current_superblock >= s->superblock_count) {
537                         av_log(s->avctx, AV_LOG_ERROR,
538                                "Invalid fully coded superblock run length\n");
539                         return -1;
540                     }
541 
542                     /* skip any superblocks already marked as partially coded */
543                     if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
544                         s->superblock_coding[current_superblock] = 2 * bit;
545                         j++;
546                     }
547                 }
548                 superblocks_decoded += current_run;
549             }
550         }
551 
552         /* if there were partial blocks, initialize bitstream for
553          * unpacking fragment codings */
554         if (num_partial_superblocks) {
555             current_run = 0;
556             bit         = get_bits1(gb);
557             /* toggle the bit because as soon as the first run length is
558              * fetched the bit will be toggled again */
559             bit ^= 1;
560         }
561     }
562 
563     /* figure out which fragments are coded; iterate through each
564      * superblock (all planes) */
565     s->total_num_coded_frags = 0;
566     memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
567 
568     s->coded_fragment_list[0] = s->keyframe ? s->kf_coded_fragment_list
569                                             : s->nkf_coded_fragment_list;
570 
571     for (plane = 0; plane < 3; plane++) {
572         int sb_start = superblock_starts[plane];
573         int sb_end   = sb_start + (plane ? s->c_superblock_count
574                                          : s->y_superblock_count);
575         int num_coded_frags = 0;
576 
577         if (s->keyframe) {
578             if (s->num_kf_coded_fragment[plane] == -1) {
579                 for (i = sb_start; i < sb_end; i++) {
580                     /* iterate through all 16 fragments in a superblock */
581                     for (j = 0; j < 16; j++) {
582                         /* if the fragment is in bounds, check its coding status */
583                         current_fragment = s->superblock_fragments[i * 16 + j];
584                         if (current_fragment != -1) {
585                             s->coded_fragment_list[plane][num_coded_frags++] =
586                                 current_fragment;
587                         }
588                     }
589                 }
590                 s->num_kf_coded_fragment[plane] = num_coded_frags;
591             } else
592                 num_coded_frags = s->num_kf_coded_fragment[plane];
593         } else {
594             for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
595                 if (get_bits_left(gb) < plane0_num_coded_frags >> 2) {
596                     return AVERROR_INVALIDDATA;
597                 }
598                 /* iterate through all 16 fragments in a superblock */
599                 for (j = 0; j < 16; j++) {
600                     /* if the fragment is in bounds, check its coding status */
601                     current_fragment = s->superblock_fragments[i * 16 + j];
602                     if (current_fragment != -1) {
603                         int coded = s->superblock_coding[i];
604 
605                         if (coded == SB_PARTIALLY_CODED) {
606                             /* fragment may or may not be coded; this is the case
607                              * that cares about the fragment coding runs */
608                             if (current_run-- == 0) {
609                                 bit        ^= 1;
610                                 current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2);
611                             }
612                             coded = bit;
613                         }
614 
615                         if (coded) {
616                             /* default mode; actual mode will be decoded in
617                              * the next phase */
618                             s->all_fragments[current_fragment].coding_method =
619                                 MODE_INTER_NO_MV;
620                             s->coded_fragment_list[plane][num_coded_frags++] =
621                                 current_fragment;
622                         } else {
623                             /* not coded; copy this fragment from the prior frame */
624                             s->all_fragments[current_fragment].coding_method =
625                                 MODE_COPY;
626                         }
627                     }
628                 }
629             }
630         }
631         if (!plane)
632             plane0_num_coded_frags = num_coded_frags;
633         s->total_num_coded_frags += num_coded_frags;
634         for (i = 0; i < 64; i++)
635             s->num_coded_frags[plane][i] = num_coded_frags;
636         if (plane < 2)
637             s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] +
638                                                 num_coded_frags;
639     }
640     return 0;
641 }
642 
643 #define BLOCK_X (2 * mb_x + (k & 1))
644 #define BLOCK_Y (2 * mb_y + (k >> 1))
645 
646 #if CONFIG_VP4_DECODER
647 /**
648  * @return number of blocks, or > yuv_macroblock_count on error.
649  *         return value is always >= 1.
650  */
vp4_get_mb_count(Vp3DecodeContext * s,GetBitContext * gb)651 static int vp4_get_mb_count(Vp3DecodeContext *s, GetBitContext *gb)
652 {
653     int v = 1;
654     int bits;
655     while ((bits = show_bits(gb, 9)) == 0x1ff) {
656         skip_bits(gb, 9);
657         v += 256;
658         if (v > s->yuv_macroblock_count) {
659             av_log(s->avctx, AV_LOG_ERROR, "Invalid run length\n");
660             return v;
661         }
662     }
663 #define body(n) { \
664     skip_bits(gb, 2 + n); \
665     v += (1 << n) + get_bits(gb, n); }
666 #define thresh(n) (0x200 - (0x80 >> n))
667 #define else_if(n) else if (bits < thresh(n)) body(n)
668     if (bits < 0x100) {
669         skip_bits(gb, 1);
670     } else if (bits < thresh(0)) {
671         skip_bits(gb, 2);
672         v += 1;
673     }
674     else_if(1)
675     else_if(2)
676     else_if(3)
677     else_if(4)
678     else_if(5)
679     else_if(6)
680     else body(7)
681 #undef body
682 #undef thresh
683 #undef else_if
684     return v;
685 }
686 
vp4_get_block_pattern(Vp3DecodeContext * s,GetBitContext * gb,int * next_block_pattern_table)687 static int vp4_get_block_pattern(Vp3DecodeContext *s, GetBitContext *gb, int *next_block_pattern_table)
688 {
689     int v = get_vlc2(gb, s->block_pattern_vlc[*next_block_pattern_table].table, 3, 2);
690     *next_block_pattern_table = vp4_block_pattern_table_selector[v];
691     return v + 1;
692 }
693 
vp4_unpack_macroblocks(Vp3DecodeContext * s,GetBitContext * gb)694 static int vp4_unpack_macroblocks(Vp3DecodeContext *s, GetBitContext *gb)
695 {
696     int plane, i, j, k, fragment;
697     int next_block_pattern_table;
698     int bit, current_run, has_partial;
699 
700     memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
701 
702     if (s->keyframe)
703         return 0;
704 
705     has_partial = 0;
706     bit         = get_bits1(gb);
707     for (i = 0; i < s->yuv_macroblock_count; i += current_run) {
708         if (get_bits_left(gb) <= 0)
709             return AVERROR_INVALIDDATA;
710         current_run = vp4_get_mb_count(s, gb);
711         if (current_run > s->yuv_macroblock_count - i)
712             return -1;
713         memset(s->superblock_coding + i, 2 * bit, current_run);
714         bit ^= 1;
715         has_partial |= bit;
716     }
717 
718     if (has_partial) {
719         if (get_bits_left(gb) <= 0)
720             return AVERROR_INVALIDDATA;
721         bit  = get_bits1(gb);
722         current_run = vp4_get_mb_count(s, gb);
723         for (i = 0; i < s->yuv_macroblock_count; i++) {
724             if (!s->superblock_coding[i]) {
725                 if (!current_run) {
726                     bit ^= 1;
727                     current_run = vp4_get_mb_count(s, gb);
728                 }
729                 s->superblock_coding[i] = bit;
730                 current_run--;
731             }
732         }
733         if (current_run) /* handle situation when vp4_get_mb_count() fails */
734             return -1;
735     }
736 
737     next_block_pattern_table = 0;
738     i = 0;
739     for (plane = 0; plane < 3; plane++) {
740         int sb_x, sb_y;
741         int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
742         int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
743         int mb_width = plane ? s->c_macroblock_width : s->macroblock_width;
744         int mb_height = plane ? s->c_macroblock_height : s->macroblock_height;
745         int fragment_width = s->fragment_width[!!plane];
746         int fragment_height = s->fragment_height[!!plane];
747 
748         for (sb_y = 0; sb_y < sb_height; sb_y++) {
749             for (sb_x = 0; sb_x < sb_width; sb_x++) {
750                 for (j = 0; j < 4; j++) {
751                     int mb_x = 2 * sb_x + (j >> 1);
752                     int mb_y = 2 * sb_y + (j >> 1) ^ (j & 1);
753                     int mb_coded, pattern, coded;
754 
755                     if (mb_x >= mb_width || mb_y >= mb_height)
756                         continue;
757 
758                     mb_coded = s->superblock_coding[i++];
759 
760                     if (mb_coded == SB_FULLY_CODED)
761                         pattern = 0xF;
762                     else if (mb_coded == SB_PARTIALLY_CODED)
763                         pattern = vp4_get_block_pattern(s, gb, &next_block_pattern_table);
764                     else
765                         pattern = 0;
766 
767                     for (k = 0; k < 4; k++) {
768                         if (BLOCK_X >= fragment_width || BLOCK_Y >= fragment_height)
769                             continue;
770                         fragment = s->fragment_start[plane] + BLOCK_Y * fragment_width + BLOCK_X;
771                         coded = pattern & (8 >> k);
772                         /* MODE_INTER_NO_MV is the default for coded fragments.
773                            the actual method is decoded in the next phase. */
774                         s->all_fragments[fragment].coding_method = coded ? MODE_INTER_NO_MV : MODE_COPY;
775                     }
776                 }
777             }
778         }
779     }
780     return 0;
781 }
782 #endif
783 
784 /*
785  * This function unpacks all the coding mode data for individual macroblocks
786  * from the bitstream.
787  */
unpack_modes(Vp3DecodeContext * s,GetBitContext * gb)788 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
789 {
790     int i, j, k, sb_x, sb_y;
791     int scheme;
792     int current_macroblock;
793     int current_fragment;
794     int coding_mode;
795     int custom_mode_alphabet[CODING_MODE_COUNT];
796     const int *alphabet;
797     Vp3Fragment *frag;
798 
799     if (s->keyframe) {
800         for (i = 0; i < s->fragment_count; i++)
801             s->all_fragments[i].coding_method = MODE_INTRA;
802     } else {
803         /* fetch the mode coding scheme for this frame */
804         scheme = get_bits(gb, 3);
805 
806         /* is it a custom coding scheme? */
807         if (scheme == 0) {
808             for (i = 0; i < 8; i++)
809                 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
810             for (i = 0; i < 8; i++)
811                 custom_mode_alphabet[get_bits(gb, 3)] = i;
812             alphabet = custom_mode_alphabet;
813         } else
814             alphabet = ModeAlphabet[scheme - 1];
815 
816         /* iterate through all of the macroblocks that contain 1 or more
817          * coded fragments */
818         for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
819             for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
820                 if (get_bits_left(gb) <= 0)
821                     return -1;
822 
823                 for (j = 0; j < 4; j++) {
824                     int mb_x = 2 * sb_x + (j >> 1);
825                     int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
826                     current_macroblock = mb_y * s->macroblock_width + mb_x;
827 
828                     if (mb_x >= s->macroblock_width ||
829                         mb_y >= s->macroblock_height)
830                         continue;
831 
832                     /* coding modes are only stored if the macroblock has
833                      * at least one luma block coded, otherwise it must be
834                      * INTER_NO_MV */
835                     for (k = 0; k < 4; k++) {
836                         current_fragment = BLOCK_Y *
837                                            s->fragment_width[0] + BLOCK_X;
838                         if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
839                             break;
840                     }
841                     if (k == 4) {
842                         s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
843                         continue;
844                     }
845 
846                     /* mode 7 means get 3 bits for each coding mode */
847                     if (scheme == 7)
848                         coding_mode = get_bits(gb, 3);
849                     else
850                         coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
851 
852                     s->macroblock_coding[current_macroblock] = coding_mode;
853                     for (k = 0; k < 4; k++) {
854                         frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
855                         if (frag->coding_method != MODE_COPY)
856                             frag->coding_method = coding_mode;
857                     }
858 
859 #define SET_CHROMA_MODES                                                      \
860     if (frag[s->fragment_start[1]].coding_method != MODE_COPY)                \
861         frag[s->fragment_start[1]].coding_method = coding_mode;               \
862     if (frag[s->fragment_start[2]].coding_method != MODE_COPY)                \
863         frag[s->fragment_start[2]].coding_method = coding_mode;
864 
865                     if (s->chroma_y_shift) {
866                         frag = s->all_fragments + mb_y *
867                                s->fragment_width[1] + mb_x;
868                         SET_CHROMA_MODES
869                     } else if (s->chroma_x_shift) {
870                         frag = s->all_fragments +
871                                2 * mb_y * s->fragment_width[1] + mb_x;
872                         for (k = 0; k < 2; k++) {
873                             SET_CHROMA_MODES
874                             frag += s->fragment_width[1];
875                         }
876                     } else {
877                         for (k = 0; k < 4; k++) {
878                             frag = s->all_fragments +
879                                    BLOCK_Y * s->fragment_width[1] + BLOCK_X;
880                             SET_CHROMA_MODES
881                         }
882                     }
883                 }
884             }
885         }
886     }
887 
888     return 0;
889 }
890 
vp4_get_mv(Vp3DecodeContext * s,GetBitContext * gb,int axis,int last_motion)891 static int vp4_get_mv(Vp3DecodeContext *s, GetBitContext *gb, int axis, int last_motion)
892 {
893     int v = get_vlc2(gb, s->vp4_mv_vlc[axis][vp4_mv_table_selector[FFABS(last_motion)]].table,
894                      VP4_MV_VLC_BITS, 2);
895     return last_motion < 0 ? -v : v;
896 }
897 
898 /*
899  * This function unpacks all the motion vectors for the individual
900  * macroblocks from the bitstream.
901  */
unpack_vectors(Vp3DecodeContext * s,GetBitContext * gb)902 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
903 {
904     int j, k, sb_x, sb_y;
905     int coding_mode;
906     int motion_x[4];
907     int motion_y[4];
908     int last_motion_x = 0;
909     int last_motion_y = 0;
910     int prior_last_motion_x = 0;
911     int prior_last_motion_y = 0;
912     int last_gold_motion_x = 0;
913     int last_gold_motion_y = 0;
914     int current_macroblock;
915     int current_fragment;
916     int frag;
917 
918     if (s->keyframe)
919         return 0;
920 
921     /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme; 2 is VP4 code scheme */
922     coding_mode = s->version < 2 ? get_bits1(gb) : 2;
923 
924     /* iterate through all of the macroblocks that contain 1 or more
925      * coded fragments */
926     for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
927         for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
928             if (get_bits_left(gb) <= 0)
929                 return -1;
930 
931             for (j = 0; j < 4; j++) {
932                 int mb_x = 2 * sb_x + (j >> 1);
933                 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
934                 current_macroblock = mb_y * s->macroblock_width + mb_x;
935 
936                 if (mb_x >= s->macroblock_width  ||
937                     mb_y >= s->macroblock_height ||
938                     s->macroblock_coding[current_macroblock] == MODE_COPY)
939                     continue;
940 
941                 switch (s->macroblock_coding[current_macroblock]) {
942                 case MODE_GOLDEN_MV:
943                     if (coding_mode == 2) { /* VP4 */
944                         last_gold_motion_x = motion_x[0] = vp4_get_mv(s, gb, 0, last_gold_motion_x);
945                         last_gold_motion_y = motion_y[0] = vp4_get_mv(s, gb, 1, last_gold_motion_y);
946                         break;
947                     } /* otherwise fall through */
948                 case MODE_INTER_PLUS_MV:
949                     /* all 6 fragments use the same motion vector */
950                     if (coding_mode == 0) {
951                         motion_x[0] = get_vlc2(gb, s->motion_vector_vlc.table,
952                                                VP3_MV_VLC_BITS, 2);
953                         motion_y[0] = get_vlc2(gb, s->motion_vector_vlc.table,
954                                                VP3_MV_VLC_BITS, 2);
955                     } else if (coding_mode == 1) {
956                         motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
957                         motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
958                     } else { /* VP4 */
959                         motion_x[0] = vp4_get_mv(s, gb, 0, last_motion_x);
960                         motion_y[0] = vp4_get_mv(s, gb, 1, last_motion_y);
961                     }
962 
963                     /* vector maintenance, only on MODE_INTER_PLUS_MV */
964                     if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
965                         prior_last_motion_x = last_motion_x;
966                         prior_last_motion_y = last_motion_y;
967                         last_motion_x       = motion_x[0];
968                         last_motion_y       = motion_y[0];
969                     }
970                     break;
971 
972                 case MODE_INTER_FOURMV:
973                     /* vector maintenance */
974                     prior_last_motion_x = last_motion_x;
975                     prior_last_motion_y = last_motion_y;
976 
977                     /* fetch 4 vectors from the bitstream, one for each
978                      * Y fragment, then average for the C fragment vectors */
979                     for (k = 0; k < 4; k++) {
980                         current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
981                         if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
982                             if (coding_mode == 0) {
983                                 motion_x[k] = get_vlc2(gb, s->motion_vector_vlc.table,
984                                                        VP3_MV_VLC_BITS, 2);
985                                 motion_y[k] = get_vlc2(gb, s->motion_vector_vlc.table,
986                                                        VP3_MV_VLC_BITS, 2);
987                             } else if (coding_mode == 1) {
988                                 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
989                                 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
990                             } else { /* VP4 */
991                                 motion_x[k] = vp4_get_mv(s, gb, 0, prior_last_motion_x);
992                                 motion_y[k] = vp4_get_mv(s, gb, 1, prior_last_motion_y);
993                             }
994                             last_motion_x = motion_x[k];
995                             last_motion_y = motion_y[k];
996                         } else {
997                             motion_x[k] = 0;
998                             motion_y[k] = 0;
999                         }
1000                     }
1001                     break;
1002 
1003                 case MODE_INTER_LAST_MV:
1004                     /* all 6 fragments use the last motion vector */
1005                     motion_x[0] = last_motion_x;
1006                     motion_y[0] = last_motion_y;
1007 
1008                     /* no vector maintenance (last vector remains the
1009                      * last vector) */
1010                     break;
1011 
1012                 case MODE_INTER_PRIOR_LAST:
1013                     /* all 6 fragments use the motion vector prior to the
1014                      * last motion vector */
1015                     motion_x[0] = prior_last_motion_x;
1016                     motion_y[0] = prior_last_motion_y;
1017 
1018                     /* vector maintenance */
1019                     prior_last_motion_x = last_motion_x;
1020                     prior_last_motion_y = last_motion_y;
1021                     last_motion_x       = motion_x[0];
1022                     last_motion_y       = motion_y[0];
1023                     break;
1024 
1025                 default:
1026                     /* covers intra, inter without MV, golden without MV */
1027                     motion_x[0] = 0;
1028                     motion_y[0] = 0;
1029 
1030                     /* no vector maintenance */
1031                     break;
1032                 }
1033 
1034                 /* assign the motion vectors to the correct fragments */
1035                 for (k = 0; k < 4; k++) {
1036                     current_fragment =
1037                         BLOCK_Y * s->fragment_width[0] + BLOCK_X;
1038                     if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1039                         s->motion_val[0][current_fragment][0] = motion_x[k];
1040                         s->motion_val[0][current_fragment][1] = motion_y[k];
1041                     } else {
1042                         s->motion_val[0][current_fragment][0] = motion_x[0];
1043                         s->motion_val[0][current_fragment][1] = motion_y[0];
1044                     }
1045                 }
1046 
1047                 if (s->chroma_y_shift) {
1048                     if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1049                         motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
1050                                              motion_x[2] + motion_x[3], 2);
1051                         motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
1052                                              motion_y[2] + motion_y[3], 2);
1053                     }
1054                     if (s->version <= 2) {
1055                         motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1056                         motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
1057                     }
1058                     frag = mb_y * s->fragment_width[1] + mb_x;
1059                     s->motion_val[1][frag][0] = motion_x[0];
1060                     s->motion_val[1][frag][1] = motion_y[0];
1061                 } else if (s->chroma_x_shift) {
1062                     if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1063                         motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
1064                         motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
1065                         motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
1066                         motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
1067                     } else {
1068                         motion_x[1] = motion_x[0];
1069                         motion_y[1] = motion_y[0];
1070                     }
1071                     if (s->version <= 2) {
1072                         motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1073                         motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
1074                     }
1075                     frag = 2 * mb_y * s->fragment_width[1] + mb_x;
1076                     for (k = 0; k < 2; k++) {
1077                         s->motion_val[1][frag][0] = motion_x[k];
1078                         s->motion_val[1][frag][1] = motion_y[k];
1079                         frag += s->fragment_width[1];
1080                     }
1081                 } else {
1082                     for (k = 0; k < 4; k++) {
1083                         frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
1084                         if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1085                             s->motion_val[1][frag][0] = motion_x[k];
1086                             s->motion_val[1][frag][1] = motion_y[k];
1087                         } else {
1088                             s->motion_val[1][frag][0] = motion_x[0];
1089                             s->motion_val[1][frag][1] = motion_y[0];
1090                         }
1091                     }
1092                 }
1093             }
1094         }
1095     }
1096 
1097     return 0;
1098 }
1099 
unpack_block_qpis(Vp3DecodeContext * s,GetBitContext * gb)1100 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
1101 {
1102     int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
1103     int num_blocks = s->total_num_coded_frags;
1104 
1105     for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
1106         i = blocks_decoded = num_blocks_at_qpi = 0;
1107 
1108         bit        = get_bits1(gb) ^ 1;
1109         run_length = 0;
1110 
1111         do {
1112             if (run_length == MAXIMUM_LONG_BIT_RUN)
1113                 bit = get_bits1(gb);
1114             else
1115                 bit ^= 1;
1116 
1117             run_length = get_vlc2(gb, s->superblock_run_length_vlc.table,
1118                                   SUPERBLOCK_VLC_BITS, 2);
1119             if (run_length == 34)
1120                 run_length += get_bits(gb, 12);
1121             blocks_decoded += run_length;
1122 
1123             if (!bit)
1124                 num_blocks_at_qpi += run_length;
1125 
1126             for (j = 0; j < run_length; i++) {
1127                 if (i >= s->total_num_coded_frags)
1128                     return -1;
1129 
1130                 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
1131                     s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
1132                     j++;
1133                 }
1134             }
1135         } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
1136 
1137         num_blocks -= num_blocks_at_qpi;
1138     }
1139 
1140     return 0;
1141 }
1142 
get_eob_run(GetBitContext * gb,int token)1143 static inline int get_eob_run(GetBitContext *gb, int token)
1144 {
1145     int v = eob_run_table[token].base;
1146     if (eob_run_table[token].bits)
1147         v += get_bits(gb, eob_run_table[token].bits);
1148     return v;
1149 }
1150 
get_coeff(GetBitContext * gb,int token,int16_t * coeff)1151 static inline int get_coeff(GetBitContext *gb, int token, int16_t *coeff)
1152 {
1153     int bits_to_get, zero_run;
1154 
1155     bits_to_get = coeff_get_bits[token];
1156     if (bits_to_get)
1157         bits_to_get = get_bits(gb, bits_to_get);
1158     *coeff = coeff_tables[token][bits_to_get];
1159 
1160     zero_run = zero_run_base[token];
1161     if (zero_run_get_bits[token])
1162         zero_run += get_bits(gb, zero_run_get_bits[token]);
1163 
1164     return zero_run;
1165 }
1166 
1167 /*
1168  * This function is called by unpack_dct_coeffs() to extract the VLCs from
1169  * the bitstream. The VLCs encode tokens which are used to unpack DCT
1170  * data. This function unpacks all the VLCs for either the Y plane or both
1171  * C planes, and is called for DC coefficients or different AC coefficient
1172  * levels (since different coefficient types require different VLC tables.
1173  *
1174  * This function returns a residual eob run. E.g, if a particular token gave
1175  * instructions to EOB the next 5 fragments and there were only 2 fragments
1176  * left in the current fragment range, 3 would be returned so that it could
1177  * be passed into the next call to this same function.
1178  */
unpack_vlcs(Vp3DecodeContext * s,GetBitContext * gb,VLC * table,int coeff_index,int plane,int eob_run)1179 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1180                        VLC *table, int coeff_index,
1181                        int plane,
1182                        int eob_run)
1183 {
1184     int i, j = 0;
1185     int token;
1186     int zero_run  = 0;
1187     int16_t coeff = 0;
1188     int blocks_ended;
1189     int coeff_i = 0;
1190     int num_coeffs      = s->num_coded_frags[plane][coeff_index];
1191     int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
1192 
1193     /* local references to structure members to avoid repeated dereferences */
1194     int *coded_fragment_list   = s->coded_fragment_list[plane];
1195     Vp3Fragment *all_fragments = s->all_fragments;
1196     VLC_TYPE(*vlc_table)[2] = table->table;
1197 
1198     if (num_coeffs < 0) {
1199         av_log(s->avctx, AV_LOG_ERROR,
1200                "Invalid number of coefficients at level %d\n", coeff_index);
1201         return AVERROR_INVALIDDATA;
1202     }
1203 
1204     if (eob_run > num_coeffs) {
1205         coeff_i      =
1206         blocks_ended = num_coeffs;
1207         eob_run     -= num_coeffs;
1208     } else {
1209         coeff_i      =
1210         blocks_ended = eob_run;
1211         eob_run      = 0;
1212     }
1213 
1214     // insert fake EOB token to cover the split between planes or zzi
1215     if (blocks_ended)
1216         dct_tokens[j++] = blocks_ended << 2;
1217 
1218     while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
1219         /* decode a VLC into a token */
1220         token = get_vlc2(gb, vlc_table, 11, 3);
1221         /* use the token to get a zero run, a coefficient, and an eob run */
1222         if ((unsigned) token <= 6U) {
1223             eob_run = get_eob_run(gb, token);
1224             if (!eob_run)
1225                 eob_run = INT_MAX;
1226 
1227             // record only the number of blocks ended in this plane,
1228             // any spill will be recorded in the next plane.
1229             if (eob_run > num_coeffs - coeff_i) {
1230                 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
1231                 blocks_ended   += num_coeffs - coeff_i;
1232                 eob_run        -= num_coeffs - coeff_i;
1233                 coeff_i         = num_coeffs;
1234             } else {
1235                 dct_tokens[j++] = TOKEN_EOB(eob_run);
1236                 blocks_ended   += eob_run;
1237                 coeff_i        += eob_run;
1238                 eob_run         = 0;
1239             }
1240         } else if (token >= 0) {
1241             zero_run = get_coeff(gb, token, &coeff);
1242 
1243             if (zero_run) {
1244                 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
1245             } else {
1246                 // Save DC into the fragment structure. DC prediction is
1247                 // done in raster order, so the actual DC can't be in with
1248                 // other tokens. We still need the token in dct_tokens[]
1249                 // however, or else the structure collapses on itself.
1250                 if (!coeff_index)
1251                     all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
1252 
1253                 dct_tokens[j++] = TOKEN_COEFF(coeff);
1254             }
1255 
1256             if (coeff_index + zero_run > 64) {
1257                 av_log(s->avctx, AV_LOG_DEBUG,
1258                        "Invalid zero run of %d with %d coeffs left\n",
1259                        zero_run, 64 - coeff_index);
1260                 zero_run = 64 - coeff_index;
1261             }
1262 
1263             // zero runs code multiple coefficients,
1264             // so don't try to decode coeffs for those higher levels
1265             for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1266                 s->num_coded_frags[plane][i]--;
1267             coeff_i++;
1268         } else {
1269             av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1270             return -1;
1271         }
1272     }
1273 
1274     if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1275         av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1276 
1277     // decrement the number of blocks that have higher coefficients for each
1278     // EOB run at this level
1279     if (blocks_ended)
1280         for (i = coeff_index + 1; i < 64; i++)
1281             s->num_coded_frags[plane][i] -= blocks_ended;
1282 
1283     // setup the next buffer
1284     if (plane < 2)
1285         s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1286     else if (coeff_index < 63)
1287         s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1288 
1289     return eob_run;
1290 }
1291 
1292 static void reverse_dc_prediction(Vp3DecodeContext *s,
1293                                   int first_fragment,
1294                                   int fragment_width,
1295                                   int fragment_height);
1296 /*
1297  * This function unpacks all of the DCT coefficient data from the
1298  * bitstream.
1299  */
unpack_dct_coeffs(Vp3DecodeContext * s,GetBitContext * gb)1300 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1301 {
1302     int i;
1303     int dc_y_table;
1304     int dc_c_table;
1305     int ac_y_table;
1306     int ac_c_table;
1307     int residual_eob_run = 0;
1308     VLC *y_tables[64];
1309     VLC *c_tables[64];
1310 
1311     s->dct_tokens[0][0] = s->dct_tokens_base;
1312 
1313     if (get_bits_left(gb) < 16)
1314         return AVERROR_INVALIDDATA;
1315 
1316     /* fetch the DC table indexes */
1317     dc_y_table = get_bits(gb, 4);
1318     dc_c_table = get_bits(gb, 4);
1319 
1320     /* unpack the Y plane DC coefficients */
1321     residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_y_table], 0,
1322                                    0, residual_eob_run);
1323     if (residual_eob_run < 0)
1324         return residual_eob_run;
1325     if (get_bits_left(gb) < 8)
1326         return AVERROR_INVALIDDATA;
1327 
1328     /* reverse prediction of the Y-plane DC coefficients */
1329     reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1330 
1331     /* unpack the C plane DC coefficients */
1332     residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_c_table], 0,
1333                                    1, residual_eob_run);
1334     if (residual_eob_run < 0)
1335         return residual_eob_run;
1336     residual_eob_run = unpack_vlcs(s, gb, &s->coeff_vlc[dc_c_table], 0,
1337                                    2, residual_eob_run);
1338     if (residual_eob_run < 0)
1339         return residual_eob_run;
1340 
1341     /* reverse prediction of the C-plane DC coefficients */
1342     if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1343         reverse_dc_prediction(s, s->fragment_start[1],
1344                               s->fragment_width[1], s->fragment_height[1]);
1345         reverse_dc_prediction(s, s->fragment_start[2],
1346                               s->fragment_width[1], s->fragment_height[1]);
1347     }
1348 
1349     if (get_bits_left(gb) < 8)
1350         return AVERROR_INVALIDDATA;
1351     /* fetch the AC table indexes */
1352     ac_y_table = get_bits(gb, 4);
1353     ac_c_table = get_bits(gb, 4);
1354 
1355     /* build tables of AC VLC tables */
1356     for (i = 1; i <= 5; i++) {
1357         /* AC VLC table group 1 */
1358         y_tables[i] = &s->coeff_vlc[ac_y_table + 16];
1359         c_tables[i] = &s->coeff_vlc[ac_c_table + 16];
1360     }
1361     for (i = 6; i <= 14; i++) {
1362         /* AC VLC table group 2 */
1363         y_tables[i] = &s->coeff_vlc[ac_y_table + 32];
1364         c_tables[i] = &s->coeff_vlc[ac_c_table + 32];
1365     }
1366     for (i = 15; i <= 27; i++) {
1367         /* AC VLC table group 3 */
1368         y_tables[i] = &s->coeff_vlc[ac_y_table + 48];
1369         c_tables[i] = &s->coeff_vlc[ac_c_table + 48];
1370     }
1371     for (i = 28; i <= 63; i++) {
1372         /* AC VLC table group 4 */
1373         y_tables[i] = &s->coeff_vlc[ac_y_table + 64];
1374         c_tables[i] = &s->coeff_vlc[ac_c_table + 64];
1375     }
1376 
1377     /* decode all AC coefficients */
1378     for (i = 1; i <= 63; i++) {
1379         residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1380                                        0, residual_eob_run);
1381         if (residual_eob_run < 0)
1382             return residual_eob_run;
1383 
1384         residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1385                                        1, residual_eob_run);
1386         if (residual_eob_run < 0)
1387             return residual_eob_run;
1388         residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1389                                        2, residual_eob_run);
1390         if (residual_eob_run < 0)
1391             return residual_eob_run;
1392     }
1393 
1394     return 0;
1395 }
1396 
1397 #if CONFIG_VP4_DECODER
1398 /**
1399  * eob_tracker[] is instead of TOKEN_EOB(value)
1400  * a dummy TOKEN_EOB(0) value is used to make vp3_dequant work
1401  *
1402  * @return < 0 on error
1403  */
vp4_unpack_vlcs(Vp3DecodeContext * s,GetBitContext * gb,VLC * vlc_tables[64],int plane,int eob_tracker[64],int fragment)1404 static int vp4_unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1405                        VLC *vlc_tables[64],
1406                        int plane, int eob_tracker[64], int fragment)
1407 {
1408     int token;
1409     int zero_run  = 0;
1410     int16_t coeff = 0;
1411     int coeff_i = 0;
1412     int eob_run;
1413 
1414     while (!eob_tracker[coeff_i]) {
1415         if (get_bits_left(gb) < 1)
1416             return AVERROR_INVALIDDATA;
1417 
1418         token = get_vlc2(gb, vlc_tables[coeff_i]->table, 11, 3);
1419 
1420         /* use the token to get a zero run, a coefficient, and an eob run */
1421         if ((unsigned) token <= 6U) {
1422             eob_run = get_eob_run(gb, token);
1423             *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1424             eob_tracker[coeff_i] = eob_run - 1;
1425             return 0;
1426         } else if (token >= 0) {
1427             zero_run = get_coeff(gb, token, &coeff);
1428 
1429             if (zero_run) {
1430                 if (coeff_i + zero_run > 64) {
1431                     av_log(s->avctx, AV_LOG_DEBUG,
1432                         "Invalid zero run of %d with %d coeffs left\n",
1433                         zero_run, 64 - coeff_i);
1434                     zero_run = 64 - coeff_i;
1435                 }
1436                 *s->dct_tokens[plane][coeff_i]++ = TOKEN_ZERO_RUN(coeff, zero_run);
1437                 coeff_i += zero_run;
1438             } else {
1439                 if (!coeff_i)
1440                     s->all_fragments[fragment].dc = coeff;
1441 
1442                 *s->dct_tokens[plane][coeff_i]++ = TOKEN_COEFF(coeff);
1443             }
1444             coeff_i++;
1445             if (coeff_i >= 64) /* > 64 occurs when there is a zero_run overflow */
1446                 return 0; /* stop */
1447         } else {
1448             av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1449             return -1;
1450         }
1451     }
1452     *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1453     eob_tracker[coeff_i]--;
1454     return 0;
1455 }
1456 
vp4_dc_predictor_reset(VP4Predictor * p)1457 static void vp4_dc_predictor_reset(VP4Predictor *p)
1458 {
1459     p->dc = 0;
1460     p->type = VP4_DC_UNDEFINED;
1461 }
1462 
vp4_dc_pred_before(const Vp3DecodeContext * s,VP4Predictor dc_pred[6][6],int sb_x)1463 static void vp4_dc_pred_before(const Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1464 {
1465     int i, j;
1466 
1467     for (i = 0; i < 4; i++)
1468         dc_pred[0][i + 1] = s->dc_pred_row[sb_x * 4 + i];
1469 
1470     for (j = 1; j < 5; j++)
1471         for (i = 0; i < 4; i++)
1472             vp4_dc_predictor_reset(&dc_pred[j][i + 1]);
1473 }
1474 
vp4_dc_pred_after(Vp3DecodeContext * s,VP4Predictor dc_pred[6][6],int sb_x)1475 static void vp4_dc_pred_after(Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1476 {
1477     int i;
1478 
1479     for (i = 0; i < 4; i++)
1480         s->dc_pred_row[sb_x * 4 + i] = dc_pred[4][i + 1];
1481 
1482     for (i = 1; i < 5; i++)
1483         dc_pred[i][0] = dc_pred[i][4];
1484 }
1485 
1486 /* note: dc_pred points to the current block */
vp4_dc_pred(const Vp3DecodeContext * s,const VP4Predictor * dc_pred,const int * last_dc,int type,int plane)1487 static int vp4_dc_pred(const Vp3DecodeContext *s, const VP4Predictor * dc_pred, const int * last_dc, int type, int plane)
1488 {
1489     int count = 0;
1490     int dc = 0;
1491 
1492     if (dc_pred[-6].type == type) {
1493         dc += dc_pred[-6].dc;
1494         count++;
1495     }
1496 
1497     if (dc_pred[6].type == type) {
1498         dc += dc_pred[6].dc;
1499         count++;
1500     }
1501 
1502     if (count != 2 && dc_pred[-1].type == type) {
1503         dc += dc_pred[-1].dc;
1504         count++;
1505     }
1506 
1507     if (count != 2 && dc_pred[1].type == type) {
1508         dc += dc_pred[1].dc;
1509         count++;
1510     }
1511 
1512     /* using division instead of shift to correctly handle negative values */
1513     return count == 2 ? dc / 2 : last_dc[type];
1514 }
1515 
vp4_set_tokens_base(Vp3DecodeContext * s)1516 static void vp4_set_tokens_base(Vp3DecodeContext *s)
1517 {
1518     int plane, i;
1519     int16_t *base = s->dct_tokens_base;
1520     for (plane = 0; plane < 3; plane++) {
1521         for (i = 0; i < 64; i++) {
1522             s->dct_tokens[plane][i] = base;
1523             base += s->fragment_width[!!plane] * s->fragment_height[!!plane];
1524         }
1525     }
1526 }
1527 
vp4_unpack_dct_coeffs(Vp3DecodeContext * s,GetBitContext * gb)1528 static int vp4_unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1529 {
1530     int i, j;
1531     int dc_y_table;
1532     int dc_c_table;
1533     int ac_y_table;
1534     int ac_c_table;
1535     VLC *tables[2][64];
1536     int plane, sb_y, sb_x;
1537     int eob_tracker[64];
1538     VP4Predictor dc_pred[6][6];
1539     int last_dc[NB_VP4_DC_TYPES];
1540 
1541     if (get_bits_left(gb) < 16)
1542         return AVERROR_INVALIDDATA;
1543 
1544     /* fetch the DC table indexes */
1545     dc_y_table = get_bits(gb, 4);
1546     dc_c_table = get_bits(gb, 4);
1547 
1548     ac_y_table = get_bits(gb, 4);
1549     ac_c_table = get_bits(gb, 4);
1550 
1551     /* build tables of DC/AC VLC tables */
1552 
1553     /* DC table group */
1554     tables[0][0] = &s->coeff_vlc[dc_y_table];
1555     tables[1][0] = &s->coeff_vlc[dc_c_table];
1556     for (i = 1; i <= 5; i++) {
1557         /* AC VLC table group 1 */
1558         tables[0][i] = &s->coeff_vlc[ac_y_table + 16];
1559         tables[1][i] = &s->coeff_vlc[ac_c_table + 16];
1560     }
1561     for (i = 6; i <= 14; i++) {
1562         /* AC VLC table group 2 */
1563         tables[0][i] = &s->coeff_vlc[ac_y_table + 32];
1564         tables[1][i] = &s->coeff_vlc[ac_c_table + 32];
1565     }
1566     for (i = 15; i <= 27; i++) {
1567         /* AC VLC table group 3 */
1568         tables[0][i] = &s->coeff_vlc[ac_y_table + 48];
1569         tables[1][i] = &s->coeff_vlc[ac_c_table + 48];
1570     }
1571     for (i = 28; i <= 63; i++) {
1572         /* AC VLC table group 4 */
1573         tables[0][i] = &s->coeff_vlc[ac_y_table + 64];
1574         tables[1][i] = &s->coeff_vlc[ac_c_table + 64];
1575     }
1576 
1577     vp4_set_tokens_base(s);
1578 
1579     memset(last_dc, 0, sizeof(last_dc));
1580 
1581     for (plane = 0; plane < ((s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 1 : 3); plane++) {
1582         memset(eob_tracker, 0, sizeof(eob_tracker));
1583 
1584         /* initialise dc prediction */
1585         for (i = 0; i < s->fragment_width[!!plane]; i++)
1586             vp4_dc_predictor_reset(&s->dc_pred_row[i]);
1587 
1588         for (j = 0; j < 6; j++)
1589             for (i = 0; i < 6; i++)
1590                 vp4_dc_predictor_reset(&dc_pred[j][i]);
1591 
1592         for (sb_y = 0; sb_y * 4 < s->fragment_height[!!plane]; sb_y++) {
1593             for (sb_x = 0; sb_x *4 < s->fragment_width[!!plane]; sb_x++) {
1594                 vp4_dc_pred_before(s, dc_pred, sb_x);
1595                 for (j = 0; j < 16; j++) {
1596                         int hx = hilbert_offset[j][0];
1597                         int hy = hilbert_offset[j][1];
1598                         int x  = 4 * sb_x + hx;
1599                         int y  = 4 * sb_y + hy;
1600                         VP4Predictor *this_dc_pred = &dc_pred[hy + 1][hx + 1];
1601                         int fragment, dc_block_type;
1602 
1603                         if (x >= s->fragment_width[!!plane] || y >= s->fragment_height[!!plane])
1604                             continue;
1605 
1606                         fragment = s->fragment_start[plane] + y * s->fragment_width[!!plane] + x;
1607 
1608                         if (s->all_fragments[fragment].coding_method == MODE_COPY)
1609                             continue;
1610 
1611                         if (vp4_unpack_vlcs(s, gb, tables[!!plane], plane, eob_tracker, fragment) < 0)
1612                             return -1;
1613 
1614                         dc_block_type = vp4_pred_block_type_map[s->all_fragments[fragment].coding_method];
1615 
1616                         s->all_fragments[fragment].dc +=
1617                             vp4_dc_pred(s, this_dc_pred, last_dc, dc_block_type, plane);
1618 
1619                         this_dc_pred->type = dc_block_type,
1620                         this_dc_pred->dc   = last_dc[dc_block_type] = s->all_fragments[fragment].dc;
1621                 }
1622                 vp4_dc_pred_after(s, dc_pred, sb_x);
1623             }
1624         }
1625     }
1626 
1627     vp4_set_tokens_base(s);
1628 
1629     return 0;
1630 }
1631 #endif
1632 
1633 /*
1634  * This function reverses the DC prediction for each coded fragment in
1635  * the frame. Much of this function is adapted directly from the original
1636  * VP3 source code.
1637  */
1638 #define COMPATIBLE_FRAME(x)                                                   \
1639     (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1640 #define DC_COEFF(u) s->all_fragments[u].dc
1641 
reverse_dc_prediction(Vp3DecodeContext * s,int first_fragment,int fragment_width,int fragment_height)1642 static void reverse_dc_prediction(Vp3DecodeContext *s,
1643                                   int first_fragment,
1644                                   int fragment_width,
1645                                   int fragment_height)
1646 {
1647 #define PUL 8
1648 #define PU 4
1649 #define PUR 2
1650 #define PL 1
1651 
1652     int x, y;
1653     int i = first_fragment;
1654 
1655     int predicted_dc;
1656 
1657     /* DC values for the left, up-left, up, and up-right fragments */
1658     int vl, vul, vu, vur;
1659 
1660     /* indexes for the left, up-left, up, and up-right fragments */
1661     int l, ul, u, ur;
1662 
1663     /*
1664      * The 6 fields mean:
1665      *   0: up-left multiplier
1666      *   1: up multiplier
1667      *   2: up-right multiplier
1668      *   3: left multiplier
1669      */
1670     static const int predictor_transform[16][4] = {
1671         {    0,   0,   0,   0 },
1672         {    0,   0,   0, 128 }, // PL
1673         {    0,   0, 128,   0 }, // PUR
1674         {    0,   0,  53,  75 }, // PUR|PL
1675         {    0, 128,   0,   0 }, // PU
1676         {    0,  64,   0,  64 }, // PU |PL
1677         {    0, 128,   0,   0 }, // PU |PUR
1678         {    0,   0,  53,  75 }, // PU |PUR|PL
1679         {  128,   0,   0,   0 }, // PUL
1680         {    0,   0,   0, 128 }, // PUL|PL
1681         {   64,   0,  64,   0 }, // PUL|PUR
1682         {    0,   0,  53,  75 }, // PUL|PUR|PL
1683         {    0, 128,   0,   0 }, // PUL|PU
1684         { -104, 116,   0, 116 }, // PUL|PU |PL
1685         {   24,  80,  24,   0 }, // PUL|PU |PUR
1686         { -104, 116,   0, 116 }  // PUL|PU |PUR|PL
1687     };
1688 
1689     /* This table shows which types of blocks can use other blocks for
1690      * prediction. For example, INTRA is the only mode in this table to
1691      * have a frame number of 0. That means INTRA blocks can only predict
1692      * from other INTRA blocks. There are 2 golden frame coding types;
1693      * blocks encoding in these modes can only predict from other blocks
1694      * that were encoded with these 1 of these 2 modes. */
1695     static const unsigned char compatible_frame[9] = {
1696         1,    /* MODE_INTER_NO_MV */
1697         0,    /* MODE_INTRA */
1698         1,    /* MODE_INTER_PLUS_MV */
1699         1,    /* MODE_INTER_LAST_MV */
1700         1,    /* MODE_INTER_PRIOR_MV */
1701         2,    /* MODE_USING_GOLDEN */
1702         2,    /* MODE_GOLDEN_MV */
1703         1,    /* MODE_INTER_FOUR_MV */
1704         3     /* MODE_COPY */
1705     };
1706     int current_frame_type;
1707 
1708     /* there is a last DC predictor for each of the 3 frame types */
1709     short last_dc[3];
1710 
1711     int transform = 0;
1712 
1713     vul =
1714     vu  =
1715     vur =
1716     vl  = 0;
1717     last_dc[0] =
1718     last_dc[1] =
1719     last_dc[2] = 0;
1720 
1721     /* for each fragment row... */
1722     for (y = 0; y < fragment_height; y++) {
1723         /* for each fragment in a row... */
1724         for (x = 0; x < fragment_width; x++, i++) {
1725 
1726             /* reverse prediction if this block was coded */
1727             if (s->all_fragments[i].coding_method != MODE_COPY) {
1728                 current_frame_type =
1729                     compatible_frame[s->all_fragments[i].coding_method];
1730 
1731                 transform = 0;
1732                 if (x) {
1733                     l  = i - 1;
1734                     vl = DC_COEFF(l);
1735                     if (COMPATIBLE_FRAME(l))
1736                         transform |= PL;
1737                 }
1738                 if (y) {
1739                     u  = i - fragment_width;
1740                     vu = DC_COEFF(u);
1741                     if (COMPATIBLE_FRAME(u))
1742                         transform |= PU;
1743                     if (x) {
1744                         ul  = i - fragment_width - 1;
1745                         vul = DC_COEFF(ul);
1746                         if (COMPATIBLE_FRAME(ul))
1747                             transform |= PUL;
1748                     }
1749                     if (x + 1 < fragment_width) {
1750                         ur  = i - fragment_width + 1;
1751                         vur = DC_COEFF(ur);
1752                         if (COMPATIBLE_FRAME(ur))
1753                             transform |= PUR;
1754                     }
1755                 }
1756 
1757                 if (transform == 0) {
1758                     /* if there were no fragments to predict from, use last
1759                      * DC saved */
1760                     predicted_dc = last_dc[current_frame_type];
1761                 } else {
1762                     /* apply the appropriate predictor transform */
1763                     predicted_dc =
1764                         (predictor_transform[transform][0] * vul) +
1765                         (predictor_transform[transform][1] * vu) +
1766                         (predictor_transform[transform][2] * vur) +
1767                         (predictor_transform[transform][3] * vl);
1768 
1769                     predicted_dc /= 128;
1770 
1771                     /* check for outranging on the [ul u l] and
1772                      * [ul u ur l] predictors */
1773                     if ((transform == 15) || (transform == 13)) {
1774                         if (FFABS(predicted_dc - vu) > 128)
1775                             predicted_dc = vu;
1776                         else if (FFABS(predicted_dc - vl) > 128)
1777                             predicted_dc = vl;
1778                         else if (FFABS(predicted_dc - vul) > 128)
1779                             predicted_dc = vul;
1780                     }
1781                 }
1782 
1783                 /* at long last, apply the predictor */
1784                 DC_COEFF(i) += predicted_dc;
1785                 /* save the DC */
1786                 last_dc[current_frame_type] = DC_COEFF(i);
1787             }
1788         }
1789     }
1790 }
1791 
apply_loop_filter(Vp3DecodeContext * s,int plane,int ystart,int yend)1792 static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1793                               int ystart, int yend)
1794 {
1795     int x, y;
1796     int *bounding_values = s->bounding_values_array + 127;
1797 
1798     int width           = s->fragment_width[!!plane];
1799     int height          = s->fragment_height[!!plane];
1800     int fragment        = s->fragment_start[plane] + ystart * width;
1801     ptrdiff_t stride    = s->current_frame.f->linesize[plane];
1802     uint8_t *plane_data = s->current_frame.f->data[plane];
1803     if (!s->flipped_image)
1804         stride = -stride;
1805     plane_data += s->data_offset[plane] + 8 * ystart * stride;
1806 
1807     for (y = ystart; y < yend; y++) {
1808         for (x = 0; x < width; x++) {
1809             /* This code basically just deblocks on the edges of coded blocks.
1810              * However, it has to be much more complicated because of the
1811              * brain damaged deblock ordering used in VP3/Theora. Order matters
1812              * because some pixels get filtered twice. */
1813             if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1814                 /* do not perform left edge filter for left columns frags */
1815                 if (x > 0) {
1816                     s->vp3dsp.h_loop_filter(
1817                         plane_data + 8 * x,
1818                         stride, bounding_values);
1819                 }
1820 
1821                 /* do not perform top edge filter for top row fragments */
1822                 if (y > 0) {
1823                     s->vp3dsp.v_loop_filter(
1824                         plane_data + 8 * x,
1825                         stride, bounding_values);
1826                 }
1827 
1828                 /* do not perform right edge filter for right column
1829                  * fragments or if right fragment neighbor is also coded
1830                  * in this frame (it will be filtered in next iteration) */
1831                 if ((x < width - 1) &&
1832                     (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1833                     s->vp3dsp.h_loop_filter(
1834                         plane_data + 8 * x + 8,
1835                         stride, bounding_values);
1836                 }
1837 
1838                 /* do not perform bottom edge filter for bottom row
1839                  * fragments or if bottom fragment neighbor is also coded
1840                  * in this frame (it will be filtered in the next row) */
1841                 if ((y < height - 1) &&
1842                     (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1843                     s->vp3dsp.v_loop_filter(
1844                         plane_data + 8 * x + 8 * stride,
1845                         stride, bounding_values);
1846                 }
1847             }
1848 
1849             fragment++;
1850         }
1851         plane_data += 8 * stride;
1852     }
1853 }
1854 
1855 /**
1856  * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1857  * for the next block in coding order
1858  */
vp3_dequant(Vp3DecodeContext * s,Vp3Fragment * frag,int plane,int inter,int16_t block[64])1859 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1860                               int plane, int inter, int16_t block[64])
1861 {
1862     int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1863     uint8_t *perm = s->idct_scantable;
1864     int i = 0;
1865 
1866     do {
1867         int token = *s->dct_tokens[plane][i];
1868         switch (token & 3) {
1869         case 0: // EOB
1870             if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1871                 s->dct_tokens[plane][i]++;
1872             else
1873                 *s->dct_tokens[plane][i] = token & ~3;
1874             goto end;
1875         case 1: // zero run
1876             s->dct_tokens[plane][i]++;
1877             i += (token >> 2) & 0x7f;
1878             if (i > 63) {
1879                 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1880                 return i;
1881             }
1882             block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1883             i++;
1884             break;
1885         case 2: // coeff
1886             block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1887             s->dct_tokens[plane][i++]++;
1888             break;
1889         default: // shouldn't happen
1890             return i;
1891         }
1892     } while (i < 64);
1893     // return value is expected to be a valid level
1894     i--;
1895 end:
1896     // the actual DC+prediction is in the fragment structure
1897     block[0] = frag->dc * s->qmat[0][inter][plane][0];
1898     return i;
1899 }
1900 
1901 /**
1902  * called when all pixels up to row y are complete
1903  */
vp3_draw_horiz_band(Vp3DecodeContext * s,int y)1904 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1905 {
1906     int h, cy, i;
1907     int offset[AV_NUM_DATA_POINTERS];
1908 
1909     if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1910         int y_flipped = s->flipped_image ? s->height - y : y;
1911 
1912         /* At the end of the frame, report INT_MAX instead of the height of
1913          * the frame. This makes the other threads' ff_thread_await_progress()
1914          * calls cheaper, because they don't have to clip their values. */
1915         ff_thread_report_progress(&s->current_frame,
1916                                   y_flipped == s->height ? INT_MAX
1917                                                          : y_flipped - 1,
1918                                   0);
1919     }
1920 
1921     if (!s->avctx->draw_horiz_band)
1922         return;
1923 
1924     h = y - s->last_slice_end;
1925     s->last_slice_end = y;
1926     y -= h;
1927 
1928     if (!s->flipped_image)
1929         y = s->height - y - h;
1930 
1931     cy        = y >> s->chroma_y_shift;
1932     offset[0] = s->current_frame.f->linesize[0] * y;
1933     offset[1] = s->current_frame.f->linesize[1] * cy;
1934     offset[2] = s->current_frame.f->linesize[2] * cy;
1935     for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1936         offset[i] = 0;
1937 
1938     emms_c();
1939     s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1940 }
1941 
1942 /**
1943  * Wait for the reference frame of the current fragment.
1944  * The progress value is in luma pixel rows.
1945  */
await_reference_row(Vp3DecodeContext * s,Vp3Fragment * fragment,int motion_y,int y)1946 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1947                                 int motion_y, int y)
1948 {
1949     ThreadFrame *ref_frame;
1950     int ref_row;
1951     int border = motion_y & 1;
1952 
1953     if (fragment->coding_method == MODE_USING_GOLDEN ||
1954         fragment->coding_method == MODE_GOLDEN_MV)
1955         ref_frame = &s->golden_frame;
1956     else
1957         ref_frame = &s->last_frame;
1958 
1959     ref_row = y + (motion_y >> 1);
1960     ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1961 
1962     ff_thread_await_progress(ref_frame, ref_row, 0);
1963 }
1964 
1965 #if CONFIG_VP4_DECODER
1966 /**
1967  * @return non-zero if temp (edge_emu_buffer) was populated
1968  */
vp4_mc_loop_filter(Vp3DecodeContext * s,int plane,int motion_x,int motion_y,int bx,int by,uint8_t * motion_source,int stride,int src_x,int src_y,uint8_t * temp)1969 static int vp4_mc_loop_filter(Vp3DecodeContext *s, int plane, int motion_x, int motion_y, int bx, int by,
1970        uint8_t * motion_source, int stride, int src_x, int src_y, uint8_t *temp)
1971 {
1972     int motion_shift = plane ? 4 : 2;
1973     int subpel_mask = plane ? 3 : 1;
1974     int *bounding_values = s->bounding_values_array + 127;
1975 
1976     int i;
1977     int x, y;
1978     int x2, y2;
1979     int x_subpel, y_subpel;
1980     int x_offset, y_offset;
1981 
1982     int block_width = plane ? 8 : 16;
1983     int plane_width  = s->width  >> (plane && s->chroma_x_shift);
1984     int plane_height = s->height >> (plane && s->chroma_y_shift);
1985 
1986 #define loop_stride 12
1987     uint8_t loop[12 * loop_stride];
1988 
1989     /* using division instead of shift to correctly handle negative values */
1990     x = 8 * bx + motion_x / motion_shift;
1991     y = 8 * by + motion_y / motion_shift;
1992 
1993     x_subpel = motion_x & subpel_mask;
1994     y_subpel = motion_y & subpel_mask;
1995 
1996     if (x_subpel || y_subpel) {
1997         x--;
1998         y--;
1999 
2000         if (x_subpel)
2001             x = FFMIN(x, x + FFSIGN(motion_x));
2002 
2003         if (y_subpel)
2004             y = FFMIN(y, y + FFSIGN(motion_y));
2005 
2006         x2 = x + block_width;
2007         y2 = y + block_width;
2008 
2009         if (x2 < 0 || x2 >= plane_width || y2 < 0 || y2 >= plane_height)
2010             return 0;
2011 
2012         x_offset = (-(x + 2) & 7) + 2;
2013         y_offset = (-(y + 2) & 7) + 2;
2014 
2015         if (x_offset > 8 + x_subpel && y_offset > 8 + y_subpel)
2016             return 0;
2017 
2018         s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2019              loop_stride, stride,
2020              12, 12, src_x - 1, src_y - 1,
2021              plane_width,
2022              plane_height);
2023 
2024         if (x_offset <= 8 + x_subpel)
2025             ff_vp3dsp_h_loop_filter_12(loop + x_offset, loop_stride, bounding_values);
2026 
2027         if (y_offset <= 8 + y_subpel)
2028             ff_vp3dsp_v_loop_filter_12(loop + y_offset*loop_stride, loop_stride, bounding_values);
2029 
2030     } else {
2031 
2032         x_offset = -x & 7;
2033         y_offset = -y & 7;
2034 
2035         if (!x_offset && !y_offset)
2036             return 0;
2037 
2038         s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2039              loop_stride, stride,
2040              12, 12, src_x - 1, src_y - 1,
2041              plane_width,
2042              plane_height);
2043 
2044 #define safe_loop_filter(name, ptr, stride, bounding_values) \
2045     if ((uintptr_t)(ptr) & 7) \
2046         s->vp3dsp.name##_unaligned(ptr, stride, bounding_values); \
2047     else \
2048         s->vp3dsp.name(ptr, stride, bounding_values);
2049 
2050         if (x_offset)
2051             safe_loop_filter(h_loop_filter, loop + loop_stride + x_offset + 1, loop_stride, bounding_values);
2052 
2053         if (y_offset)
2054             safe_loop_filter(v_loop_filter, loop + (y_offset + 1)*loop_stride + 1, loop_stride, bounding_values);
2055     }
2056 
2057     for (i = 0; i < 9; i++)
2058         memcpy(temp + i*stride, loop + (i + 1) * loop_stride + 1, 9);
2059 
2060     return 1;
2061 }
2062 #endif
2063 
2064 /*
2065  * Perform the final rendering for a particular slice of data.
2066  * The slice number ranges from 0..(c_superblock_height - 1).
2067  */
render_slice(Vp3DecodeContext * s,int slice)2068 static void render_slice(Vp3DecodeContext *s, int slice)
2069 {
2070     int x, y, i, j, fragment;
2071     int16_t *block = s->block;
2072     int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
2073     int motion_halfpel_index;
2074     uint8_t *motion_source;
2075     int plane, first_pixel;
2076 
2077     if (slice >= s->c_superblock_height)
2078         return;
2079 
2080     for (plane = 0; plane < 3; plane++) {
2081         uint8_t *output_plane = s->current_frame.f->data[plane] +
2082                                 s->data_offset[plane];
2083         uint8_t *last_plane = s->last_frame.f->data[plane] +
2084                               s->data_offset[plane];
2085         uint8_t *golden_plane = s->golden_frame.f->data[plane] +
2086                                 s->data_offset[plane];
2087         ptrdiff_t stride = s->current_frame.f->linesize[plane];
2088         int plane_width  = s->width  >> (plane && s->chroma_x_shift);
2089         int plane_height = s->height >> (plane && s->chroma_y_shift);
2090         int8_t(*motion_val)[2] = s->motion_val[!!plane];
2091 
2092         int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
2093         int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
2094         int slice_width  = plane ? s->c_superblock_width
2095                                  : s->y_superblock_width;
2096 
2097         int fragment_width  = s->fragment_width[!!plane];
2098         int fragment_height = s->fragment_height[!!plane];
2099         int fragment_start  = s->fragment_start[plane];
2100 
2101         int do_await = !plane && HAVE_THREADS &&
2102                        (s->avctx->active_thread_type & FF_THREAD_FRAME);
2103 
2104         if (!s->flipped_image)
2105             stride = -stride;
2106         if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY))
2107             continue;
2108 
2109         /* for each superblock row in the slice (both of them)... */
2110         for (; sb_y < slice_height; sb_y++) {
2111             /* for each superblock in a row... */
2112             for (sb_x = 0; sb_x < slice_width; sb_x++) {
2113                 /* for each block in a superblock... */
2114                 for (j = 0; j < 16; j++) {
2115                     x        = 4 * sb_x + hilbert_offset[j][0];
2116                     y        = 4 * sb_y + hilbert_offset[j][1];
2117                     fragment = y * fragment_width + x;
2118 
2119                     i = fragment_start + fragment;
2120 
2121                     // bounds check
2122                     if (x >= fragment_width || y >= fragment_height)
2123                         continue;
2124 
2125                     first_pixel = 8 * y * stride + 8 * x;
2126 
2127                     if (do_await &&
2128                         s->all_fragments[i].coding_method != MODE_INTRA)
2129                         await_reference_row(s, &s->all_fragments[i],
2130                                             motion_val[fragment][1],
2131                                             (16 * y) >> s->chroma_y_shift);
2132 
2133                     /* transform if this block was coded */
2134                     if (s->all_fragments[i].coding_method != MODE_COPY) {
2135                         if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
2136                             (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
2137                             motion_source = golden_plane;
2138                         else
2139                             motion_source = last_plane;
2140 
2141                         motion_source       += first_pixel;
2142                         motion_halfpel_index = 0;
2143 
2144                         /* sort out the motion vector if this fragment is coded
2145                          * using a motion vector method */
2146                         if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
2147                             (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
2148                             int src_x, src_y;
2149                             int standard_mc = 1;
2150                             motion_x = motion_val[fragment][0];
2151                             motion_y = motion_val[fragment][1];
2152 #if CONFIG_VP4_DECODER
2153                             if (plane && s->version >= 2) {
2154                                 motion_x = (motion_x >> 1) | (motion_x & 1);
2155                                 motion_y = (motion_y >> 1) | (motion_y & 1);
2156                             }
2157 #endif
2158 
2159                             src_x = (motion_x >> 1) + 8 * x;
2160                             src_y = (motion_y >> 1) + 8 * y;
2161 
2162                             motion_halfpel_index = motion_x & 0x01;
2163                             motion_source       += (motion_x >> 1);
2164 
2165                             motion_halfpel_index |= (motion_y & 0x01) << 1;
2166                             motion_source        += ((motion_y >> 1) * stride);
2167 
2168 #if CONFIG_VP4_DECODER
2169                             if (s->version >= 2) {
2170                                 uint8_t *temp = s->edge_emu_buffer;
2171                                 if (stride < 0)
2172                                     temp -= 8 * stride;
2173                                 if (vp4_mc_loop_filter(s, plane, motion_val[fragment][0], motion_val[fragment][1], x, y, motion_source, stride, src_x, src_y, temp)) {
2174                                     motion_source = temp;
2175                                     standard_mc = 0;
2176                                 }
2177                             }
2178 #endif
2179 
2180                             if (standard_mc && (
2181                                 src_x < 0 || src_y < 0 ||
2182                                 src_x + 9 >= plane_width ||
2183                                 src_y + 9 >= plane_height)) {
2184                                 uint8_t *temp = s->edge_emu_buffer;
2185                                 if (stride < 0)
2186                                     temp -= 8 * stride;
2187 
2188                                 s->vdsp.emulated_edge_mc(temp, motion_source,
2189                                                          stride, stride,
2190                                                          9, 9, src_x, src_y,
2191                                                          plane_width,
2192                                                          plane_height);
2193                                 motion_source = temp;
2194                             }
2195                         }
2196 
2197                         /* first, take care of copying a block from either the
2198                          * previous or the golden frame */
2199                         if (s->all_fragments[i].coding_method != MODE_INTRA) {
2200                             /* Note, it is possible to implement all MC cases
2201                              * with put_no_rnd_pixels_l2 which would look more
2202                              * like the VP3 source but this would be slower as
2203                              * put_no_rnd_pixels_tab is better optimized */
2204                             if (motion_halfpel_index != 3) {
2205                                 s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
2206                                     output_plane + first_pixel,
2207                                     motion_source, stride, 8);
2208                             } else {
2209                                 /* d is 0 if motion_x and _y have the same sign,
2210                                  * else -1 */
2211                                 int d = (motion_x ^ motion_y) >> 31;
2212                                 s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
2213                                                                motion_source - d,
2214                                                                motion_source + stride + 1 + d,
2215                                                                stride, 8);
2216                             }
2217                         }
2218 
2219                         /* invert DCT and place (or add) in final output */
2220 
2221                         if (s->all_fragments[i].coding_method == MODE_INTRA) {
2222                             vp3_dequant(s, s->all_fragments + i,
2223                                         plane, 0, block);
2224                             s->vp3dsp.idct_put(output_plane + first_pixel,
2225                                                stride,
2226                                                block);
2227                         } else {
2228                             if (vp3_dequant(s, s->all_fragments + i,
2229                                             plane, 1, block)) {
2230                                 s->vp3dsp.idct_add(output_plane + first_pixel,
2231                                                    stride,
2232                                                    block);
2233                             } else {
2234                                 s->vp3dsp.idct_dc_add(output_plane + first_pixel,
2235                                                       stride, block);
2236                             }
2237                         }
2238                     } else {
2239                         /* copy directly from the previous frame */
2240                         s->hdsp.put_pixels_tab[1][0](
2241                             output_plane + first_pixel,
2242                             last_plane + first_pixel,
2243                             stride, 8);
2244                     }
2245                 }
2246             }
2247 
2248             // Filter up to the last row in the superblock row
2249             if (s->version < 2 && !s->skip_loop_filter)
2250                 apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
2251                                   FFMIN(4 * sb_y + 3, fragment_height - 1));
2252         }
2253     }
2254 
2255     /* this looks like a good place for slice dispatch... */
2256     /* algorithm:
2257      *   if (slice == s->macroblock_height - 1)
2258      *     dispatch (both last slice & 2nd-to-last slice);
2259      *   else if (slice > 0)
2260      *     dispatch (slice - 1);
2261      */
2262 
2263     vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
2264                                  s->height - 16));
2265 }
2266 
2267 /// Allocate tables for per-frame data in Vp3DecodeContext
allocate_tables(AVCodecContext * avctx)2268 static av_cold int allocate_tables(AVCodecContext *avctx)
2269 {
2270     Vp3DecodeContext *s = avctx->priv_data;
2271     int y_fragment_count, c_fragment_count;
2272 
2273     free_tables(avctx);
2274 
2275     y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
2276     c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
2277 
2278     /* superblock_coding is used by unpack_superblocks (VP3/Theora) and vp4_unpack_macroblocks (VP4) */
2279     s->superblock_coding = av_mallocz(FFMAX(s->superblock_count, s->yuv_macroblock_count));
2280     s->all_fragments     = av_mallocz_array(s->fragment_count, sizeof(Vp3Fragment));
2281 
2282     s-> kf_coded_fragment_list = av_mallocz_array(s->fragment_count, sizeof(int));
2283     s->nkf_coded_fragment_list = av_mallocz_array(s->fragment_count, sizeof(int));
2284     memset(s-> num_kf_coded_fragment, -1, sizeof(s-> num_kf_coded_fragment));
2285 
2286     s->dct_tokens_base = av_mallocz_array(s->fragment_count,
2287                                           64 * sizeof(*s->dct_tokens_base));
2288     s->motion_val[0] = av_mallocz_array(y_fragment_count, sizeof(*s->motion_val[0]));
2289     s->motion_val[1] = av_mallocz_array(c_fragment_count, sizeof(*s->motion_val[1]));
2290 
2291     /* work out the block mapping tables */
2292     s->superblock_fragments = av_mallocz_array(s->superblock_count, 16 * sizeof(int));
2293     s->macroblock_coding    = av_mallocz(s->macroblock_count + 1);
2294 
2295     s->dc_pred_row = av_malloc_array(s->y_superblock_width * 4, sizeof(*s->dc_pred_row));
2296 
2297     if (!s->superblock_coding    || !s->all_fragments          ||
2298         !s->dct_tokens_base      || !s->kf_coded_fragment_list ||
2299         !s->nkf_coded_fragment_list ||
2300         !s->superblock_fragments || !s->macroblock_coding      ||
2301         !s->dc_pred_row ||
2302         !s->motion_val[0]        || !s->motion_val[1]) {
2303         return -1;
2304     }
2305 
2306     init_block_mapping(s);
2307 
2308     return 0;
2309 }
2310 
init_frames(Vp3DecodeContext * s)2311 static av_cold int init_frames(Vp3DecodeContext *s)
2312 {
2313     s->current_frame.f = av_frame_alloc();
2314     s->last_frame.f    = av_frame_alloc();
2315     s->golden_frame.f  = av_frame_alloc();
2316 
2317     if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f)
2318         return AVERROR(ENOMEM);
2319 
2320     return 0;
2321 }
2322 
vp3_decode_init(AVCodecContext * avctx)2323 static av_cold int vp3_decode_init(AVCodecContext *avctx)
2324 {
2325     Vp3DecodeContext *s = avctx->priv_data;
2326     int i, inter, plane, ret;
2327     int c_width;
2328     int c_height;
2329     int y_fragment_count, c_fragment_count;
2330 #if CONFIG_VP4_DECODER
2331     int j;
2332 #endif
2333 
2334     ret = init_frames(s);
2335     if (ret < 0)
2336         return ret;
2337 
2338     if (avctx->codec_tag == MKTAG('V', 'P', '4', '0'))
2339         s->version = 3;
2340     else if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
2341         s->version = 0;
2342     else
2343         s->version = 1;
2344 
2345     s->avctx  = avctx;
2346     s->width  = FFALIGN(avctx->coded_width, 16);
2347     s->height = FFALIGN(avctx->coded_height, 16);
2348     if (avctx->codec_id != AV_CODEC_ID_THEORA)
2349         avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2350     avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
2351     ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT);
2352     ff_videodsp_init(&s->vdsp, 8);
2353     ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
2354 
2355     for (i = 0; i < 64; i++) {
2356 #define TRANSPOSE(x) (((x) >> 3) | (((x) & 7) << 3))
2357         s->idct_permutation[i] = TRANSPOSE(i);
2358         s->idct_scantable[i]   = TRANSPOSE(ff_zigzag_direct[i]);
2359 #undef TRANSPOSE
2360     }
2361 
2362     /* initialize to an impossible value which will force a recalculation
2363      * in the first frame decode */
2364     for (i = 0; i < 3; i++)
2365         s->qps[i] = -1;
2366 
2367     ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
2368     if (ret)
2369         return ret;
2370 
2371     s->y_superblock_width  = (s->width  + 31) / 32;
2372     s->y_superblock_height = (s->height + 31) / 32;
2373     s->y_superblock_count  = s->y_superblock_width * s->y_superblock_height;
2374 
2375     /* work out the dimensions for the C planes */
2376     c_width                = s->width >> s->chroma_x_shift;
2377     c_height               = s->height >> s->chroma_y_shift;
2378     s->c_superblock_width  = (c_width  + 31) / 32;
2379     s->c_superblock_height = (c_height + 31) / 32;
2380     s->c_superblock_count  = s->c_superblock_width * s->c_superblock_height;
2381 
2382     s->superblock_count   = s->y_superblock_count + (s->c_superblock_count * 2);
2383     s->u_superblock_start = s->y_superblock_count;
2384     s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
2385 
2386     s->macroblock_width  = (s->width  + 15) / 16;
2387     s->macroblock_height = (s->height + 15) / 16;
2388     s->macroblock_count  = s->macroblock_width * s->macroblock_height;
2389     s->c_macroblock_width  = (c_width  + 15) / 16;
2390     s->c_macroblock_height = (c_height + 15) / 16;
2391     s->c_macroblock_count  = s->c_macroblock_width * s->c_macroblock_height;
2392     s->yuv_macroblock_count = s->macroblock_count + 2 * s->c_macroblock_count;
2393 
2394     s->fragment_width[0]  = s->width / FRAGMENT_PIXELS;
2395     s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
2396     s->fragment_width[1]  = s->fragment_width[0] >> s->chroma_x_shift;
2397     s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
2398 
2399     /* fragment count covers all 8x8 blocks for all 3 planes */
2400     y_fragment_count     = s->fragment_width[0] * s->fragment_height[0];
2401     c_fragment_count     = s->fragment_width[1] * s->fragment_height[1];
2402     s->fragment_count    = y_fragment_count + 2 * c_fragment_count;
2403     s->fragment_start[1] = y_fragment_count;
2404     s->fragment_start[2] = y_fragment_count + c_fragment_count;
2405 
2406     if (!s->theora_tables) {
2407         for (i = 0; i < 64; i++) {
2408             s->coded_dc_scale_factor[0][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_y_dc_scale_factor[i];
2409             s->coded_dc_scale_factor[1][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_uv_dc_scale_factor[i];
2410             s->coded_ac_scale_factor[i] = s->version < 2 ? vp31_ac_scale_factor[i] : vp4_ac_scale_factor[i];
2411             s->base_matrix[0][i]        = s->version < 2 ? vp31_intra_y_dequant[i] : vp4_generic_dequant[i];
2412             s->base_matrix[1][i]        = s->version < 2 ? vp31_intra_c_dequant[i] : vp4_generic_dequant[i];
2413             s->base_matrix[2][i]        = s->version < 2 ? vp31_inter_dequant[i]   : vp4_generic_dequant[i];
2414             s->filter_limit_values[i]   = s->version < 2 ? vp31_filter_limit_values[i] : vp4_filter_limit_values[i];
2415         }
2416 
2417         for (inter = 0; inter < 2; inter++) {
2418             for (plane = 0; plane < 3; plane++) {
2419                 s->qr_count[inter][plane]   = 1;
2420                 s->qr_size[inter][plane][0] = 63;
2421                 s->qr_base[inter][plane][0] =
2422                 s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
2423             }
2424         }
2425 
2426         /* init VLC tables */
2427         if (s->version < 2) {
2428             for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++) {
2429                 ret = ff_init_vlc_from_lengths(&s->coeff_vlc[i], 11, 32,
2430                                                &vp3_bias[i][0][1], 2,
2431                                                &vp3_bias[i][0][0], 2, 1,
2432                                                0, 0, avctx);
2433                 if (ret < 0)
2434                     return ret;
2435             }
2436 #if CONFIG_VP4_DECODER
2437         } else { /* version >= 2 */
2438             for (i = 0; i < FF_ARRAY_ELEMS(s->coeff_vlc); i++) {
2439                 ret = ff_init_vlc_from_lengths(&s->coeff_vlc[i], 11, 32,
2440                                                &vp4_bias[i][0][1], 2,
2441                                                &vp4_bias[i][0][0], 2, 1,
2442                                                0, 0, avctx);
2443                 if (ret < 0)
2444                     return ret;
2445             }
2446 #endif
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
update_frames(AVCodecContext * avctx)2509 static 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_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_buffer(avctx, &s->golden_frame);
2522         ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
2523     }
2524 
2525 fail:
2526     ff_thread_release_buffer(avctx, &s->current_frame);
2527     return ret;
2528 }
2529 
2530 #if HAVE_THREADS
ref_frame(Vp3DecodeContext * s,ThreadFrame * dst,ThreadFrame * src)2531 static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
2532 {
2533     ff_thread_release_buffer(s->avctx, dst);
2534     if (src->f->data[0])
2535         return ff_thread_ref_frame(dst, src);
2536     return 0;
2537 }
2538 
ref_frames(Vp3DecodeContext * dst,Vp3DecodeContext * src)2539 static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
2540 {
2541     int ret;
2542     if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
2543         (ret = ref_frame(dst, &dst->golden_frame,  &src->golden_frame)) < 0  ||
2544         (ret = ref_frame(dst, &dst->last_frame,    &src->last_frame)) < 0)
2545         return ret;
2546     return 0;
2547 }
2548 
vp3_update_thread_context(AVCodecContext * dst,const AVCodecContext * src)2549 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
2550 {
2551     Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
2552     int qps_changed = 0, i, err;
2553 
2554     if (!s1->current_frame.f->data[0] ||
2555         s->width != s1->width || s->height != s1->height) {
2556         if (s != s1)
2557             ref_frames(s, s1);
2558         return -1;
2559     }
2560 
2561     if (s != s1) {
2562         // copy previous frame data
2563         if ((err = ref_frames(s, s1)) < 0)
2564             return err;
2565 
2566         s->keyframe = s1->keyframe;
2567 
2568         // copy qscale data if necessary
2569         for (i = 0; i < 3; i++) {
2570             if (s->qps[i] != s1->qps[1]) {
2571                 qps_changed = 1;
2572                 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
2573             }
2574         }
2575 
2576         if (s->qps[0] != s1->qps[0])
2577             memcpy(&s->bounding_values_array, &s1->bounding_values_array,
2578                    sizeof(s->bounding_values_array));
2579 
2580         if (qps_changed) {
2581             memcpy(s->qps,      s1->qps,      sizeof(s->qps));
2582             memcpy(s->last_qps, s1->last_qps, sizeof(s->last_qps));
2583             s->nqps = s1->nqps;
2584         }
2585     }
2586 
2587     return update_frames(dst);
2588 }
2589 #endif
2590 
vp3_decode_frame(AVCodecContext * avctx,void * data,int * got_frame,AVPacket * avpkt)2591 static int vp3_decode_frame(AVCodecContext *avctx,
2592                             void *data, int *got_frame,
2593                             AVPacket *avpkt)
2594 {
2595     AVFrame     *frame  = data;
2596     const uint8_t *buf  = avpkt->data;
2597     int buf_size        = avpkt->size;
2598     Vp3DecodeContext *s = avctx->priv_data;
2599     GetBitContext gb;
2600     int i, ret;
2601 
2602     if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0)
2603         return ret;
2604 
2605 #if CONFIG_THEORA_DECODER
2606     if (s->theora && get_bits1(&gb)) {
2607         int type = get_bits(&gb, 7);
2608         skip_bits_long(&gb, 6*8); /* "theora" */
2609 
2610         if (s->avctx->active_thread_type&FF_THREAD_FRAME) {
2611             av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n");
2612             return AVERROR_PATCHWELCOME;
2613         }
2614         if (type == 0) {
2615             vp3_decode_end(avctx);
2616             ret = theora_decode_header(avctx, &gb);
2617 
2618             if (ret >= 0)
2619                 ret = vp3_decode_init(avctx);
2620             if (ret < 0) {
2621                 vp3_decode_end(avctx);
2622                 return ret;
2623             }
2624             return buf_size;
2625         } else if (type == 2) {
2626             vp3_decode_end(avctx);
2627             ret = theora_decode_tables(avctx, &gb);
2628             if (ret >= 0)
2629                 ret = vp3_decode_init(avctx);
2630             if (ret < 0) {
2631                 vp3_decode_end(avctx);
2632                 return ret;
2633             }
2634             return buf_size;
2635         }
2636 
2637         av_log(avctx, AV_LOG_ERROR,
2638                "Header packet passed to frame decoder, skipping\n");
2639         return -1;
2640     }
2641 #endif
2642 
2643     s->keyframe = !get_bits1(&gb);
2644     if (!s->all_fragments) {
2645         av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n");
2646         return -1;
2647     }
2648     if (!s->theora)
2649         skip_bits(&gb, 1);
2650     for (i = 0; i < 3; i++)
2651         s->last_qps[i] = s->qps[i];
2652 
2653     s->nqps = 0;
2654     do {
2655         s->qps[s->nqps++] = get_bits(&gb, 6);
2656     } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2657     for (i = s->nqps; i < 3; i++)
2658         s->qps[i] = -1;
2659 
2660     if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2661         av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2662                s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2663 
2664     s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2665                           avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2666                                                                   : AVDISCARD_NONKEY);
2667 
2668     if (s->qps[0] != s->last_qps[0])
2669         init_loop_filter(s);
2670 
2671     for (i = 0; i < s->nqps; i++)
2672         // reinit all dequantizers if the first one changed, because
2673         // the DC of the first quantizer must be used for all matrices
2674         if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2675             init_dequantizer(s, i);
2676 
2677     if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2678         return buf_size;
2679 
2680     s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2681                                                 : AV_PICTURE_TYPE_P;
2682     s->current_frame.f->key_frame = s->keyframe;
2683     if ((ret = ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF)) < 0)
2684         goto error;
2685 
2686     if (!s->edge_emu_buffer) {
2687         s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2688         if (!s->edge_emu_buffer) {
2689             ret = AVERROR(ENOMEM);
2690             goto error;
2691         }
2692     }
2693 
2694     if (s->keyframe) {
2695         if (!s->theora) {
2696             skip_bits(&gb, 4); /* width code */
2697             skip_bits(&gb, 4); /* height code */
2698             if (s->version) {
2699                 s->version = get_bits(&gb, 5);
2700                 if (avctx->frame_number == 0)
2701                     av_log(s->avctx, AV_LOG_DEBUG,
2702                            "VP version: %d\n", s->version);
2703             }
2704         }
2705         if (s->version || s->theora) {
2706             if (get_bits1(&gb))
2707                 av_log(s->avctx, AV_LOG_ERROR,
2708                        "Warning, unsupported keyframe coding type?!\n");
2709             skip_bits(&gb, 2); /* reserved? */
2710 
2711 #if CONFIG_VP4_DECODER
2712             if (s->version >= 2) {
2713                 int mb_height, mb_width;
2714                 int mb_width_mul, mb_width_div, mb_height_mul, mb_height_div;
2715 
2716                 mb_height = get_bits(&gb, 8);
2717                 mb_width  = get_bits(&gb, 8);
2718                 if (mb_height != s->macroblock_height ||
2719                     mb_width != s->macroblock_width)
2720                     avpriv_request_sample(s->avctx, "macroblock dimension mismatch");
2721 
2722                 mb_width_mul = get_bits(&gb, 5);
2723                 mb_width_div = get_bits(&gb, 3);
2724                 mb_height_mul = get_bits(&gb, 5);
2725                 mb_height_div = get_bits(&gb, 3);
2726                 if (mb_width_mul != 1 || mb_width_div != 1 || mb_height_mul != 1 || mb_height_div != 1)
2727                     avpriv_request_sample(s->avctx, "unexpected macroblock dimension multipler/divider");
2728 
2729                 if (get_bits(&gb, 2))
2730                     avpriv_request_sample(s->avctx, "unknown bits");
2731             }
2732 #endif
2733         }
2734     } else {
2735         if (!s->golden_frame.f->data[0]) {
2736             av_log(s->avctx, AV_LOG_WARNING,
2737                    "vp3: first frame not a keyframe\n");
2738 
2739             s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2740             if ((ret = ff_thread_get_buffer(avctx, &s->golden_frame,
2741                                      AV_GET_BUFFER_FLAG_REF)) < 0)
2742                 goto error;
2743             ff_thread_release_buffer(avctx, &s->last_frame);
2744             if ((ret = ff_thread_ref_frame(&s->last_frame,
2745                                            &s->golden_frame)) < 0)
2746                 goto error;
2747             ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2748         }
2749     }
2750 
2751     memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2752     ff_thread_finish_setup(avctx);
2753 
2754     if (s->version < 2) {
2755         if ((ret = unpack_superblocks(s, &gb)) < 0) {
2756             av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2757             goto error;
2758         }
2759 #if CONFIG_VP4_DECODER
2760     } else {
2761         if ((ret = vp4_unpack_macroblocks(s, &gb)) < 0) {
2762             av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_macroblocks\n");
2763             goto error;
2764     }
2765 #endif
2766     }
2767     if ((ret = unpack_modes(s, &gb)) < 0) {
2768         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2769         goto error;
2770     }
2771     if (ret = unpack_vectors(s, &gb)) {
2772         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2773         goto error;
2774     }
2775     if ((ret = unpack_block_qpis(s, &gb)) < 0) {
2776         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2777         goto error;
2778     }
2779 
2780     if (s->version < 2) {
2781         if ((ret = unpack_dct_coeffs(s, &gb)) < 0) {
2782             av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2783             goto error;
2784         }
2785 #if CONFIG_VP4_DECODER
2786     } else {
2787         if ((ret = vp4_unpack_dct_coeffs(s, &gb)) < 0) {
2788             av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_dct_coeffs\n");
2789             goto error;
2790         }
2791 #endif
2792     }
2793 
2794     for (i = 0; i < 3; i++) {
2795         int height = s->height >> (i && s->chroma_y_shift);
2796         if (s->flipped_image)
2797             s->data_offset[i] = 0;
2798         else
2799             s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2800     }
2801 
2802     s->last_slice_end = 0;
2803     for (i = 0; i < s->c_superblock_height; i++)
2804         render_slice(s, i);
2805 
2806     // filter the last row
2807     if (s->version < 2)
2808         for (i = 0; i < 3; i++) {
2809             int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2810             apply_loop_filter(s, i, row, row + 1);
2811         }
2812     vp3_draw_horiz_band(s, s->height);
2813 
2814     /* output frame, offset as needed */
2815     if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2816         return ret;
2817 
2818     frame->crop_left   = s->offset_x;
2819     frame->crop_right  = avctx->coded_width - avctx->width - s->offset_x;
2820     frame->crop_top    = s->offset_y;
2821     frame->crop_bottom = avctx->coded_height - avctx->height - s->offset_y;
2822 
2823     *got_frame = 1;
2824 
2825     if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2826         ret = update_frames(avctx);
2827         if (ret < 0)
2828             return ret;
2829     }
2830 
2831     return buf_size;
2832 
2833 error:
2834     ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2835 
2836     if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2837         av_frame_unref(s->current_frame.f);
2838 
2839     return ret;
2840 }
2841 
read_huffman_tree(HuffTable * huff,GetBitContext * gb,int length,AVCodecContext * avctx)2842 static int read_huffman_tree(HuffTable *huff, GetBitContext *gb, int length,
2843                              AVCodecContext *avctx)
2844 {
2845     if (get_bits1(gb)) {
2846         int token;
2847         if (huff->nb_entries >= 32) { /* overflow */
2848             av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2849             return -1;
2850         }
2851         token = get_bits(gb, 5);
2852         ff_dlog(avctx, "code length %d, curr entry %d, token %d\n",
2853                 length, huff->nb_entries, token);
2854         huff->entries[huff->nb_entries++] = (HuffEntry){ length, token };
2855     } else {
2856         /* The following bound follows from the fact that nb_entries <= 32. */
2857         if (length >= 31) { /* overflow */
2858             av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2859             return -1;
2860         }
2861         length++;
2862         if (read_huffman_tree(huff, gb, length, avctx))
2863             return -1;
2864         if (read_huffman_tree(huff, gb, length, avctx))
2865             return -1;
2866     }
2867     return 0;
2868 }
2869 
2870 #if CONFIG_THEORA_DECODER
2871 static const enum AVPixelFormat theora_pix_fmts[4] = {
2872     AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2873 };
2874 
theora_decode_header(AVCodecContext * avctx,GetBitContext * gb)2875 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2876 {
2877     Vp3DecodeContext *s = avctx->priv_data;
2878     int visible_width, visible_height, colorspace;
2879     uint8_t offset_x = 0, offset_y = 0;
2880     int ret;
2881     AVRational fps, aspect;
2882 
2883     if (get_bits_left(gb) < 206)
2884         return AVERROR_INVALIDDATA;
2885 
2886     s->theora_header = 0;
2887     s->theora = get_bits(gb, 24);
2888     av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2889     if (!s->theora) {
2890         s->theora = 1;
2891         avpriv_request_sample(s->avctx, "theora 0");
2892     }
2893 
2894     /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2895      * but previous versions have the image flipped relative to vp3 */
2896     if (s->theora < 0x030200) {
2897         s->flipped_image = 1;
2898         av_log(avctx, AV_LOG_DEBUG,
2899                "Old (<alpha3) Theora bitstream, flipped image\n");
2900     }
2901 
2902     visible_width  =
2903     s->width       = get_bits(gb, 16) << 4;
2904     visible_height =
2905     s->height      = get_bits(gb, 16) << 4;
2906 
2907     if (s->theora >= 0x030200) {
2908         visible_width  = get_bits(gb, 24);
2909         visible_height = get_bits(gb, 24);
2910 
2911         offset_x = get_bits(gb, 8); /* offset x */
2912         offset_y = get_bits(gb, 8); /* offset y, from bottom */
2913     }
2914 
2915     /* sanity check */
2916     if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 ||
2917         visible_width  + offset_x > s->width ||
2918         visible_height + offset_y > s->height) {
2919         av_log(avctx, AV_LOG_ERROR,
2920                "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n",
2921                visible_width, visible_height, offset_x, offset_y,
2922                s->width, s->height);
2923         return AVERROR_INVALIDDATA;
2924     }
2925 
2926     fps.num = get_bits_long(gb, 32);
2927     fps.den = get_bits_long(gb, 32);
2928     if (fps.num && fps.den) {
2929         if (fps.num < 0 || fps.den < 0) {
2930             av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2931             return AVERROR_INVALIDDATA;
2932         }
2933         av_reduce(&avctx->framerate.den, &avctx->framerate.num,
2934                   fps.den, fps.num, 1 << 30);
2935     }
2936 
2937     aspect.num = get_bits(gb, 24);
2938     aspect.den = get_bits(gb, 24);
2939     if (aspect.num && aspect.den) {
2940         av_reduce(&avctx->sample_aspect_ratio.num,
2941                   &avctx->sample_aspect_ratio.den,
2942                   aspect.num, aspect.den, 1 << 30);
2943         ff_set_sar(avctx, avctx->sample_aspect_ratio);
2944     }
2945 
2946     if (s->theora < 0x030200)
2947         skip_bits(gb, 5); /* keyframe frequency force */
2948     colorspace = get_bits(gb, 8);
2949     skip_bits(gb, 24); /* bitrate */
2950 
2951     skip_bits(gb, 6); /* quality hint */
2952 
2953     if (s->theora >= 0x030200) {
2954         skip_bits(gb, 5); /* keyframe frequency force */
2955         avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2956         if (avctx->pix_fmt == AV_PIX_FMT_NONE) {
2957             av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n");
2958             return AVERROR_INVALIDDATA;
2959         }
2960         skip_bits(gb, 3); /* reserved */
2961     } else
2962         avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2963 
2964     ret = ff_set_dimensions(avctx, s->width, s->height);
2965     if (ret < 0)
2966         return ret;
2967     if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP)) {
2968         avctx->width  = visible_width;
2969         avctx->height = visible_height;
2970         // translate offsets from theora axis ([0,0] lower left)
2971         // to normal axis ([0,0] upper left)
2972         s->offset_x = offset_x;
2973         s->offset_y = s->height - visible_height - offset_y;
2974     }
2975 
2976     if (colorspace == 1)
2977         avctx->color_primaries = AVCOL_PRI_BT470M;
2978     else if (colorspace == 2)
2979         avctx->color_primaries = AVCOL_PRI_BT470BG;
2980 
2981     if (colorspace == 1 || colorspace == 2) {
2982         avctx->colorspace = AVCOL_SPC_BT470BG;
2983         avctx->color_trc  = AVCOL_TRC_BT709;
2984     }
2985 
2986     s->theora_header = 1;
2987     return 0;
2988 }
2989 
theora_decode_tables(AVCodecContext * avctx,GetBitContext * gb)2990 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2991 {
2992     Vp3DecodeContext *s = avctx->priv_data;
2993     int i, n, matrices, inter, plane, ret;
2994 
2995     if (!s->theora_header)
2996         return AVERROR_INVALIDDATA;
2997 
2998     if (s->theora >= 0x030200) {
2999         n = get_bits(gb, 3);
3000         /* loop filter limit values table */
3001         if (n)
3002             for (i = 0; i < 64; i++)
3003                 s->filter_limit_values[i] = get_bits(gb, n);
3004     }
3005 
3006     if (s->theora >= 0x030200)
3007         n = get_bits(gb, 4) + 1;
3008     else
3009         n = 16;
3010     /* quality threshold table */
3011     for (i = 0; i < 64; i++)
3012         s->coded_ac_scale_factor[i] = get_bits(gb, n);
3013 
3014     if (s->theora >= 0x030200)
3015         n = get_bits(gb, 4) + 1;
3016     else
3017         n = 16;
3018     /* dc scale factor table */
3019     for (i = 0; i < 64; i++)
3020         s->coded_dc_scale_factor[0][i] =
3021         s->coded_dc_scale_factor[1][i] = get_bits(gb, n);
3022 
3023     if (s->theora >= 0x030200)
3024         matrices = get_bits(gb, 9) + 1;
3025     else
3026         matrices = 3;
3027 
3028     if (matrices > 384) {
3029         av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
3030         return -1;
3031     }
3032 
3033     for (n = 0; n < matrices; n++)
3034         for (i = 0; i < 64; i++)
3035             s->base_matrix[n][i] = get_bits(gb, 8);
3036 
3037     for (inter = 0; inter <= 1; inter++) {
3038         for (plane = 0; plane <= 2; plane++) {
3039             int newqr = 1;
3040             if (inter || plane > 0)
3041                 newqr = get_bits1(gb);
3042             if (!newqr) {
3043                 int qtj, plj;
3044                 if (inter && get_bits1(gb)) {
3045                     qtj = 0;
3046                     plj = plane;
3047                 } else {
3048                     qtj = (3 * inter + plane - 1) / 3;
3049                     plj = (plane + 2) % 3;
3050                 }
3051                 s->qr_count[inter][plane] = s->qr_count[qtj][plj];
3052                 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
3053                        sizeof(s->qr_size[0][0]));
3054                 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
3055                        sizeof(s->qr_base[0][0]));
3056             } else {
3057                 int qri = 0;
3058                 int qi  = 0;
3059 
3060                 for (;;) {
3061                     i = get_bits(gb, av_log2(matrices - 1) + 1);
3062                     if (i >= matrices) {
3063                         av_log(avctx, AV_LOG_ERROR,
3064                                "invalid base matrix index\n");
3065                         return -1;
3066                     }
3067                     s->qr_base[inter][plane][qri] = i;
3068                     if (qi >= 63)
3069                         break;
3070                     i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
3071                     s->qr_size[inter][plane][qri++] = i;
3072                     qi += i;
3073                 }
3074 
3075                 if (qi > 63) {
3076                     av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
3077                     return -1;
3078                 }
3079                 s->qr_count[inter][plane] = qri;
3080             }
3081         }
3082     }
3083 
3084     /* Huffman tables */
3085     for (int i = 0; i < FF_ARRAY_ELEMS(s->huffman_table); i++) {
3086         s->huffman_table[i].nb_entries = 0;
3087         if ((ret = read_huffman_tree(&s->huffman_table[i], gb, 0, avctx)) < 0)
3088             return ret;
3089     }
3090 
3091     s->theora_tables = 1;
3092 
3093     return 0;
3094 }
3095 
theora_decode_init(AVCodecContext * avctx)3096 static av_cold int theora_decode_init(AVCodecContext *avctx)
3097 {
3098     Vp3DecodeContext *s = avctx->priv_data;
3099     GetBitContext gb;
3100     int ptype;
3101     const uint8_t *header_start[3];
3102     int header_len[3];
3103     int i;
3104     int ret;
3105 
3106     avctx->pix_fmt = AV_PIX_FMT_YUV420P;
3107 
3108     s->theora = 1;
3109 
3110     if (!avctx->extradata_size) {
3111         av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
3112         return -1;
3113     }
3114 
3115     if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
3116                                   42, header_start, header_len) < 0) {
3117         av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
3118         return -1;
3119     }
3120 
3121     for (i = 0; i < 3; i++) {
3122         if (header_len[i] <= 0)
3123             continue;
3124         ret = init_get_bits8(&gb, header_start[i], header_len[i]);
3125         if (ret < 0)
3126             return ret;
3127 
3128         ptype = get_bits(&gb, 8);
3129 
3130         if (!(ptype & 0x80)) {
3131             av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
3132 //          return -1;
3133         }
3134 
3135         // FIXME: Check for this as well.
3136         skip_bits_long(&gb, 6 * 8); /* "theora" */
3137 
3138         switch (ptype) {
3139         case 0x80:
3140             if (theora_decode_header(avctx, &gb) < 0)
3141                 return -1;
3142             break;
3143         case 0x81:
3144 // FIXME: is this needed? it breaks sometimes
3145 //            theora_decode_comments(avctx, gb);
3146             break;
3147         case 0x82:
3148             if (theora_decode_tables(avctx, &gb))
3149                 return -1;
3150             break;
3151         default:
3152             av_log(avctx, AV_LOG_ERROR,
3153                    "Unknown Theora config packet: %d\n", ptype & ~0x80);
3154             break;
3155         }
3156         if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb))
3157             av_log(avctx, AV_LOG_WARNING,
3158                    "%d bits left in packet %X\n",
3159                    8 * header_len[i] - get_bits_count(&gb), ptype);
3160         if (s->theora < 0x030200)
3161             break;
3162     }
3163 
3164     return vp3_decode_init(avctx);
3165 }
3166 
3167 AVCodec ff_theora_decoder = {
3168     .name                  = "theora",
3169     .long_name             = NULL_IF_CONFIG_SMALL("Theora"),
3170     .type                  = AVMEDIA_TYPE_VIDEO,
3171     .id                    = AV_CODEC_ID_THEORA,
3172     .priv_data_size        = sizeof(Vp3DecodeContext),
3173     .init                  = theora_decode_init,
3174     .close                 = vp3_decode_end,
3175     .decode                = vp3_decode_frame,
3176     .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3177                              AV_CODEC_CAP_FRAME_THREADS,
3178     .flush                 = vp3_decode_flush,
3179     .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3180     .caps_internal         = FF_CODEC_CAP_EXPORTS_CROPPING | FF_CODEC_CAP_ALLOCATE_PROGRESS |
3181                              FF_CODEC_CAP_INIT_CLEANUP,
3182 };
3183 #endif
3184 
3185 AVCodec ff_vp3_decoder = {
3186     .name                  = "vp3",
3187     .long_name             = NULL_IF_CONFIG_SMALL("On2 VP3"),
3188     .type                  = AVMEDIA_TYPE_VIDEO,
3189     .id                    = AV_CODEC_ID_VP3,
3190     .priv_data_size        = sizeof(Vp3DecodeContext),
3191     .init                  = vp3_decode_init,
3192     .close                 = vp3_decode_end,
3193     .decode                = vp3_decode_frame,
3194     .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3195                              AV_CODEC_CAP_FRAME_THREADS,
3196     .flush                 = vp3_decode_flush,
3197     .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3198     .caps_internal         = FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
3199 };
3200 
3201 #if CONFIG_VP4_DECODER
3202 AVCodec ff_vp4_decoder = {
3203     .name                  = "vp4",
3204     .long_name             = NULL_IF_CONFIG_SMALL("On2 VP4"),
3205     .type                  = AVMEDIA_TYPE_VIDEO,
3206     .id                    = AV_CODEC_ID_VP4,
3207     .priv_data_size        = sizeof(Vp3DecodeContext),
3208     .init                  = vp3_decode_init,
3209     .close                 = vp3_decode_end,
3210     .decode                = vp3_decode_frame,
3211     .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3212                              AV_CODEC_CAP_FRAME_THREADS,
3213     .flush                 = vp3_decode_flush,
3214     .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3215     .caps_internal         = FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
3216 };
3217 #endif
3218