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
39 #include "avcodec.h"
40 #include "get_bits.h"
41 #include "hpeldsp.h"
42 #include "internal.h"
43 #include "mathops.h"
44 #include "thread.h"
45 #include "videodsp.h"
46 #include "vp3data.h"
47 #include "vp4data.h"
48 #include "vp3dsp.h"
49 #include "xiph.h"
50
51 #define FRAGMENT_PIXELS 8
52
53 // FIXME split things out into their own arrays
54 typedef struct Vp3Fragment {
55 int16_t dc;
56 uint8_t coding_method;
57 uint8_t qpi;
58 } Vp3Fragment;
59
60 #define SB_NOT_CODED 0
61 #define SB_PARTIALLY_CODED 1
62 #define SB_FULLY_CODED 2
63
64 // This is the maximum length of a single long bit run that can be encoded
65 // for superblock coding or block qps. Theora special-cases this to read a
66 // bit instead of flipping the current bit to allow for runs longer than 4129.
67 #define MAXIMUM_LONG_BIT_RUN 4129
68
69 #define MODE_INTER_NO_MV 0
70 #define MODE_INTRA 1
71 #define MODE_INTER_PLUS_MV 2
72 #define MODE_INTER_LAST_MV 3
73 #define MODE_INTER_PRIOR_LAST 4
74 #define MODE_USING_GOLDEN 5
75 #define MODE_GOLDEN_MV 6
76 #define MODE_INTER_FOURMV 7
77 #define CODING_MODE_COUNT 8
78
79 /* special internal mode */
80 #define MODE_COPY 8
81
82 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb);
83 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb);
84
85
86 /* There are 6 preset schemes, plus a free-form scheme */
87 static const int ModeAlphabet[6][CODING_MODE_COUNT] = {
88 /* scheme 1: Last motion vector dominates */
89 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
90 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
91 MODE_INTRA, MODE_USING_GOLDEN,
92 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
93
94 /* scheme 2 */
95 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
96 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
97 MODE_INTRA, MODE_USING_GOLDEN,
98 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
99
100 /* scheme 3 */
101 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
102 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
103 MODE_INTRA, MODE_USING_GOLDEN,
104 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
105
106 /* scheme 4 */
107 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
108 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
109 MODE_INTRA, MODE_USING_GOLDEN,
110 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
111
112 /* scheme 5: No motion vector dominates */
113 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
114 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
115 MODE_INTRA, MODE_USING_GOLDEN,
116 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
117
118 /* scheme 6 */
119 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
120 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
121 MODE_INTER_PLUS_MV, MODE_INTRA,
122 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
123 };
124
125 static const uint8_t hilbert_offset[16][2] = {
126 { 0, 0 }, { 1, 0 }, { 1, 1 }, { 0, 1 },
127 { 0, 2 }, { 0, 3 }, { 1, 3 }, { 1, 2 },
128 { 2, 2 }, { 2, 3 }, { 3, 3 }, { 3, 2 },
129 { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 }
130 };
131
132 enum {
133 VP4_DC_INTRA = 0,
134 VP4_DC_INTER = 1,
135 VP4_DC_GOLDEN = 2,
136 NB_VP4_DC_TYPES,
137 VP4_DC_UNDEFINED = NB_VP4_DC_TYPES
138 };
139
140 static const uint8_t vp4_pred_block_type_map[8] = {
141 [MODE_INTER_NO_MV] = VP4_DC_INTER,
142 [MODE_INTRA] = VP4_DC_INTRA,
143 [MODE_INTER_PLUS_MV] = VP4_DC_INTER,
144 [MODE_INTER_LAST_MV] = VP4_DC_INTER,
145 [MODE_INTER_PRIOR_LAST] = VP4_DC_INTER,
146 [MODE_USING_GOLDEN] = VP4_DC_GOLDEN,
147 [MODE_GOLDEN_MV] = VP4_DC_GOLDEN,
148 [MODE_INTER_FOURMV] = VP4_DC_INTER,
149 };
150
151 typedef struct {
152 int dc;
153 int type;
154 } VP4Predictor;
155
156 #define MIN_DEQUANT_VAL 2
157
158 typedef struct Vp3DecodeContext {
159 AVCodecContext *avctx;
160 int theora, theora_tables, theora_header;
161 int version;
162 int width, height;
163 int chroma_x_shift, chroma_y_shift;
164 ThreadFrame golden_frame;
165 ThreadFrame last_frame;
166 ThreadFrame current_frame;
167 int keyframe;
168 uint8_t idct_permutation[64];
169 uint8_t idct_scantable[64];
170 HpelDSPContext hdsp;
171 VideoDSPContext vdsp;
172 VP3DSPContext vp3dsp;
173 DECLARE_ALIGNED(16, int16_t, block)[64];
174 int flipped_image;
175 int last_slice_end;
176 int skip_loop_filter;
177
178 int qps[3];
179 int nqps;
180 int last_qps[3];
181
182 int superblock_count;
183 int y_superblock_width;
184 int y_superblock_height;
185 int y_superblock_count;
186 int c_superblock_width;
187 int c_superblock_height;
188 int c_superblock_count;
189 int u_superblock_start;
190 int v_superblock_start;
191 unsigned char *superblock_coding;
192
193 int macroblock_count; /* y macroblock count */
194 int macroblock_width;
195 int macroblock_height;
196 int c_macroblock_count;
197 int c_macroblock_width;
198 int c_macroblock_height;
199 int yuv_macroblock_count; /* y+u+v macroblock count */
200
201 int fragment_count;
202 int fragment_width[2];
203 int fragment_height[2];
204
205 Vp3Fragment *all_fragments;
206 int fragment_start[3];
207 int data_offset[3];
208 uint8_t offset_x;
209 uint8_t offset_y;
210 int offset_x_warned;
211
212 int8_t (*motion_val[2])[2];
213
214 /* tables */
215 uint16_t coded_dc_scale_factor[2][64];
216 uint32_t coded_ac_scale_factor[64];
217 uint8_t base_matrix[384][64];
218 uint8_t qr_count[2][3];
219 uint8_t qr_size[2][3][64];
220 uint16_t qr_base[2][3][64];
221
222 /**
223 * This is a list of all tokens in bitstream order. Reordering takes place
224 * by pulling from each level during IDCT. As a consequence, IDCT must be
225 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
226 * otherwise. The 32 different tokens with up to 12 bits of extradata are
227 * collapsed into 3 types, packed as follows:
228 * (from the low to high bits)
229 *
230 * 2 bits: type (0,1,2)
231 * 0: EOB run, 14 bits for run length (12 needed)
232 * 1: zero run, 7 bits for run length
233 * 7 bits for the next coefficient (3 needed)
234 * 2: coefficient, 14 bits (11 needed)
235 *
236 * Coefficients are signed, so are packed in the highest bits for automatic
237 * sign extension.
238 */
239 int16_t *dct_tokens[3][64];
240 int16_t *dct_tokens_base;
241 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
242 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) * 512) + ((zero_run) << 2) + 1)
243 #define TOKEN_COEFF(coeff) (((coeff) * 4) + 2)
244
245 /**
246 * number of blocks that contain DCT coefficients at
247 * the given level or higher
248 */
249 int num_coded_frags[3][64];
250 int total_num_coded_frags;
251
252 /* this is a list of indexes into the all_fragments array indicating
253 * which of the fragments are coded */
254 int *coded_fragment_list[3];
255
256 int *kf_coded_fragment_list;
257 int *nkf_coded_fragment_list;
258 int num_kf_coded_fragment[3];
259
260 VLC dc_vlc[16];
261 VLC ac_vlc_1[16];
262 VLC ac_vlc_2[16];
263 VLC ac_vlc_3[16];
264 VLC ac_vlc_4[16];
265
266 VLC superblock_run_length_vlc; /* version < 2 */
267 VLC fragment_run_length_vlc; /* version < 2 */
268 VLC block_pattern_vlc[2]; /* version >= 2*/
269 VLC mode_code_vlc;
270 VLC motion_vector_vlc; /* version < 2 */
271 VLC vp4_mv_vlc[2][7]; /* version >=2 */
272
273 /* these arrays need to be on 16-byte boundaries since SSE2 operations
274 * index into them */
275 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
276
277 /* This table contains superblock_count * 16 entries. Each set of 16
278 * numbers corresponds to the fragment indexes 0..15 of the superblock.
279 * An entry will be -1 to indicate that no entry corresponds to that
280 * index. */
281 int *superblock_fragments;
282
283 /* This is an array that indicates how a particular macroblock
284 * is coded. */
285 unsigned char *macroblock_coding;
286
287 uint8_t *edge_emu_buffer;
288
289 /* Huffman decode */
290 int hti;
291 unsigned int hbits;
292 int entries;
293 int huff_code_size;
294 uint32_t huffman_table[80][32][2];
295
296 uint8_t filter_limit_values[64];
297 DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
298
299 VP4Predictor * dc_pred_row; /* dc_pred_row[y_superblock_width * 4] */
300 } Vp3DecodeContext;
301
302 /************************************************************************
303 * VP3 specific functions
304 ************************************************************************/
305
free_tables(AVCodecContext * avctx)306 static av_cold void free_tables(AVCodecContext *avctx)
307 {
308 Vp3DecodeContext *s = avctx->priv_data;
309
310 av_freep(&s->superblock_coding);
311 av_freep(&s->all_fragments);
312 av_freep(&s->nkf_coded_fragment_list);
313 av_freep(&s->kf_coded_fragment_list);
314 av_freep(&s->dct_tokens_base);
315 av_freep(&s->superblock_fragments);
316 av_freep(&s->macroblock_coding);
317 av_freep(&s->dc_pred_row);
318 av_freep(&s->motion_val[0]);
319 av_freep(&s->motion_val[1]);
320 }
321
vp3_decode_flush(AVCodecContext * avctx)322 static void vp3_decode_flush(AVCodecContext *avctx)
323 {
324 Vp3DecodeContext *s = avctx->priv_data;
325
326 if (s->golden_frame.f)
327 ff_thread_release_buffer(avctx, &s->golden_frame);
328 if (s->last_frame.f)
329 ff_thread_release_buffer(avctx, &s->last_frame);
330 if (s->current_frame.f)
331 ff_thread_release_buffer(avctx, &s->current_frame);
332 }
333
vp3_decode_end(AVCodecContext * avctx)334 static av_cold int vp3_decode_end(AVCodecContext *avctx)
335 {
336 Vp3DecodeContext *s = avctx->priv_data;
337 int i, j;
338
339 free_tables(avctx);
340 av_freep(&s->edge_emu_buffer);
341
342 s->theora_tables = 0;
343
344 /* release all frames */
345 vp3_decode_flush(avctx);
346 av_frame_free(&s->current_frame.f);
347 av_frame_free(&s->last_frame.f);
348 av_frame_free(&s->golden_frame.f);
349
350 for (i = 0; i < 16; i++) {
351 ff_free_vlc(&s->dc_vlc[i]);
352 ff_free_vlc(&s->ac_vlc_1[i]);
353 ff_free_vlc(&s->ac_vlc_2[i]);
354 ff_free_vlc(&s->ac_vlc_3[i]);
355 ff_free_vlc(&s->ac_vlc_4[i]);
356 }
357
358 ff_free_vlc(&s->superblock_run_length_vlc);
359 ff_free_vlc(&s->fragment_run_length_vlc);
360 ff_free_vlc(&s->mode_code_vlc);
361 ff_free_vlc(&s->motion_vector_vlc);
362
363 for (j = 0; j < 2; j++)
364 for (i = 0; i < 7; i++)
365 ff_free_vlc(&s->vp4_mv_vlc[j][i]);
366
367 for (i = 0; i < 2; i++)
368 ff_free_vlc(&s->block_pattern_vlc[i]);
369 return 0;
370 }
371
372 /**
373 * This function sets up all of the various blocks mappings:
374 * superblocks <-> fragments, macroblocks <-> fragments,
375 * superblocks <-> macroblocks
376 *
377 * @return 0 is successful; returns 1 if *anything* went wrong.
378 */
init_block_mapping(Vp3DecodeContext * s)379 static int init_block_mapping(Vp3DecodeContext *s)
380 {
381 int sb_x, sb_y, plane;
382 int x, y, i, j = 0;
383
384 for (plane = 0; plane < 3; plane++) {
385 int sb_width = plane ? s->c_superblock_width
386 : s->y_superblock_width;
387 int sb_height = plane ? s->c_superblock_height
388 : s->y_superblock_height;
389 int frag_width = s->fragment_width[!!plane];
390 int frag_height = s->fragment_height[!!plane];
391
392 for (sb_y = 0; sb_y < sb_height; sb_y++)
393 for (sb_x = 0; sb_x < sb_width; sb_x++)
394 for (i = 0; i < 16; i++) {
395 x = 4 * sb_x + hilbert_offset[i][0];
396 y = 4 * sb_y + hilbert_offset[i][1];
397
398 if (x < frag_width && y < frag_height)
399 s->superblock_fragments[j++] = s->fragment_start[plane] +
400 y * frag_width + x;
401 else
402 s->superblock_fragments[j++] = -1;
403 }
404 }
405
406 return 0; /* successful path out */
407 }
408
409 /*
410 * This function sets up the dequantization tables used for a particular
411 * frame.
412 */
init_dequantizer(Vp3DecodeContext * s,int qpi)413 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
414 {
415 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
416 int i, plane, inter, qri, bmi, bmj, qistart;
417
418 for (inter = 0; inter < 2; inter++) {
419 for (plane = 0; plane < 3; plane++) {
420 int dc_scale_factor = s->coded_dc_scale_factor[!!plane][s->qps[qpi]];
421 int sum = 0;
422 for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
423 sum += s->qr_size[inter][plane][qri];
424 if (s->qps[qpi] <= sum)
425 break;
426 }
427 qistart = sum - s->qr_size[inter][plane][qri];
428 bmi = s->qr_base[inter][plane][qri];
429 bmj = s->qr_base[inter][plane][qri + 1];
430 for (i = 0; i < 64; i++) {
431 int coeff = (2 * (sum - s->qps[qpi]) * s->base_matrix[bmi][i] -
432 2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
433 s->qr_size[inter][plane][qri]) /
434 (2 * s->qr_size[inter][plane][qri]);
435
436 int qmin = 8 << (inter + !i);
437 int qscale = i ? ac_scale_factor : dc_scale_factor;
438 int qbias = (1 + inter) * 3;
439 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
440 (i == 0 || s->version < 2) ? av_clip((qscale * coeff) / 100 * 4, qmin, 4096)
441 : (qscale * (coeff - qbias) / 100 + qbias) * 4;
442 }
443 /* all DC coefficients use the same quant so as not to interfere
444 * with DC prediction */
445 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
446 }
447 }
448 }
449
450 /*
451 * This function initializes the loop filter boundary limits if the frame's
452 * quality index is different from the previous frame's.
453 *
454 * The filter_limit_values may not be larger than 127.
455 */
init_loop_filter(Vp3DecodeContext * s)456 static void init_loop_filter(Vp3DecodeContext *s)
457 {
458 ff_vp3dsp_set_bounding_values(s->bounding_values_array, s->filter_limit_values[s->qps[0]]);
459 }
460
461 /*
462 * This function unpacks all of the superblock/macroblock/fragment coding
463 * information from the bitstream.
464 */
unpack_superblocks(Vp3DecodeContext * s,GetBitContext * gb)465 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
466 {
467 int superblock_starts[3] = {
468 0, s->u_superblock_start, s->v_superblock_start
469 };
470 int bit = 0;
471 int current_superblock = 0;
472 int current_run = 0;
473 int num_partial_superblocks = 0;
474
475 int i, j;
476 int current_fragment;
477 int plane;
478 int plane0_num_coded_frags = 0;
479
480 if (s->keyframe) {
481 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
482 } else {
483 /* unpack the list of partially-coded superblocks */
484 bit = get_bits1(gb) ^ 1;
485 current_run = 0;
486
487 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
488 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
489 bit = get_bits1(gb);
490 else
491 bit ^= 1;
492
493 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
494 6, 2) + 1;
495 if (current_run == 34)
496 current_run += get_bits(gb, 12);
497
498 if (current_run > s->superblock_count - current_superblock) {
499 av_log(s->avctx, AV_LOG_ERROR,
500 "Invalid partially coded superblock run length\n");
501 return -1;
502 }
503
504 memset(s->superblock_coding + current_superblock, bit, current_run);
505
506 current_superblock += current_run;
507 if (bit)
508 num_partial_superblocks += current_run;
509 }
510
511 /* unpack the list of fully coded superblocks if any of the blocks were
512 * not marked as partially coded in the previous step */
513 if (num_partial_superblocks < s->superblock_count) {
514 int superblocks_decoded = 0;
515
516 current_superblock = 0;
517 bit = get_bits1(gb) ^ 1;
518 current_run = 0;
519
520 while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
521 get_bits_left(gb) > 0) {
522 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
523 bit = get_bits1(gb);
524 else
525 bit ^= 1;
526
527 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
528 6, 2) + 1;
529 if (current_run == 34)
530 current_run += get_bits(gb, 12);
531
532 for (j = 0; j < current_run; current_superblock++) {
533 if (current_superblock >= s->superblock_count) {
534 av_log(s->avctx, AV_LOG_ERROR,
535 "Invalid fully coded superblock run length\n");
536 return -1;
537 }
538
539 /* skip any superblocks already marked as partially coded */
540 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
541 s->superblock_coding[current_superblock] = 2 * bit;
542 j++;
543 }
544 }
545 superblocks_decoded += current_run;
546 }
547 }
548
549 /* if there were partial blocks, initialize bitstream for
550 * unpacking fragment codings */
551 if (num_partial_superblocks) {
552 current_run = 0;
553 bit = get_bits1(gb);
554 /* toggle the bit because as soon as the first run length is
555 * fetched the bit will be toggled again */
556 bit ^= 1;
557 }
558 }
559
560 /* figure out which fragments are coded; iterate through each
561 * superblock (all planes) */
562 s->total_num_coded_frags = 0;
563 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
564
565 s->coded_fragment_list[0] = s->keyframe ? s->kf_coded_fragment_list
566 : s->nkf_coded_fragment_list;
567
568 for (plane = 0; plane < 3; plane++) {
569 int sb_start = superblock_starts[plane];
570 int sb_end = sb_start + (plane ? s->c_superblock_count
571 : s->y_superblock_count);
572 int num_coded_frags = 0;
573
574 if (s->keyframe) {
575 if (s->num_kf_coded_fragment[plane] == -1) {
576 for (i = sb_start; i < sb_end; i++) {
577 /* iterate through all 16 fragments in a superblock */
578 for (j = 0; j < 16; j++) {
579 /* if the fragment is in bounds, check its coding status */
580 current_fragment = s->superblock_fragments[i * 16 + j];
581 if (current_fragment != -1) {
582 s->coded_fragment_list[plane][num_coded_frags++] =
583 current_fragment;
584 }
585 }
586 }
587 s->num_kf_coded_fragment[plane] = num_coded_frags;
588 } else
589 num_coded_frags = s->num_kf_coded_fragment[plane];
590 } else {
591 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
592 if (get_bits_left(gb) < plane0_num_coded_frags >> 2) {
593 return AVERROR_INVALIDDATA;
594 }
595 /* iterate through all 16 fragments in a superblock */
596 for (j = 0; j < 16; j++) {
597 /* if the fragment is in bounds, check its coding status */
598 current_fragment = s->superblock_fragments[i * 16 + j];
599 if (current_fragment != -1) {
600 int coded = s->superblock_coding[i];
601
602 if (coded == SB_PARTIALLY_CODED) {
603 /* fragment may or may not be coded; this is the case
604 * that cares about the fragment coding runs */
605 if (current_run-- == 0) {
606 bit ^= 1;
607 current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2);
608 }
609 coded = bit;
610 }
611
612 if (coded) {
613 /* default mode; actual mode will be decoded in
614 * the next phase */
615 s->all_fragments[current_fragment].coding_method =
616 MODE_INTER_NO_MV;
617 s->coded_fragment_list[plane][num_coded_frags++] =
618 current_fragment;
619 } else {
620 /* not coded; copy this fragment from the prior frame */
621 s->all_fragments[current_fragment].coding_method =
622 MODE_COPY;
623 }
624 }
625 }
626 }
627 }
628 if (!plane)
629 plane0_num_coded_frags = num_coded_frags;
630 s->total_num_coded_frags += num_coded_frags;
631 for (i = 0; i < 64; i++)
632 s->num_coded_frags[plane][i] = num_coded_frags;
633 if (plane < 2)
634 s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] +
635 num_coded_frags;
636 }
637 return 0;
638 }
639
640 #define BLOCK_X (2 * mb_x + (k & 1))
641 #define BLOCK_Y (2 * mb_y + (k >> 1))
642
643 #if CONFIG_VP4_DECODER
644 /**
645 * @return number of blocks, or > yuv_macroblock_count on error.
646 * return value is always >= 1.
647 */
vp4_get_mb_count(Vp3DecodeContext * s,GetBitContext * gb)648 static int vp4_get_mb_count(Vp3DecodeContext *s, GetBitContext *gb)
649 {
650 int v = 1;
651 int bits;
652 while ((bits = show_bits(gb, 9)) == 0x1ff) {
653 skip_bits(gb, 9);
654 v += 256;
655 if (v > s->yuv_macroblock_count) {
656 av_log(s->avctx, AV_LOG_ERROR, "Invalid run length\n");
657 return v;
658 }
659 }
660 #define body(n) { \
661 skip_bits(gb, 2 + n); \
662 v += (1 << n) + get_bits(gb, n); }
663 #define thresh(n) (0x200 - (0x80 >> n))
664 #define else_if(n) else if (bits < thresh(n)) body(n)
665 if (bits < 0x100) {
666 skip_bits(gb, 1);
667 } else if (bits < thresh(0)) {
668 skip_bits(gb, 2);
669 v += 1;
670 }
671 else_if(1)
672 else_if(2)
673 else_if(3)
674 else_if(4)
675 else_if(5)
676 else_if(6)
677 else body(7)
678 #undef body
679 #undef thresh
680 #undef else_if
681 return v;
682 }
683
vp4_get_block_pattern(Vp3DecodeContext * s,GetBitContext * gb,int * next_block_pattern_table)684 static int vp4_get_block_pattern(Vp3DecodeContext *s, GetBitContext *gb, int *next_block_pattern_table)
685 {
686 int v = get_vlc2(gb, s->block_pattern_vlc[*next_block_pattern_table].table, 3, 2);
687 if (v == -1) {
688 av_log(s->avctx, AV_LOG_ERROR, "Invalid block pattern\n");
689 *next_block_pattern_table = 0;
690 return 0;
691 }
692 *next_block_pattern_table = vp4_block_pattern_table_selector[v];
693 return v + 1;
694 }
695
vp4_unpack_macroblocks(Vp3DecodeContext * s,GetBitContext * gb)696 static int vp4_unpack_macroblocks(Vp3DecodeContext *s, GetBitContext *gb)
697 {
698 int plane, i, j, k, fragment;
699 int next_block_pattern_table;
700 int bit, current_run, has_partial;
701
702 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
703
704 if (s->keyframe)
705 return 0;
706
707 has_partial = 0;
708 bit = get_bits1(gb);
709 for (i = 0; i < s->yuv_macroblock_count; i += current_run) {
710 if (get_bits_left(gb) <= 0)
711 return AVERROR_INVALIDDATA;
712 current_run = vp4_get_mb_count(s, gb);
713 if (current_run > s->yuv_macroblock_count - i)
714 return -1;
715 memset(s->superblock_coding + i, 2 * bit, current_run);
716 bit ^= 1;
717 has_partial |= bit;
718 }
719
720 if (has_partial) {
721 if (get_bits_left(gb) <= 0)
722 return AVERROR_INVALIDDATA;
723 bit = get_bits1(gb);
724 current_run = vp4_get_mb_count(s, gb);
725 for (i = 0; i < s->yuv_macroblock_count; i++) {
726 if (!s->superblock_coding[i]) {
727 if (!current_run) {
728 bit ^= 1;
729 current_run = vp4_get_mb_count(s, gb);
730 }
731 s->superblock_coding[i] = bit;
732 current_run--;
733 }
734 }
735 if (current_run) /* handle situation when vp4_get_mb_count() fails */
736 return -1;
737 }
738
739 next_block_pattern_table = 0;
740 i = 0;
741 for (plane = 0; plane < 3; plane++) {
742 int sb_x, sb_y;
743 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
744 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
745 int mb_width = plane ? s->c_macroblock_width : s->macroblock_width;
746 int mb_height = plane ? s->c_macroblock_height : s->macroblock_height;
747 int fragment_width = s->fragment_width[!!plane];
748 int fragment_height = s->fragment_height[!!plane];
749
750 for (sb_y = 0; sb_y < sb_height; sb_y++) {
751 for (sb_x = 0; sb_x < sb_width; sb_x++) {
752 for (j = 0; j < 4; j++) {
753 int mb_x = 2 * sb_x + (j >> 1);
754 int mb_y = 2 * sb_y + (j >> 1) ^ (j & 1);
755 int mb_coded, pattern, coded;
756
757 if (mb_x >= mb_width || mb_y >= mb_height)
758 continue;
759
760 mb_coded = s->superblock_coding[i++];
761
762 if (mb_coded == SB_FULLY_CODED)
763 pattern = 0xF;
764 else if (mb_coded == SB_PARTIALLY_CODED)
765 pattern = vp4_get_block_pattern(s, gb, &next_block_pattern_table);
766 else
767 pattern = 0;
768
769 for (k = 0; k < 4; k++) {
770 if (BLOCK_X >= fragment_width || BLOCK_Y >= fragment_height)
771 continue;
772 fragment = s->fragment_start[plane] + BLOCK_Y * fragment_width + BLOCK_X;
773 coded = pattern & (8 >> k);
774 /* MODE_INTER_NO_MV is the default for coded fragments.
775 the actual method is decoded in the next phase. */
776 s->all_fragments[fragment].coding_method = coded ? MODE_INTER_NO_MV : MODE_COPY;
777 }
778 }
779 }
780 }
781 }
782 return 0;
783 }
784 #endif
785
786 /*
787 * This function unpacks all the coding mode data for individual macroblocks
788 * from the bitstream.
789 */
unpack_modes(Vp3DecodeContext * s,GetBitContext * gb)790 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
791 {
792 int i, j, k, sb_x, sb_y;
793 int scheme;
794 int current_macroblock;
795 int current_fragment;
796 int coding_mode;
797 int custom_mode_alphabet[CODING_MODE_COUNT];
798 const int *alphabet;
799 Vp3Fragment *frag;
800
801 if (s->keyframe) {
802 for (i = 0; i < s->fragment_count; i++)
803 s->all_fragments[i].coding_method = MODE_INTRA;
804 } else {
805 /* fetch the mode coding scheme for this frame */
806 scheme = get_bits(gb, 3);
807
808 /* is it a custom coding scheme? */
809 if (scheme == 0) {
810 for (i = 0; i < 8; i++)
811 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
812 for (i = 0; i < 8; i++)
813 custom_mode_alphabet[get_bits(gb, 3)] = i;
814 alphabet = custom_mode_alphabet;
815 } else
816 alphabet = ModeAlphabet[scheme - 1];
817
818 /* iterate through all of the macroblocks that contain 1 or more
819 * coded fragments */
820 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
821 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
822 if (get_bits_left(gb) <= 0)
823 return -1;
824
825 for (j = 0; j < 4; j++) {
826 int mb_x = 2 * sb_x + (j >> 1);
827 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
828 current_macroblock = mb_y * s->macroblock_width + mb_x;
829
830 if (mb_x >= s->macroblock_width ||
831 mb_y >= s->macroblock_height)
832 continue;
833
834 /* coding modes are only stored if the macroblock has
835 * at least one luma block coded, otherwise it must be
836 * INTER_NO_MV */
837 for (k = 0; k < 4; k++) {
838 current_fragment = BLOCK_Y *
839 s->fragment_width[0] + BLOCK_X;
840 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
841 break;
842 }
843 if (k == 4) {
844 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
845 continue;
846 }
847
848 /* mode 7 means get 3 bits for each coding mode */
849 if (scheme == 7)
850 coding_mode = get_bits(gb, 3);
851 else
852 coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
853
854 s->macroblock_coding[current_macroblock] = coding_mode;
855 for (k = 0; k < 4; k++) {
856 frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
857 if (frag->coding_method != MODE_COPY)
858 frag->coding_method = coding_mode;
859 }
860
861 #define SET_CHROMA_MODES \
862 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
863 frag[s->fragment_start[1]].coding_method = coding_mode; \
864 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
865 frag[s->fragment_start[2]].coding_method = coding_mode;
866
867 if (s->chroma_y_shift) {
868 frag = s->all_fragments + mb_y *
869 s->fragment_width[1] + mb_x;
870 SET_CHROMA_MODES
871 } else if (s->chroma_x_shift) {
872 frag = s->all_fragments +
873 2 * mb_y * s->fragment_width[1] + mb_x;
874 for (k = 0; k < 2; k++) {
875 SET_CHROMA_MODES
876 frag += s->fragment_width[1];
877 }
878 } else {
879 for (k = 0; k < 4; k++) {
880 frag = s->all_fragments +
881 BLOCK_Y * s->fragment_width[1] + BLOCK_X;
882 SET_CHROMA_MODES
883 }
884 }
885 }
886 }
887 }
888 }
889
890 return 0;
891 }
892
vp4_get_mv(Vp3DecodeContext * s,GetBitContext * gb,int axis,int last_motion)893 static int vp4_get_mv(Vp3DecodeContext *s, GetBitContext *gb, int axis, int last_motion)
894 {
895 int v = get_vlc2(gb, s->vp4_mv_vlc[axis][vp4_mv_table_selector[FFABS(last_motion)]].table, 6, 2) - 31;
896 return last_motion < 0 ? -v : v;
897 }
898
899 /*
900 * This function unpacks all the motion vectors for the individual
901 * macroblocks from the bitstream.
902 */
unpack_vectors(Vp3DecodeContext * s,GetBitContext * gb)903 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
904 {
905 int j, k, sb_x, sb_y;
906 int coding_mode;
907 int motion_x[4];
908 int motion_y[4];
909 int last_motion_x = 0;
910 int last_motion_y = 0;
911 int prior_last_motion_x = 0;
912 int prior_last_motion_y = 0;
913 int last_gold_motion_x = 0;
914 int last_gold_motion_y = 0;
915 int current_macroblock;
916 int current_fragment;
917 int frag;
918
919 if (s->keyframe)
920 return 0;
921
922 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme; 2 is VP4 code scheme */
923 coding_mode = s->version < 2 ? get_bits1(gb) : 2;
924
925 /* iterate through all of the macroblocks that contain 1 or more
926 * coded fragments */
927 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
928 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
929 if (get_bits_left(gb) <= 0)
930 return -1;
931
932 for (j = 0; j < 4; j++) {
933 int mb_x = 2 * sb_x + (j >> 1);
934 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
935 current_macroblock = mb_y * s->macroblock_width + mb_x;
936
937 if (mb_x >= s->macroblock_width ||
938 mb_y >= s->macroblock_height ||
939 s->macroblock_coding[current_macroblock] == MODE_COPY)
940 continue;
941
942 switch (s->macroblock_coding[current_macroblock]) {
943 case MODE_GOLDEN_MV:
944 if (coding_mode == 2) { /* VP4 */
945 last_gold_motion_x = motion_x[0] = vp4_get_mv(s, gb, 0, last_gold_motion_x);
946 last_gold_motion_y = motion_y[0] = vp4_get_mv(s, gb, 1, last_gold_motion_y);
947 break;
948 } /* otherwise fall through */
949 case MODE_INTER_PLUS_MV:
950 /* all 6 fragments use the same motion vector */
951 if (coding_mode == 0) {
952 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
953 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
954 } else if (coding_mode == 1) {
955 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
956 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
957 } else { /* VP4 */
958 motion_x[0] = vp4_get_mv(s, gb, 0, last_motion_x);
959 motion_y[0] = vp4_get_mv(s, gb, 1, last_motion_y);
960 }
961
962 /* vector maintenance, only on MODE_INTER_PLUS_MV */
963 if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
964 prior_last_motion_x = last_motion_x;
965 prior_last_motion_y = last_motion_y;
966 last_motion_x = motion_x[0];
967 last_motion_y = motion_y[0];
968 }
969 break;
970
971 case MODE_INTER_FOURMV:
972 /* vector maintenance */
973 prior_last_motion_x = last_motion_x;
974 prior_last_motion_y = last_motion_y;
975
976 /* fetch 4 vectors from the bitstream, one for each
977 * Y fragment, then average for the C fragment vectors */
978 for (k = 0; k < 4; k++) {
979 current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
980 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
981 if (coding_mode == 0) {
982 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
983 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
984 } else if (coding_mode == 1) {
985 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
986 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
987 } else { /* VP4 */
988 motion_x[k] = vp4_get_mv(s, gb, 0, prior_last_motion_x);
989 motion_y[k] = vp4_get_mv(s, gb, 1, prior_last_motion_y);
990 }
991 last_motion_x = motion_x[k];
992 last_motion_y = motion_y[k];
993 } else {
994 motion_x[k] = 0;
995 motion_y[k] = 0;
996 }
997 }
998 break;
999
1000 case MODE_INTER_LAST_MV:
1001 /* all 6 fragments use the last motion vector */
1002 motion_x[0] = last_motion_x;
1003 motion_y[0] = last_motion_y;
1004
1005 /* no vector maintenance (last vector remains the
1006 * last vector) */
1007 break;
1008
1009 case MODE_INTER_PRIOR_LAST:
1010 /* all 6 fragments use the motion vector prior to the
1011 * last motion vector */
1012 motion_x[0] = prior_last_motion_x;
1013 motion_y[0] = prior_last_motion_y;
1014
1015 /* vector maintenance */
1016 prior_last_motion_x = last_motion_x;
1017 prior_last_motion_y = last_motion_y;
1018 last_motion_x = motion_x[0];
1019 last_motion_y = motion_y[0];
1020 break;
1021
1022 default:
1023 /* covers intra, inter without MV, golden without MV */
1024 motion_x[0] = 0;
1025 motion_y[0] = 0;
1026
1027 /* no vector maintenance */
1028 break;
1029 }
1030
1031 /* assign the motion vectors to the correct fragments */
1032 for (k = 0; k < 4; k++) {
1033 current_fragment =
1034 BLOCK_Y * s->fragment_width[0] + BLOCK_X;
1035 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1036 s->motion_val[0][current_fragment][0] = motion_x[k];
1037 s->motion_val[0][current_fragment][1] = motion_y[k];
1038 } else {
1039 s->motion_val[0][current_fragment][0] = motion_x[0];
1040 s->motion_val[0][current_fragment][1] = motion_y[0];
1041 }
1042 }
1043
1044 if (s->chroma_y_shift) {
1045 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1046 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
1047 motion_x[2] + motion_x[3], 2);
1048 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
1049 motion_y[2] + motion_y[3], 2);
1050 }
1051 if (s->version <= 2) {
1052 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1053 motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
1054 }
1055 frag = mb_y * s->fragment_width[1] + mb_x;
1056 s->motion_val[1][frag][0] = motion_x[0];
1057 s->motion_val[1][frag][1] = motion_y[0];
1058 } else if (s->chroma_x_shift) {
1059 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1060 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
1061 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
1062 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
1063 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
1064 } else {
1065 motion_x[1] = motion_x[0];
1066 motion_y[1] = motion_y[0];
1067 }
1068 if (s->version <= 2) {
1069 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
1070 motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
1071 }
1072 frag = 2 * mb_y * s->fragment_width[1] + mb_x;
1073 for (k = 0; k < 2; k++) {
1074 s->motion_val[1][frag][0] = motion_x[k];
1075 s->motion_val[1][frag][1] = motion_y[k];
1076 frag += s->fragment_width[1];
1077 }
1078 } else {
1079 for (k = 0; k < 4; k++) {
1080 frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
1081 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
1082 s->motion_val[1][frag][0] = motion_x[k];
1083 s->motion_val[1][frag][1] = motion_y[k];
1084 } else {
1085 s->motion_val[1][frag][0] = motion_x[0];
1086 s->motion_val[1][frag][1] = motion_y[0];
1087 }
1088 }
1089 }
1090 }
1091 }
1092 }
1093
1094 return 0;
1095 }
1096
unpack_block_qpis(Vp3DecodeContext * s,GetBitContext * gb)1097 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
1098 {
1099 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
1100 int num_blocks = s->total_num_coded_frags;
1101
1102 for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
1103 i = blocks_decoded = num_blocks_at_qpi = 0;
1104
1105 bit = get_bits1(gb) ^ 1;
1106 run_length = 0;
1107
1108 do {
1109 if (run_length == MAXIMUM_LONG_BIT_RUN)
1110 bit = get_bits1(gb);
1111 else
1112 bit ^= 1;
1113
1114 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
1115 if (run_length == 34)
1116 run_length += get_bits(gb, 12);
1117 blocks_decoded += run_length;
1118
1119 if (!bit)
1120 num_blocks_at_qpi += run_length;
1121
1122 for (j = 0; j < run_length; i++) {
1123 if (i >= s->total_num_coded_frags)
1124 return -1;
1125
1126 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
1127 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
1128 j++;
1129 }
1130 }
1131 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
1132
1133 num_blocks -= num_blocks_at_qpi;
1134 }
1135
1136 return 0;
1137 }
1138
get_eob_run(GetBitContext * gb,int token)1139 static inline int get_eob_run(GetBitContext *gb, int token)
1140 {
1141 int v = eob_run_table[token].base;
1142 if (eob_run_table[token].bits)
1143 v += get_bits(gb, eob_run_table[token].bits);
1144 return v;
1145 }
1146
get_coeff(GetBitContext * gb,int token,int16_t * coeff)1147 static inline int get_coeff(GetBitContext *gb, int token, int16_t *coeff)
1148 {
1149 int bits_to_get, zero_run;
1150
1151 bits_to_get = coeff_get_bits[token];
1152 if (bits_to_get)
1153 bits_to_get = get_bits(gb, bits_to_get);
1154 *coeff = coeff_tables[token][bits_to_get];
1155
1156 zero_run = zero_run_base[token];
1157 if (zero_run_get_bits[token])
1158 zero_run += get_bits(gb, zero_run_get_bits[token]);
1159
1160 return zero_run;
1161 }
1162
1163 /*
1164 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1165 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1166 * data. This function unpacks all the VLCs for either the Y plane or both
1167 * C planes, and is called for DC coefficients or different AC coefficient
1168 * levels (since different coefficient types require different VLC tables.
1169 *
1170 * This function returns a residual eob run. E.g, if a particular token gave
1171 * instructions to EOB the next 5 fragments and there were only 2 fragments
1172 * left in the current fragment range, 3 would be returned so that it could
1173 * be passed into the next call to this same function.
1174 */
unpack_vlcs(Vp3DecodeContext * s,GetBitContext * gb,VLC * table,int coeff_index,int plane,int eob_run)1175 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1176 VLC *table, int coeff_index,
1177 int plane,
1178 int eob_run)
1179 {
1180 int i, j = 0;
1181 int token;
1182 int zero_run = 0;
1183 int16_t coeff = 0;
1184 int blocks_ended;
1185 int coeff_i = 0;
1186 int num_coeffs = s->num_coded_frags[plane][coeff_index];
1187 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
1188
1189 /* local references to structure members to avoid repeated dereferences */
1190 int *coded_fragment_list = s->coded_fragment_list[plane];
1191 Vp3Fragment *all_fragments = s->all_fragments;
1192 VLC_TYPE(*vlc_table)[2] = table->table;
1193
1194 if (num_coeffs < 0) {
1195 av_log(s->avctx, AV_LOG_ERROR,
1196 "Invalid number of coefficients at level %d\n", coeff_index);
1197 return AVERROR_INVALIDDATA;
1198 }
1199
1200 if (eob_run > num_coeffs) {
1201 coeff_i =
1202 blocks_ended = num_coeffs;
1203 eob_run -= num_coeffs;
1204 } else {
1205 coeff_i =
1206 blocks_ended = eob_run;
1207 eob_run = 0;
1208 }
1209
1210 // insert fake EOB token to cover the split between planes or zzi
1211 if (blocks_ended)
1212 dct_tokens[j++] = blocks_ended << 2;
1213
1214 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
1215 /* decode a VLC into a token */
1216 token = get_vlc2(gb, vlc_table, 11, 3);
1217 /* use the token to get a zero run, a coefficient, and an eob run */
1218 if ((unsigned) token <= 6U) {
1219 eob_run = get_eob_run(gb, token);
1220 if (!eob_run)
1221 eob_run = INT_MAX;
1222
1223 // record only the number of blocks ended in this plane,
1224 // any spill will be recorded in the next plane.
1225 if (eob_run > num_coeffs - coeff_i) {
1226 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
1227 blocks_ended += num_coeffs - coeff_i;
1228 eob_run -= num_coeffs - coeff_i;
1229 coeff_i = num_coeffs;
1230 } else {
1231 dct_tokens[j++] = TOKEN_EOB(eob_run);
1232 blocks_ended += eob_run;
1233 coeff_i += eob_run;
1234 eob_run = 0;
1235 }
1236 } else if (token >= 0) {
1237 zero_run = get_coeff(gb, token, &coeff);
1238
1239 if (zero_run) {
1240 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
1241 } else {
1242 // Save DC into the fragment structure. DC prediction is
1243 // done in raster order, so the actual DC can't be in with
1244 // other tokens. We still need the token in dct_tokens[]
1245 // however, or else the structure collapses on itself.
1246 if (!coeff_index)
1247 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
1248
1249 dct_tokens[j++] = TOKEN_COEFF(coeff);
1250 }
1251
1252 if (coeff_index + zero_run > 64) {
1253 av_log(s->avctx, AV_LOG_DEBUG,
1254 "Invalid zero run of %d with %d coeffs left\n",
1255 zero_run, 64 - coeff_index);
1256 zero_run = 64 - coeff_index;
1257 }
1258
1259 // zero runs code multiple coefficients,
1260 // so don't try to decode coeffs for those higher levels
1261 for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1262 s->num_coded_frags[plane][i]--;
1263 coeff_i++;
1264 } else {
1265 av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1266 return -1;
1267 }
1268 }
1269
1270 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1271 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1272
1273 // decrement the number of blocks that have higher coefficients for each
1274 // EOB run at this level
1275 if (blocks_ended)
1276 for (i = coeff_index + 1; i < 64; i++)
1277 s->num_coded_frags[plane][i] -= blocks_ended;
1278
1279 // setup the next buffer
1280 if (plane < 2)
1281 s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1282 else if (coeff_index < 63)
1283 s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1284
1285 return eob_run;
1286 }
1287
1288 static void reverse_dc_prediction(Vp3DecodeContext *s,
1289 int first_fragment,
1290 int fragment_width,
1291 int fragment_height);
1292 /*
1293 * This function unpacks all of the DCT coefficient data from the
1294 * bitstream.
1295 */
unpack_dct_coeffs(Vp3DecodeContext * s,GetBitContext * gb)1296 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1297 {
1298 int i;
1299 int dc_y_table;
1300 int dc_c_table;
1301 int ac_y_table;
1302 int ac_c_table;
1303 int residual_eob_run = 0;
1304 VLC *y_tables[64];
1305 VLC *c_tables[64];
1306
1307 s->dct_tokens[0][0] = s->dct_tokens_base;
1308
1309 if (get_bits_left(gb) < 16)
1310 return AVERROR_INVALIDDATA;
1311
1312 /* fetch the DC table indexes */
1313 dc_y_table = get_bits(gb, 4);
1314 dc_c_table = get_bits(gb, 4);
1315
1316 /* unpack the Y plane DC coefficients */
1317 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1318 0, residual_eob_run);
1319 if (residual_eob_run < 0)
1320 return residual_eob_run;
1321 if (get_bits_left(gb) < 8)
1322 return AVERROR_INVALIDDATA;
1323
1324 /* reverse prediction of the Y-plane DC coefficients */
1325 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1326
1327 /* unpack the C plane DC coefficients */
1328 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1329 1, residual_eob_run);
1330 if (residual_eob_run < 0)
1331 return residual_eob_run;
1332 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1333 2, residual_eob_run);
1334 if (residual_eob_run < 0)
1335 return residual_eob_run;
1336
1337 /* reverse prediction of the C-plane DC coefficients */
1338 if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1339 reverse_dc_prediction(s, s->fragment_start[1],
1340 s->fragment_width[1], s->fragment_height[1]);
1341 reverse_dc_prediction(s, s->fragment_start[2],
1342 s->fragment_width[1], s->fragment_height[1]);
1343 }
1344
1345 if (get_bits_left(gb) < 8)
1346 return AVERROR_INVALIDDATA;
1347 /* fetch the AC table indexes */
1348 ac_y_table = get_bits(gb, 4);
1349 ac_c_table = get_bits(gb, 4);
1350
1351 /* build tables of AC VLC tables */
1352 for (i = 1; i <= 5; i++) {
1353 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1354 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1355 }
1356 for (i = 6; i <= 14; i++) {
1357 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1358 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1359 }
1360 for (i = 15; i <= 27; i++) {
1361 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1362 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1363 }
1364 for (i = 28; i <= 63; i++) {
1365 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1366 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1367 }
1368
1369 /* decode all AC coefficients */
1370 for (i = 1; i <= 63; i++) {
1371 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1372 0, residual_eob_run);
1373 if (residual_eob_run < 0)
1374 return residual_eob_run;
1375
1376 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1377 1, residual_eob_run);
1378 if (residual_eob_run < 0)
1379 return residual_eob_run;
1380 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1381 2, residual_eob_run);
1382 if (residual_eob_run < 0)
1383 return residual_eob_run;
1384 }
1385
1386 return 0;
1387 }
1388
1389 #if CONFIG_VP4_DECODER
1390 /**
1391 * eob_tracker[] is instead of TOKEN_EOB(value)
1392 * a dummy TOKEN_EOB(0) value is used to make vp3_dequant work
1393 *
1394 * @return < 0 on error
1395 */
vp4_unpack_vlcs(Vp3DecodeContext * s,GetBitContext * gb,VLC * vlc_tables[64],int plane,int eob_tracker[64],int fragment)1396 static int vp4_unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1397 VLC *vlc_tables[64],
1398 int plane, int eob_tracker[64], int fragment)
1399 {
1400 int token;
1401 int zero_run = 0;
1402 int16_t coeff = 0;
1403 int coeff_i = 0;
1404 int eob_run;
1405
1406 while (!eob_tracker[coeff_i]) {
1407 if (get_bits_left(gb) < 1)
1408 return AVERROR_INVALIDDATA;
1409
1410 token = get_vlc2(gb, vlc_tables[coeff_i]->table, 11, 3);
1411
1412 /* use the token to get a zero run, a coefficient, and an eob run */
1413 if ((unsigned) token <= 6U) {
1414 eob_run = get_eob_run(gb, token);
1415 *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1416 eob_tracker[coeff_i] = eob_run - 1;
1417 return 0;
1418 } else if (token >= 0) {
1419 zero_run = get_coeff(gb, token, &coeff);
1420
1421 if (zero_run) {
1422 if (coeff_i + zero_run > 64) {
1423 av_log(s->avctx, AV_LOG_DEBUG,
1424 "Invalid zero run of %d with %d coeffs left\n",
1425 zero_run, 64 - coeff_i);
1426 zero_run = 64 - coeff_i;
1427 }
1428 *s->dct_tokens[plane][coeff_i]++ = TOKEN_ZERO_RUN(coeff, zero_run);
1429 coeff_i += zero_run;
1430 } else {
1431 if (!coeff_i)
1432 s->all_fragments[fragment].dc = coeff;
1433
1434 *s->dct_tokens[plane][coeff_i]++ = TOKEN_COEFF(coeff);
1435 }
1436 coeff_i++;
1437 if (coeff_i >= 64) /* > 64 occurs when there is a zero_run overflow */
1438 return 0; /* stop */
1439 } else {
1440 av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1441 return -1;
1442 }
1443 }
1444 *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
1445 eob_tracker[coeff_i]--;
1446 return 0;
1447 }
1448
vp4_dc_predictor_reset(VP4Predictor * p)1449 static void vp4_dc_predictor_reset(VP4Predictor *p)
1450 {
1451 p->dc = 0;
1452 p->type = VP4_DC_UNDEFINED;
1453 }
1454
vp4_dc_pred_before(const Vp3DecodeContext * s,VP4Predictor dc_pred[6][6],int sb_x)1455 static void vp4_dc_pred_before(const Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1456 {
1457 int i, j;
1458
1459 for (i = 0; i < 4; i++)
1460 dc_pred[0][i + 1] = s->dc_pred_row[sb_x * 4 + i];
1461
1462 for (j = 1; j < 5; j++)
1463 for (i = 0; i < 4; i++)
1464 vp4_dc_predictor_reset(&dc_pred[j][i + 1]);
1465 }
1466
vp4_dc_pred_after(Vp3DecodeContext * s,VP4Predictor dc_pred[6][6],int sb_x)1467 static void vp4_dc_pred_after(Vp3DecodeContext *s, VP4Predictor dc_pred[6][6], int sb_x)
1468 {
1469 int i;
1470
1471 for (i = 0; i < 4; i++)
1472 s->dc_pred_row[sb_x * 4 + i] = dc_pred[4][i + 1];
1473
1474 for (i = 1; i < 5; i++)
1475 dc_pred[i][0] = dc_pred[i][4];
1476 }
1477
1478 /* 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)1479 static int vp4_dc_pred(const Vp3DecodeContext *s, const VP4Predictor * dc_pred, const int * last_dc, int type, int plane)
1480 {
1481 int count = 0;
1482 int dc = 0;
1483
1484 if (dc_pred[-6].type == type) {
1485 dc += dc_pred[-6].dc;
1486 count++;
1487 }
1488
1489 if (dc_pred[6].type == type) {
1490 dc += dc_pred[6].dc;
1491 count++;
1492 }
1493
1494 if (count != 2 && dc_pred[-1].type == type) {
1495 dc += dc_pred[-1].dc;
1496 count++;
1497 }
1498
1499 if (count != 2 && dc_pred[1].type == type) {
1500 dc += dc_pred[1].dc;
1501 count++;
1502 }
1503
1504 /* using division instead of shift to correctly handle negative values */
1505 return count == 2 ? dc / 2 : last_dc[type];
1506 }
1507
vp4_set_tokens_base(Vp3DecodeContext * s)1508 static void vp4_set_tokens_base(Vp3DecodeContext *s)
1509 {
1510 int plane, i;
1511 int16_t *base = s->dct_tokens_base;
1512 for (plane = 0; plane < 3; plane++) {
1513 for (i = 0; i < 64; i++) {
1514 s->dct_tokens[plane][i] = base;
1515 base += s->fragment_width[!!plane] * s->fragment_height[!!plane];
1516 }
1517 }
1518 }
1519
vp4_unpack_dct_coeffs(Vp3DecodeContext * s,GetBitContext * gb)1520 static int vp4_unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1521 {
1522 int i, j;
1523 int dc_y_table;
1524 int dc_c_table;
1525 int ac_y_table;
1526 int ac_c_table;
1527 VLC *tables[2][64];
1528 int plane, sb_y, sb_x;
1529 int eob_tracker[64];
1530 VP4Predictor dc_pred[6][6];
1531 int last_dc[NB_VP4_DC_TYPES];
1532
1533 if (get_bits_left(gb) < 16)
1534 return AVERROR_INVALIDDATA;
1535
1536 /* fetch the DC table indexes */
1537 dc_y_table = get_bits(gb, 4);
1538 dc_c_table = get_bits(gb, 4);
1539
1540 ac_y_table = get_bits(gb, 4);
1541 ac_c_table = get_bits(gb, 4);
1542
1543 /* build tables of DC/AC VLC tables */
1544
1545 tables[0][0] = &s->dc_vlc[dc_y_table];
1546 tables[1][0] = &s->dc_vlc[dc_c_table];
1547 for (i = 1; i <= 5; i++) {
1548 tables[0][i] = &s->ac_vlc_1[ac_y_table];
1549 tables[1][i] = &s->ac_vlc_1[ac_c_table];
1550 }
1551 for (i = 6; i <= 14; i++) {
1552 tables[0][i] = &s->ac_vlc_2[ac_y_table];
1553 tables[1][i] = &s->ac_vlc_2[ac_c_table];
1554 }
1555 for (i = 15; i <= 27; i++) {
1556 tables[0][i] = &s->ac_vlc_3[ac_y_table];
1557 tables[1][i] = &s->ac_vlc_3[ac_c_table];
1558 }
1559 for (i = 28; i <= 63; i++) {
1560 tables[0][i] = &s->ac_vlc_4[ac_y_table];
1561 tables[1][i] = &s->ac_vlc_4[ac_c_table];
1562 }
1563
1564 vp4_set_tokens_base(s);
1565
1566 memset(last_dc, 0, sizeof(last_dc));
1567
1568 for (plane = 0; plane < ((s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 1 : 3); plane++) {
1569 memset(eob_tracker, 0, sizeof(eob_tracker));
1570
1571 /* initialise dc prediction */
1572 for (i = 0; i < s->fragment_width[!!plane]; i++)
1573 vp4_dc_predictor_reset(&s->dc_pred_row[i]);
1574
1575 for (j = 0; j < 6; j++)
1576 for (i = 0; i < 6; i++)
1577 vp4_dc_predictor_reset(&dc_pred[j][i]);
1578
1579 for (sb_y = 0; sb_y * 4 < s->fragment_height[!!plane]; sb_y++) {
1580 for (sb_x = 0; sb_x *4 < s->fragment_width[!!plane]; sb_x++) {
1581 vp4_dc_pred_before(s, dc_pred, sb_x);
1582 for (j = 0; j < 16; j++) {
1583 int hx = hilbert_offset[j][0];
1584 int hy = hilbert_offset[j][1];
1585 int x = 4 * sb_x + hx;
1586 int y = 4 * sb_y + hy;
1587 VP4Predictor *this_dc_pred = &dc_pred[hy + 1][hx + 1];
1588 int fragment, dc_block_type;
1589
1590 if (x >= s->fragment_width[!!plane] || y >= s->fragment_height[!!plane])
1591 continue;
1592
1593 fragment = s->fragment_start[plane] + y * s->fragment_width[!!plane] + x;
1594
1595 if (s->all_fragments[fragment].coding_method == MODE_COPY)
1596 continue;
1597
1598 if (vp4_unpack_vlcs(s, gb, tables[!!plane], plane, eob_tracker, fragment) < 0)
1599 return -1;
1600
1601 dc_block_type = vp4_pred_block_type_map[s->all_fragments[fragment].coding_method];
1602
1603 s->all_fragments[fragment].dc +=
1604 vp4_dc_pred(s, this_dc_pred, last_dc, dc_block_type, plane);
1605
1606 this_dc_pred->type = dc_block_type,
1607 this_dc_pred->dc = last_dc[dc_block_type] = s->all_fragments[fragment].dc;
1608 }
1609 vp4_dc_pred_after(s, dc_pred, sb_x);
1610 }
1611 }
1612 }
1613
1614 vp4_set_tokens_base(s);
1615
1616 return 0;
1617 }
1618 #endif
1619
1620 /*
1621 * This function reverses the DC prediction for each coded fragment in
1622 * the frame. Much of this function is adapted directly from the original
1623 * VP3 source code.
1624 */
1625 #define COMPATIBLE_FRAME(x) \
1626 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1627 #define DC_COEFF(u) s->all_fragments[u].dc
1628
reverse_dc_prediction(Vp3DecodeContext * s,int first_fragment,int fragment_width,int fragment_height)1629 static void reverse_dc_prediction(Vp3DecodeContext *s,
1630 int first_fragment,
1631 int fragment_width,
1632 int fragment_height)
1633 {
1634 #define PUL 8
1635 #define PU 4
1636 #define PUR 2
1637 #define PL 1
1638
1639 int x, y;
1640 int i = first_fragment;
1641
1642 int predicted_dc;
1643
1644 /* DC values for the left, up-left, up, and up-right fragments */
1645 int vl, vul, vu, vur;
1646
1647 /* indexes for the left, up-left, up, and up-right fragments */
1648 int l, ul, u, ur;
1649
1650 /*
1651 * The 6 fields mean:
1652 * 0: up-left multiplier
1653 * 1: up multiplier
1654 * 2: up-right multiplier
1655 * 3: left multiplier
1656 */
1657 static const int predictor_transform[16][4] = {
1658 { 0, 0, 0, 0 },
1659 { 0, 0, 0, 128 }, // PL
1660 { 0, 0, 128, 0 }, // PUR
1661 { 0, 0, 53, 75 }, // PUR|PL
1662 { 0, 128, 0, 0 }, // PU
1663 { 0, 64, 0, 64 }, // PU |PL
1664 { 0, 128, 0, 0 }, // PU |PUR
1665 { 0, 0, 53, 75 }, // PU |PUR|PL
1666 { 128, 0, 0, 0 }, // PUL
1667 { 0, 0, 0, 128 }, // PUL|PL
1668 { 64, 0, 64, 0 }, // PUL|PUR
1669 { 0, 0, 53, 75 }, // PUL|PUR|PL
1670 { 0, 128, 0, 0 }, // PUL|PU
1671 { -104, 116, 0, 116 }, // PUL|PU |PL
1672 { 24, 80, 24, 0 }, // PUL|PU |PUR
1673 { -104, 116, 0, 116 } // PUL|PU |PUR|PL
1674 };
1675
1676 /* This table shows which types of blocks can use other blocks for
1677 * prediction. For example, INTRA is the only mode in this table to
1678 * have a frame number of 0. That means INTRA blocks can only predict
1679 * from other INTRA blocks. There are 2 golden frame coding types;
1680 * blocks encoding in these modes can only predict from other blocks
1681 * that were encoded with these 1 of these 2 modes. */
1682 static const unsigned char compatible_frame[9] = {
1683 1, /* MODE_INTER_NO_MV */
1684 0, /* MODE_INTRA */
1685 1, /* MODE_INTER_PLUS_MV */
1686 1, /* MODE_INTER_LAST_MV */
1687 1, /* MODE_INTER_PRIOR_MV */
1688 2, /* MODE_USING_GOLDEN */
1689 2, /* MODE_GOLDEN_MV */
1690 1, /* MODE_INTER_FOUR_MV */
1691 3 /* MODE_COPY */
1692 };
1693 int current_frame_type;
1694
1695 /* there is a last DC predictor for each of the 3 frame types */
1696 short last_dc[3];
1697
1698 int transform = 0;
1699
1700 vul =
1701 vu =
1702 vur =
1703 vl = 0;
1704 last_dc[0] =
1705 last_dc[1] =
1706 last_dc[2] = 0;
1707
1708 /* for each fragment row... */
1709 for (y = 0; y < fragment_height; y++) {
1710 /* for each fragment in a row... */
1711 for (x = 0; x < fragment_width; x++, i++) {
1712
1713 /* reverse prediction if this block was coded */
1714 if (s->all_fragments[i].coding_method != MODE_COPY) {
1715 current_frame_type =
1716 compatible_frame[s->all_fragments[i].coding_method];
1717
1718 transform = 0;
1719 if (x) {
1720 l = i - 1;
1721 vl = DC_COEFF(l);
1722 if (COMPATIBLE_FRAME(l))
1723 transform |= PL;
1724 }
1725 if (y) {
1726 u = i - fragment_width;
1727 vu = DC_COEFF(u);
1728 if (COMPATIBLE_FRAME(u))
1729 transform |= PU;
1730 if (x) {
1731 ul = i - fragment_width - 1;
1732 vul = DC_COEFF(ul);
1733 if (COMPATIBLE_FRAME(ul))
1734 transform |= PUL;
1735 }
1736 if (x + 1 < fragment_width) {
1737 ur = i - fragment_width + 1;
1738 vur = DC_COEFF(ur);
1739 if (COMPATIBLE_FRAME(ur))
1740 transform |= PUR;
1741 }
1742 }
1743
1744 if (transform == 0) {
1745 /* if there were no fragments to predict from, use last
1746 * DC saved */
1747 predicted_dc = last_dc[current_frame_type];
1748 } else {
1749 /* apply the appropriate predictor transform */
1750 predicted_dc =
1751 (predictor_transform[transform][0] * vul) +
1752 (predictor_transform[transform][1] * vu) +
1753 (predictor_transform[transform][2] * vur) +
1754 (predictor_transform[transform][3] * vl);
1755
1756 predicted_dc /= 128;
1757
1758 /* check for outranging on the [ul u l] and
1759 * [ul u ur l] predictors */
1760 if ((transform == 15) || (transform == 13)) {
1761 if (FFABS(predicted_dc - vu) > 128)
1762 predicted_dc = vu;
1763 else if (FFABS(predicted_dc - vl) > 128)
1764 predicted_dc = vl;
1765 else if (FFABS(predicted_dc - vul) > 128)
1766 predicted_dc = vul;
1767 }
1768 }
1769
1770 /* at long last, apply the predictor */
1771 DC_COEFF(i) += predicted_dc;
1772 /* save the DC */
1773 last_dc[current_frame_type] = DC_COEFF(i);
1774 }
1775 }
1776 }
1777 }
1778
apply_loop_filter(Vp3DecodeContext * s,int plane,int ystart,int yend)1779 static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1780 int ystart, int yend)
1781 {
1782 int x, y;
1783 int *bounding_values = s->bounding_values_array + 127;
1784
1785 int width = s->fragment_width[!!plane];
1786 int height = s->fragment_height[!!plane];
1787 int fragment = s->fragment_start[plane] + ystart * width;
1788 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1789 uint8_t *plane_data = s->current_frame.f->data[plane];
1790 if (!s->flipped_image)
1791 stride = -stride;
1792 plane_data += s->data_offset[plane] + 8 * ystart * stride;
1793
1794 for (y = ystart; y < yend; y++) {
1795 for (x = 0; x < width; x++) {
1796 /* This code basically just deblocks on the edges of coded blocks.
1797 * However, it has to be much more complicated because of the
1798 * brain damaged deblock ordering used in VP3/Theora. Order matters
1799 * because some pixels get filtered twice. */
1800 if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1801 /* do not perform left edge filter for left columns frags */
1802 if (x > 0) {
1803 s->vp3dsp.h_loop_filter(
1804 plane_data + 8 * x,
1805 stride, bounding_values);
1806 }
1807
1808 /* do not perform top edge filter for top row fragments */
1809 if (y > 0) {
1810 s->vp3dsp.v_loop_filter(
1811 plane_data + 8 * x,
1812 stride, bounding_values);
1813 }
1814
1815 /* do not perform right edge filter for right column
1816 * fragments or if right fragment neighbor is also coded
1817 * in this frame (it will be filtered in next iteration) */
1818 if ((x < width - 1) &&
1819 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1820 s->vp3dsp.h_loop_filter(
1821 plane_data + 8 * x + 8,
1822 stride, bounding_values);
1823 }
1824
1825 /* do not perform bottom edge filter for bottom row
1826 * fragments or if bottom fragment neighbor is also coded
1827 * in this frame (it will be filtered in the next row) */
1828 if ((y < height - 1) &&
1829 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1830 s->vp3dsp.v_loop_filter(
1831 plane_data + 8 * x + 8 * stride,
1832 stride, bounding_values);
1833 }
1834 }
1835
1836 fragment++;
1837 }
1838 plane_data += 8 * stride;
1839 }
1840 }
1841
1842 /**
1843 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1844 * for the next block in coding order
1845 */
vp3_dequant(Vp3DecodeContext * s,Vp3Fragment * frag,int plane,int inter,int16_t block[64])1846 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1847 int plane, int inter, int16_t block[64])
1848 {
1849 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1850 uint8_t *perm = s->idct_scantable;
1851 int i = 0;
1852
1853 do {
1854 int token = *s->dct_tokens[plane][i];
1855 switch (token & 3) {
1856 case 0: // EOB
1857 if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1858 s->dct_tokens[plane][i]++;
1859 else
1860 *s->dct_tokens[plane][i] = token & ~3;
1861 goto end;
1862 case 1: // zero run
1863 s->dct_tokens[plane][i]++;
1864 i += (token >> 2) & 0x7f;
1865 if (i > 63) {
1866 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1867 return i;
1868 }
1869 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1870 i++;
1871 break;
1872 case 2: // coeff
1873 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1874 s->dct_tokens[plane][i++]++;
1875 break;
1876 default: // shouldn't happen
1877 return i;
1878 }
1879 } while (i < 64);
1880 // return value is expected to be a valid level
1881 i--;
1882 end:
1883 // the actual DC+prediction is in the fragment structure
1884 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1885 return i;
1886 }
1887
1888 /**
1889 * called when all pixels up to row y are complete
1890 */
vp3_draw_horiz_band(Vp3DecodeContext * s,int y)1891 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1892 {
1893 int h, cy, i;
1894 int offset[AV_NUM_DATA_POINTERS];
1895
1896 if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1897 int y_flipped = s->flipped_image ? s->height - y : y;
1898
1899 /* At the end of the frame, report INT_MAX instead of the height of
1900 * the frame. This makes the other threads' ff_thread_await_progress()
1901 * calls cheaper, because they don't have to clip their values. */
1902 ff_thread_report_progress(&s->current_frame,
1903 y_flipped == s->height ? INT_MAX
1904 : y_flipped - 1,
1905 0);
1906 }
1907
1908 if (!s->avctx->draw_horiz_band)
1909 return;
1910
1911 h = y - s->last_slice_end;
1912 s->last_slice_end = y;
1913 y -= h;
1914
1915 if (!s->flipped_image)
1916 y = s->height - y - h;
1917
1918 cy = y >> s->chroma_y_shift;
1919 offset[0] = s->current_frame.f->linesize[0] * y;
1920 offset[1] = s->current_frame.f->linesize[1] * cy;
1921 offset[2] = s->current_frame.f->linesize[2] * cy;
1922 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1923 offset[i] = 0;
1924
1925 emms_c();
1926 s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1927 }
1928
1929 /**
1930 * Wait for the reference frame of the current fragment.
1931 * The progress value is in luma pixel rows.
1932 */
await_reference_row(Vp3DecodeContext * s,Vp3Fragment * fragment,int motion_y,int y)1933 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1934 int motion_y, int y)
1935 {
1936 ThreadFrame *ref_frame;
1937 int ref_row;
1938 int border = motion_y & 1;
1939
1940 if (fragment->coding_method == MODE_USING_GOLDEN ||
1941 fragment->coding_method == MODE_GOLDEN_MV)
1942 ref_frame = &s->golden_frame;
1943 else
1944 ref_frame = &s->last_frame;
1945
1946 ref_row = y + (motion_y >> 1);
1947 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1948
1949 ff_thread_await_progress(ref_frame, ref_row, 0);
1950 }
1951
1952 #if CONFIG_VP4_DECODER
1953 /**
1954 * @return non-zero if temp (edge_emu_buffer) was populated
1955 */
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)1956 static int vp4_mc_loop_filter(Vp3DecodeContext *s, int plane, int motion_x, int motion_y, int bx, int by,
1957 uint8_t * motion_source, int stride, int src_x, int src_y, uint8_t *temp)
1958 {
1959 int motion_shift = plane ? 4 : 2;
1960 int subpel_mask = plane ? 3 : 1;
1961 int *bounding_values = s->bounding_values_array + 127;
1962
1963 int i;
1964 int x, y;
1965 int x2, y2;
1966 int x_subpel, y_subpel;
1967 int x_offset, y_offset;
1968
1969 int block_width = plane ? 8 : 16;
1970 int plane_width = s->width >> (plane && s->chroma_x_shift);
1971 int plane_height = s->height >> (plane && s->chroma_y_shift);
1972
1973 #define loop_stride 12
1974 uint8_t loop[12 * loop_stride];
1975
1976 /* using division instead of shift to correctly handle negative values */
1977 x = 8 * bx + motion_x / motion_shift;
1978 y = 8 * by + motion_y / motion_shift;
1979
1980 x_subpel = motion_x & subpel_mask;
1981 y_subpel = motion_y & subpel_mask;
1982
1983 if (x_subpel || y_subpel) {
1984 x--;
1985 y--;
1986
1987 if (x_subpel)
1988 x = FFMIN(x, x + FFSIGN(motion_x));
1989
1990 if (y_subpel)
1991 y = FFMIN(y, y + FFSIGN(motion_y));
1992
1993 x2 = x + block_width;
1994 y2 = y + block_width;
1995
1996 if (x2 < 0 || x2 >= plane_width || y2 < 0 || y2 >= plane_height)
1997 return 0;
1998
1999 x_offset = (-(x + 2) & 7) + 2;
2000 y_offset = (-(y + 2) & 7) + 2;
2001
2002 if (x_offset > 8 + x_subpel && y_offset > 8 + y_subpel)
2003 return 0;
2004
2005 s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2006 loop_stride, stride,
2007 12, 12, src_x - 1, src_y - 1,
2008 plane_width,
2009 plane_height);
2010
2011 if (x_offset <= 8 + x_subpel)
2012 ff_vp3dsp_h_loop_filter_12(loop + x_offset, loop_stride, bounding_values);
2013
2014 if (y_offset <= 8 + y_subpel)
2015 ff_vp3dsp_v_loop_filter_12(loop + y_offset*loop_stride, loop_stride, bounding_values);
2016
2017 } else {
2018
2019 x_offset = -x & 7;
2020 y_offset = -y & 7;
2021
2022 if (!x_offset && !y_offset)
2023 return 0;
2024
2025 s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
2026 loop_stride, stride,
2027 12, 12, src_x - 1, src_y - 1,
2028 plane_width,
2029 plane_height);
2030
2031 #define safe_loop_filter(name, ptr, stride, bounding_values) \
2032 if ((uintptr_t)(ptr) & 7) \
2033 s->vp3dsp.name##_unaligned(ptr, stride, bounding_values); \
2034 else \
2035 s->vp3dsp.name(ptr, stride, bounding_values);
2036
2037 if (x_offset)
2038 safe_loop_filter(h_loop_filter, loop + loop_stride + x_offset + 1, loop_stride, bounding_values);
2039
2040 if (y_offset)
2041 safe_loop_filter(v_loop_filter, loop + (y_offset + 1)*loop_stride + 1, loop_stride, bounding_values);
2042 }
2043
2044 for (i = 0; i < 9; i++)
2045 memcpy(temp + i*stride, loop + (i + 1) * loop_stride + 1, 9);
2046
2047 return 1;
2048 }
2049 #endif
2050
2051 /*
2052 * Perform the final rendering for a particular slice of data.
2053 * The slice number ranges from 0..(c_superblock_height - 1).
2054 */
render_slice(Vp3DecodeContext * s,int slice)2055 static void render_slice(Vp3DecodeContext *s, int slice)
2056 {
2057 int x, y, i, j, fragment;
2058 int16_t *block = s->block;
2059 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
2060 int motion_halfpel_index;
2061 uint8_t *motion_source;
2062 int plane, first_pixel;
2063
2064 if (slice >= s->c_superblock_height)
2065 return;
2066
2067 for (plane = 0; plane < 3; plane++) {
2068 uint8_t *output_plane = s->current_frame.f->data[plane] +
2069 s->data_offset[plane];
2070 uint8_t *last_plane = s->last_frame.f->data[plane] +
2071 s->data_offset[plane];
2072 uint8_t *golden_plane = s->golden_frame.f->data[plane] +
2073 s->data_offset[plane];
2074 ptrdiff_t stride = s->current_frame.f->linesize[plane];
2075 int plane_width = s->width >> (plane && s->chroma_x_shift);
2076 int plane_height = s->height >> (plane && s->chroma_y_shift);
2077 int8_t(*motion_val)[2] = s->motion_val[!!plane];
2078
2079 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
2080 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
2081 int slice_width = plane ? s->c_superblock_width
2082 : s->y_superblock_width;
2083
2084 int fragment_width = s->fragment_width[!!plane];
2085 int fragment_height = s->fragment_height[!!plane];
2086 int fragment_start = s->fragment_start[plane];
2087
2088 int do_await = !plane && HAVE_THREADS &&
2089 (s->avctx->active_thread_type & FF_THREAD_FRAME);
2090
2091 if (!s->flipped_image)
2092 stride = -stride;
2093 if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY))
2094 continue;
2095
2096 /* for each superblock row in the slice (both of them)... */
2097 for (; sb_y < slice_height; sb_y++) {
2098 /* for each superblock in a row... */
2099 for (sb_x = 0; sb_x < slice_width; sb_x++) {
2100 /* for each block in a superblock... */
2101 for (j = 0; j < 16; j++) {
2102 x = 4 * sb_x + hilbert_offset[j][0];
2103 y = 4 * sb_y + hilbert_offset[j][1];
2104 fragment = y * fragment_width + x;
2105
2106 i = fragment_start + fragment;
2107
2108 // bounds check
2109 if (x >= fragment_width || y >= fragment_height)
2110 continue;
2111
2112 first_pixel = 8 * y * stride + 8 * x;
2113
2114 if (do_await &&
2115 s->all_fragments[i].coding_method != MODE_INTRA)
2116 await_reference_row(s, &s->all_fragments[i],
2117 motion_val[fragment][1],
2118 (16 * y) >> s->chroma_y_shift);
2119
2120 /* transform if this block was coded */
2121 if (s->all_fragments[i].coding_method != MODE_COPY) {
2122 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
2123 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
2124 motion_source = golden_plane;
2125 else
2126 motion_source = last_plane;
2127
2128 motion_source += first_pixel;
2129 motion_halfpel_index = 0;
2130
2131 /* sort out the motion vector if this fragment is coded
2132 * using a motion vector method */
2133 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
2134 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
2135 int src_x, src_y;
2136 int standard_mc = 1;
2137 motion_x = motion_val[fragment][0];
2138 motion_y = motion_val[fragment][1];
2139 #if CONFIG_VP4_DECODER
2140 if (plane && s->version >= 2) {
2141 motion_x = (motion_x >> 1) | (motion_x & 1);
2142 motion_y = (motion_y >> 1) | (motion_y & 1);
2143 }
2144 #endif
2145
2146 src_x = (motion_x >> 1) + 8 * x;
2147 src_y = (motion_y >> 1) + 8 * y;
2148
2149 motion_halfpel_index = motion_x & 0x01;
2150 motion_source += (motion_x >> 1);
2151
2152 motion_halfpel_index |= (motion_y & 0x01) << 1;
2153 motion_source += ((motion_y >> 1) * stride);
2154
2155 #if CONFIG_VP4_DECODER
2156 if (s->version >= 2) {
2157 uint8_t *temp = s->edge_emu_buffer;
2158 if (stride < 0)
2159 temp -= 8 * stride;
2160 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)) {
2161 motion_source = temp;
2162 standard_mc = 0;
2163 }
2164 }
2165 #endif
2166
2167 if (standard_mc && (
2168 src_x < 0 || src_y < 0 ||
2169 src_x + 9 >= plane_width ||
2170 src_y + 9 >= plane_height)) {
2171 uint8_t *temp = s->edge_emu_buffer;
2172 if (stride < 0)
2173 temp -= 8 * stride;
2174
2175 s->vdsp.emulated_edge_mc(temp, motion_source,
2176 stride, stride,
2177 9, 9, src_x, src_y,
2178 plane_width,
2179 plane_height);
2180 motion_source = temp;
2181 }
2182 }
2183
2184 /* first, take care of copying a block from either the
2185 * previous or the golden frame */
2186 if (s->all_fragments[i].coding_method != MODE_INTRA) {
2187 /* Note, it is possible to implement all MC cases
2188 * with put_no_rnd_pixels_l2 which would look more
2189 * like the VP3 source but this would be slower as
2190 * put_no_rnd_pixels_tab is better optimized */
2191 if (motion_halfpel_index != 3) {
2192 s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
2193 output_plane + first_pixel,
2194 motion_source, stride, 8);
2195 } else {
2196 /* d is 0 if motion_x and _y have the same sign,
2197 * else -1 */
2198 int d = (motion_x ^ motion_y) >> 31;
2199 s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
2200 motion_source - d,
2201 motion_source + stride + 1 + d,
2202 stride, 8);
2203 }
2204 }
2205
2206 /* invert DCT and place (or add) in final output */
2207
2208 if (s->all_fragments[i].coding_method == MODE_INTRA) {
2209 vp3_dequant(s, s->all_fragments + i,
2210 plane, 0, block);
2211 s->vp3dsp.idct_put(output_plane + first_pixel,
2212 stride,
2213 block);
2214 } else {
2215 if (vp3_dequant(s, s->all_fragments + i,
2216 plane, 1, block)) {
2217 s->vp3dsp.idct_add(output_plane + first_pixel,
2218 stride,
2219 block);
2220 } else {
2221 s->vp3dsp.idct_dc_add(output_plane + first_pixel,
2222 stride, block);
2223 }
2224 }
2225 } else {
2226 /* copy directly from the previous frame */
2227 s->hdsp.put_pixels_tab[1][0](
2228 output_plane + first_pixel,
2229 last_plane + first_pixel,
2230 stride, 8);
2231 }
2232 }
2233 }
2234
2235 // Filter up to the last row in the superblock row
2236 if (s->version < 2 && !s->skip_loop_filter)
2237 apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
2238 FFMIN(4 * sb_y + 3, fragment_height - 1));
2239 }
2240 }
2241
2242 /* this looks like a good place for slice dispatch... */
2243 /* algorithm:
2244 * if (slice == s->macroblock_height - 1)
2245 * dispatch (both last slice & 2nd-to-last slice);
2246 * else if (slice > 0)
2247 * dispatch (slice - 1);
2248 */
2249
2250 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
2251 s->height - 16));
2252 }
2253
2254 /// Allocate tables for per-frame data in Vp3DecodeContext
allocate_tables(AVCodecContext * avctx)2255 static av_cold int allocate_tables(AVCodecContext *avctx)
2256 {
2257 Vp3DecodeContext *s = avctx->priv_data;
2258 int y_fragment_count, c_fragment_count;
2259
2260 free_tables(avctx);
2261
2262 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
2263 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
2264
2265 /* superblock_coding is used by unpack_superblocks (VP3/Theora) and vp4_unpack_macroblocks (VP4) */
2266 s->superblock_coding = av_mallocz(FFMAX(s->superblock_count, s->yuv_macroblock_count));
2267 s->all_fragments = av_mallocz_array(s->fragment_count, sizeof(Vp3Fragment));
2268
2269 s-> kf_coded_fragment_list = av_mallocz_array(s->fragment_count, sizeof(int));
2270 s->nkf_coded_fragment_list = av_mallocz_array(s->fragment_count, sizeof(int));
2271 memset(s-> num_kf_coded_fragment, -1, sizeof(s-> num_kf_coded_fragment));
2272
2273 s->dct_tokens_base = av_mallocz_array(s->fragment_count,
2274 64 * sizeof(*s->dct_tokens_base));
2275 s->motion_val[0] = av_mallocz_array(y_fragment_count, sizeof(*s->motion_val[0]));
2276 s->motion_val[1] = av_mallocz_array(c_fragment_count, sizeof(*s->motion_val[1]));
2277
2278 /* work out the block mapping tables */
2279 s->superblock_fragments = av_mallocz_array(s->superblock_count, 16 * sizeof(int));
2280 s->macroblock_coding = av_mallocz(s->macroblock_count + 1);
2281
2282 s->dc_pred_row = av_malloc_array(s->y_superblock_width * 4, sizeof(*s->dc_pred_row));
2283
2284 if (!s->superblock_coding || !s->all_fragments ||
2285 !s->dct_tokens_base || !s->kf_coded_fragment_list ||
2286 !s->nkf_coded_fragment_list ||
2287 !s->superblock_fragments || !s->macroblock_coding ||
2288 !s->dc_pred_row ||
2289 !s->motion_val[0] || !s->motion_val[1]) {
2290 vp3_decode_end(avctx);
2291 return -1;
2292 }
2293
2294 init_block_mapping(s);
2295
2296 return 0;
2297 }
2298
init_frames(Vp3DecodeContext * s)2299 static av_cold int init_frames(Vp3DecodeContext *s)
2300 {
2301 s->current_frame.f = av_frame_alloc();
2302 s->last_frame.f = av_frame_alloc();
2303 s->golden_frame.f = av_frame_alloc();
2304
2305 if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) {
2306 av_frame_free(&s->current_frame.f);
2307 av_frame_free(&s->last_frame.f);
2308 av_frame_free(&s->golden_frame.f);
2309 return AVERROR(ENOMEM);
2310 }
2311
2312 return 0;
2313 }
2314
vp3_decode_init(AVCodecContext * avctx)2315 static av_cold int vp3_decode_init(AVCodecContext *avctx)
2316 {
2317 Vp3DecodeContext *s = avctx->priv_data;
2318 int i, inter, plane, ret;
2319 int c_width;
2320 int c_height;
2321 int y_fragment_count, c_fragment_count;
2322 #if CONFIG_VP4_DECODER
2323 int j;
2324 #endif
2325
2326 ret = init_frames(s);
2327 if (ret < 0)
2328 return ret;
2329
2330 if (avctx->codec_tag == MKTAG('V', 'P', '4', '0'))
2331 s->version = 3;
2332 else if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
2333 s->version = 0;
2334 else
2335 s->version = 1;
2336
2337 s->avctx = avctx;
2338 s->width = FFALIGN(avctx->coded_width, 16);
2339 s->height = FFALIGN(avctx->coded_height, 16);
2340 if (avctx->codec_id != AV_CODEC_ID_THEORA)
2341 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
2342 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
2343 ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT);
2344 ff_videodsp_init(&s->vdsp, 8);
2345 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
2346
2347 for (i = 0; i < 64; i++) {
2348 #define TRANSPOSE(x) (((x) >> 3) | (((x) & 7) << 3))
2349 s->idct_permutation[i] = TRANSPOSE(i);
2350 s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]);
2351 #undef TRANSPOSE
2352 }
2353
2354 /* initialize to an impossible value which will force a recalculation
2355 * in the first frame decode */
2356 for (i = 0; i < 3; i++)
2357 s->qps[i] = -1;
2358
2359 ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
2360 if (ret)
2361 return ret;
2362
2363 s->y_superblock_width = (s->width + 31) / 32;
2364 s->y_superblock_height = (s->height + 31) / 32;
2365 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
2366
2367 /* work out the dimensions for the C planes */
2368 c_width = s->width >> s->chroma_x_shift;
2369 c_height = s->height >> s->chroma_y_shift;
2370 s->c_superblock_width = (c_width + 31) / 32;
2371 s->c_superblock_height = (c_height + 31) / 32;
2372 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
2373
2374 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
2375 s->u_superblock_start = s->y_superblock_count;
2376 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
2377
2378 s->macroblock_width = (s->width + 15) / 16;
2379 s->macroblock_height = (s->height + 15) / 16;
2380 s->macroblock_count = s->macroblock_width * s->macroblock_height;
2381 s->c_macroblock_width = (c_width + 15) / 16;
2382 s->c_macroblock_height = (c_height + 15) / 16;
2383 s->c_macroblock_count = s->c_macroblock_width * s->c_macroblock_height;
2384 s->yuv_macroblock_count = s->macroblock_count + 2 * s->c_macroblock_count;
2385
2386 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
2387 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
2388 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
2389 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
2390
2391 /* fragment count covers all 8x8 blocks for all 3 planes */
2392 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
2393 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
2394 s->fragment_count = y_fragment_count + 2 * c_fragment_count;
2395 s->fragment_start[1] = y_fragment_count;
2396 s->fragment_start[2] = y_fragment_count + c_fragment_count;
2397
2398 if (!s->theora_tables) {
2399 for (i = 0; i < 64; i++) {
2400 s->coded_dc_scale_factor[0][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_y_dc_scale_factor[i];
2401 s->coded_dc_scale_factor[1][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_uv_dc_scale_factor[i];
2402 s->coded_ac_scale_factor[i] = s->version < 2 ? vp31_ac_scale_factor[i] : vp4_ac_scale_factor[i];
2403 s->base_matrix[0][i] = s->version < 2 ? vp31_intra_y_dequant[i] : vp4_generic_dequant[i];
2404 s->base_matrix[1][i] = s->version < 2 ? vp31_intra_c_dequant[i] : vp4_generic_dequant[i];
2405 s->base_matrix[2][i] = s->version < 2 ? vp31_inter_dequant[i] : vp4_generic_dequant[i];
2406 s->filter_limit_values[i] = s->version < 2 ? vp31_filter_limit_values[i] : vp4_filter_limit_values[i];
2407 }
2408
2409 for (inter = 0; inter < 2; inter++) {
2410 for (plane = 0; plane < 3; plane++) {
2411 s->qr_count[inter][plane] = 1;
2412 s->qr_size[inter][plane][0] = 63;
2413 s->qr_base[inter][plane][0] =
2414 s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
2415 }
2416 }
2417
2418 /* init VLC tables */
2419 if (s->version < 2) {
2420 for (i = 0; i < 16; i++) {
2421 /* DC histograms */
2422 init_vlc(&s->dc_vlc[i], 11, 32,
2423 &dc_bias[i][0][1], 4, 2,
2424 &dc_bias[i][0][0], 4, 2, 0);
2425
2426 /* group 1 AC histograms */
2427 init_vlc(&s->ac_vlc_1[i], 11, 32,
2428 &ac_bias_0[i][0][1], 4, 2,
2429 &ac_bias_0[i][0][0], 4, 2, 0);
2430
2431 /* group 2 AC histograms */
2432 init_vlc(&s->ac_vlc_2[i], 11, 32,
2433 &ac_bias_1[i][0][1], 4, 2,
2434 &ac_bias_1[i][0][0], 4, 2, 0);
2435
2436 /* group 3 AC histograms */
2437 init_vlc(&s->ac_vlc_3[i], 11, 32,
2438 &ac_bias_2[i][0][1], 4, 2,
2439 &ac_bias_2[i][0][0], 4, 2, 0);
2440
2441 /* group 4 AC histograms */
2442 init_vlc(&s->ac_vlc_4[i], 11, 32,
2443 &ac_bias_3[i][0][1], 4, 2,
2444 &ac_bias_3[i][0][0], 4, 2, 0);
2445 }
2446 #if CONFIG_VP4_DECODER
2447 } else { /* version >= 2 */
2448 for (i = 0; i < 16; i++) {
2449 /* DC histograms */
2450 init_vlc(&s->dc_vlc[i], 11, 32,
2451 &vp4_dc_bias[i][0][1], 4, 2,
2452 &vp4_dc_bias[i][0][0], 4, 2, 0);
2453
2454 /* group 1 AC histograms */
2455 init_vlc(&s->ac_vlc_1[i], 11, 32,
2456 &vp4_ac_bias_0[i][0][1], 4, 2,
2457 &vp4_ac_bias_0[i][0][0], 4, 2, 0);
2458
2459 /* group 2 AC histograms */
2460 init_vlc(&s->ac_vlc_2[i], 11, 32,
2461 &vp4_ac_bias_1[i][0][1], 4, 2,
2462 &vp4_ac_bias_1[i][0][0], 4, 2, 0);
2463
2464 /* group 3 AC histograms */
2465 init_vlc(&s->ac_vlc_3[i], 11, 32,
2466 &vp4_ac_bias_2[i][0][1], 4, 2,
2467 &vp4_ac_bias_2[i][0][0], 4, 2, 0);
2468
2469 /* group 4 AC histograms */
2470 init_vlc(&s->ac_vlc_4[i], 11, 32,
2471 &vp4_ac_bias_3[i][0][1], 4, 2,
2472 &vp4_ac_bias_3[i][0][0], 4, 2, 0);
2473 }
2474 #endif
2475 }
2476 } else {
2477 for (i = 0; i < 16; i++) {
2478 /* DC histograms */
2479 if (init_vlc(&s->dc_vlc[i], 11, 32,
2480 &s->huffman_table[i][0][1], 8, 4,
2481 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
2482 goto vlc_fail;
2483
2484 /* group 1 AC histograms */
2485 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
2486 &s->huffman_table[i + 16][0][1], 8, 4,
2487 &s->huffman_table[i + 16][0][0], 8, 4, 0) < 0)
2488 goto vlc_fail;
2489
2490 /* group 2 AC histograms */
2491 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
2492 &s->huffman_table[i + 16 * 2][0][1], 8, 4,
2493 &s->huffman_table[i + 16 * 2][0][0], 8, 4, 0) < 0)
2494 goto vlc_fail;
2495
2496 /* group 3 AC histograms */
2497 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
2498 &s->huffman_table[i + 16 * 3][0][1], 8, 4,
2499 &s->huffman_table[i + 16 * 3][0][0], 8, 4, 0) < 0)
2500 goto vlc_fail;
2501
2502 /* group 4 AC histograms */
2503 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
2504 &s->huffman_table[i + 16 * 4][0][1], 8, 4,
2505 &s->huffman_table[i + 16 * 4][0][0], 8, 4, 0) < 0)
2506 goto vlc_fail;
2507 }
2508 }
2509
2510 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2511 &superblock_run_length_vlc_table[0][1], 4, 2,
2512 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2513
2514 init_vlc(&s->fragment_run_length_vlc, 5, 30,
2515 &fragment_run_length_vlc_table[0][1], 4, 2,
2516 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2517
2518 init_vlc(&s->mode_code_vlc, 3, 8,
2519 &mode_code_vlc_table[0][1], 2, 1,
2520 &mode_code_vlc_table[0][0], 2, 1, 0);
2521
2522 init_vlc(&s->motion_vector_vlc, 6, 63,
2523 &motion_vector_vlc_table[0][1], 2, 1,
2524 &motion_vector_vlc_table[0][0], 2, 1, 0);
2525
2526 #if CONFIG_VP4_DECODER
2527 for (j = 0; j < 2; j++)
2528 for (i = 0; i < 7; i++)
2529 init_vlc(&s->vp4_mv_vlc[j][i], 6, 63,
2530 &vp4_mv_vlc[j][i][0][1], 4, 2,
2531 &vp4_mv_vlc[j][i][0][0], 4, 2, 0);
2532
2533 /* version >= 2 */
2534 for (i = 0; i < 2; i++)
2535 init_vlc(&s->block_pattern_vlc[i], 3, 14,
2536 &vp4_block_pattern_vlc[i][0][1], 2, 1,
2537 &vp4_block_pattern_vlc[i][0][0], 2, 1, 0);
2538 #endif
2539
2540 return allocate_tables(avctx);
2541
2542 vlc_fail:
2543 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
2544 return -1;
2545 }
2546
2547 /// Release and shuffle frames after decode finishes
update_frames(AVCodecContext * avctx)2548 static int update_frames(AVCodecContext *avctx)
2549 {
2550 Vp3DecodeContext *s = avctx->priv_data;
2551 int ret = 0;
2552
2553 /* shuffle frames (last = current) */
2554 ff_thread_release_buffer(avctx, &s->last_frame);
2555 ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
2556 if (ret < 0)
2557 goto fail;
2558
2559 if (s->keyframe) {
2560 ff_thread_release_buffer(avctx, &s->golden_frame);
2561 ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
2562 }
2563
2564 fail:
2565 ff_thread_release_buffer(avctx, &s->current_frame);
2566 return ret;
2567 }
2568
2569 #if HAVE_THREADS
ref_frame(Vp3DecodeContext * s,ThreadFrame * dst,ThreadFrame * src)2570 static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
2571 {
2572 ff_thread_release_buffer(s->avctx, dst);
2573 if (src->f->data[0])
2574 return ff_thread_ref_frame(dst, src);
2575 return 0;
2576 }
2577
ref_frames(Vp3DecodeContext * dst,Vp3DecodeContext * src)2578 static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
2579 {
2580 int ret;
2581 if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
2582 (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 ||
2583 (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0)
2584 return ret;
2585 return 0;
2586 }
2587
vp3_update_thread_context(AVCodecContext * dst,const AVCodecContext * src)2588 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
2589 {
2590 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
2591 int qps_changed = 0, i, err;
2592
2593 if (!s1->current_frame.f->data[0] ||
2594 s->width != s1->width || s->height != s1->height) {
2595 if (s != s1)
2596 ref_frames(s, s1);
2597 return -1;
2598 }
2599
2600 if (s != s1) {
2601 // copy previous frame data
2602 if ((err = ref_frames(s, s1)) < 0)
2603 return err;
2604
2605 s->keyframe = s1->keyframe;
2606
2607 // copy qscale data if necessary
2608 for (i = 0; i < 3; i++) {
2609 if (s->qps[i] != s1->qps[1]) {
2610 qps_changed = 1;
2611 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
2612 }
2613 }
2614
2615 if (s->qps[0] != s1->qps[0])
2616 memcpy(&s->bounding_values_array, &s1->bounding_values_array,
2617 sizeof(s->bounding_values_array));
2618
2619 if (qps_changed) {
2620 memcpy(s->qps, s1->qps, sizeof(s->qps));
2621 memcpy(s->last_qps, s1->last_qps, sizeof(s->last_qps));
2622 s->nqps = s1->nqps;
2623 }
2624 }
2625
2626 return update_frames(dst);
2627 }
2628 #endif
2629
vp3_decode_frame(AVCodecContext * avctx,void * data,int * got_frame,AVPacket * avpkt)2630 static int vp3_decode_frame(AVCodecContext *avctx,
2631 void *data, int *got_frame,
2632 AVPacket *avpkt)
2633 {
2634 AVFrame *frame = data;
2635 const uint8_t *buf = avpkt->data;
2636 int buf_size = avpkt->size;
2637 Vp3DecodeContext *s = avctx->priv_data;
2638 GetBitContext gb;
2639 int i, ret;
2640
2641 if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0)
2642 return ret;
2643
2644 #if CONFIG_THEORA_DECODER
2645 if (s->theora && get_bits1(&gb)) {
2646 int type = get_bits(&gb, 7);
2647 skip_bits_long(&gb, 6*8); /* "theora" */
2648
2649 if (s->avctx->active_thread_type&FF_THREAD_FRAME) {
2650 av_log(avctx, AV_LOG_ERROR, "midstream reconfiguration with multithreading is unsupported, try -threads 1\n");
2651 return AVERROR_PATCHWELCOME;
2652 }
2653 if (type == 0) {
2654 vp3_decode_end(avctx);
2655 ret = theora_decode_header(avctx, &gb);
2656
2657 if (ret >= 0)
2658 ret = vp3_decode_init(avctx);
2659 if (ret < 0) {
2660 vp3_decode_end(avctx);
2661 return ret;
2662 }
2663 return buf_size;
2664 } else if (type == 2) {
2665 vp3_decode_end(avctx);
2666 ret = theora_decode_tables(avctx, &gb);
2667 if (ret >= 0)
2668 ret = vp3_decode_init(avctx);
2669 if (ret < 0) {
2670 vp3_decode_end(avctx);
2671 return ret;
2672 }
2673 return buf_size;
2674 }
2675
2676 av_log(avctx, AV_LOG_ERROR,
2677 "Header packet passed to frame decoder, skipping\n");
2678 return -1;
2679 }
2680 #endif
2681
2682 s->keyframe = !get_bits1(&gb);
2683 if (!s->all_fragments) {
2684 av_log(avctx, AV_LOG_ERROR, "Data packet without prior valid headers\n");
2685 return -1;
2686 }
2687 if (!s->theora)
2688 skip_bits(&gb, 1);
2689 for (i = 0; i < 3; i++)
2690 s->last_qps[i] = s->qps[i];
2691
2692 s->nqps = 0;
2693 do {
2694 s->qps[s->nqps++] = get_bits(&gb, 6);
2695 } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2696 for (i = s->nqps; i < 3; i++)
2697 s->qps[i] = -1;
2698
2699 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2700 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2701 s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2702
2703 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2704 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2705 : AVDISCARD_NONKEY);
2706
2707 if (s->qps[0] != s->last_qps[0])
2708 init_loop_filter(s);
2709
2710 for (i = 0; i < s->nqps; i++)
2711 // reinit all dequantizers if the first one changed, because
2712 // the DC of the first quantizer must be used for all matrices
2713 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2714 init_dequantizer(s, i);
2715
2716 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2717 return buf_size;
2718
2719 s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2720 : AV_PICTURE_TYPE_P;
2721 s->current_frame.f->key_frame = s->keyframe;
2722 if ((ret = ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF)) < 0)
2723 goto error;
2724
2725 if (!s->edge_emu_buffer) {
2726 s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2727 if (!s->edge_emu_buffer) {
2728 ret = AVERROR(ENOMEM);
2729 goto error;
2730 }
2731 }
2732
2733 if (s->keyframe) {
2734 if (!s->theora) {
2735 skip_bits(&gb, 4); /* width code */
2736 skip_bits(&gb, 4); /* height code */
2737 if (s->version) {
2738 s->version = get_bits(&gb, 5);
2739 if (avctx->frame_number == 0)
2740 av_log(s->avctx, AV_LOG_DEBUG,
2741 "VP version: %d\n", s->version);
2742 }
2743 }
2744 if (s->version || s->theora) {
2745 if (get_bits1(&gb))
2746 av_log(s->avctx, AV_LOG_ERROR,
2747 "Warning, unsupported keyframe coding type?!\n");
2748 skip_bits(&gb, 2); /* reserved? */
2749
2750 #if CONFIG_VP4_DECODER
2751 if (s->version >= 2) {
2752 int mb_height, mb_width;
2753 int mb_width_mul, mb_width_div, mb_height_mul, mb_height_div;
2754
2755 mb_height = get_bits(&gb, 8);
2756 mb_width = get_bits(&gb, 8);
2757 if (mb_height != s->macroblock_height ||
2758 mb_width != s->macroblock_width)
2759 avpriv_request_sample(s->avctx, "macroblock dimension mismatch");
2760
2761 mb_width_mul = get_bits(&gb, 5);
2762 mb_width_div = get_bits(&gb, 3);
2763 mb_height_mul = get_bits(&gb, 5);
2764 mb_height_div = get_bits(&gb, 3);
2765 if (mb_width_mul != 1 || mb_width_div != 1 || mb_height_mul != 1 || mb_height_div != 1)
2766 avpriv_request_sample(s->avctx, "unexpected macroblock dimension multipler/divider");
2767
2768 if (get_bits(&gb, 2))
2769 avpriv_request_sample(s->avctx, "unknown bits");
2770 }
2771 #endif
2772 }
2773 } else {
2774 if (!s->golden_frame.f->data[0]) {
2775 av_log(s->avctx, AV_LOG_WARNING,
2776 "vp3: first frame not a keyframe\n");
2777
2778 s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2779 if ((ret = ff_thread_get_buffer(avctx, &s->golden_frame,
2780 AV_GET_BUFFER_FLAG_REF)) < 0)
2781 goto error;
2782 ff_thread_release_buffer(avctx, &s->last_frame);
2783 if ((ret = ff_thread_ref_frame(&s->last_frame,
2784 &s->golden_frame)) < 0)
2785 goto error;
2786 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2787 }
2788 }
2789
2790 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2791 ff_thread_finish_setup(avctx);
2792
2793 if (s->version < 2) {
2794 if ((ret = unpack_superblocks(s, &gb)) < 0) {
2795 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2796 goto error;
2797 }
2798 #if CONFIG_VP4_DECODER
2799 } else {
2800 if ((ret = vp4_unpack_macroblocks(s, &gb)) < 0) {
2801 av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_macroblocks\n");
2802 goto error;
2803 }
2804 #endif
2805 }
2806 if ((ret = unpack_modes(s, &gb)) < 0) {
2807 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2808 goto error;
2809 }
2810 if (ret = unpack_vectors(s, &gb)) {
2811 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2812 goto error;
2813 }
2814 if ((ret = unpack_block_qpis(s, &gb)) < 0) {
2815 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2816 goto error;
2817 }
2818
2819 if (s->version < 2) {
2820 if ((ret = unpack_dct_coeffs(s, &gb)) < 0) {
2821 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2822 goto error;
2823 }
2824 #if CONFIG_VP4_DECODER
2825 } else {
2826 if ((ret = vp4_unpack_dct_coeffs(s, &gb)) < 0) {
2827 av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_dct_coeffs\n");
2828 goto error;
2829 }
2830 #endif
2831 }
2832
2833 for (i = 0; i < 3; i++) {
2834 int height = s->height >> (i && s->chroma_y_shift);
2835 if (s->flipped_image)
2836 s->data_offset[i] = 0;
2837 else
2838 s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2839 }
2840
2841 s->last_slice_end = 0;
2842 for (i = 0; i < s->c_superblock_height; i++)
2843 render_slice(s, i);
2844
2845 // filter the last row
2846 if (s->version < 2)
2847 for (i = 0; i < 3; i++) {
2848 int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2849 apply_loop_filter(s, i, row, row + 1);
2850 }
2851 vp3_draw_horiz_band(s, s->height);
2852
2853 /* output frame, offset as needed */
2854 if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2855 return ret;
2856
2857 frame->crop_left = s->offset_x;
2858 frame->crop_right = avctx->coded_width - avctx->width - s->offset_x;
2859 frame->crop_top = s->offset_y;
2860 frame->crop_bottom = avctx->coded_height - avctx->height - s->offset_y;
2861
2862 *got_frame = 1;
2863
2864 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2865 ret = update_frames(avctx);
2866 if (ret < 0)
2867 return ret;
2868 }
2869
2870 return buf_size;
2871
2872 error:
2873 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2874
2875 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2876 av_frame_unref(s->current_frame.f);
2877
2878 return ret;
2879 }
2880
read_huffman_tree(AVCodecContext * avctx,GetBitContext * gb)2881 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2882 {
2883 Vp3DecodeContext *s = avctx->priv_data;
2884
2885 if (get_bits1(gb)) {
2886 int token;
2887 if (s->entries >= 32) { /* overflow */
2888 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2889 return -1;
2890 }
2891 token = get_bits(gb, 5);
2892 ff_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2893 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2894 s->huffman_table[s->hti][token][0] = s->hbits;
2895 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2896 s->entries++;
2897 } else {
2898 if (s->huff_code_size >= 32) { /* overflow */
2899 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2900 return -1;
2901 }
2902 s->huff_code_size++;
2903 s->hbits <<= 1;
2904 if (read_huffman_tree(avctx, gb))
2905 return -1;
2906 s->hbits |= 1;
2907 if (read_huffman_tree(avctx, gb))
2908 return -1;
2909 s->hbits >>= 1;
2910 s->huff_code_size--;
2911 }
2912 return 0;
2913 }
2914
2915 #if CONFIG_THEORA_DECODER
2916 static const enum AVPixelFormat theora_pix_fmts[4] = {
2917 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2918 };
2919
theora_decode_header(AVCodecContext * avctx,GetBitContext * gb)2920 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2921 {
2922 Vp3DecodeContext *s = avctx->priv_data;
2923 int visible_width, visible_height, colorspace;
2924 uint8_t offset_x = 0, offset_y = 0;
2925 int ret;
2926 AVRational fps, aspect;
2927
2928 if (get_bits_left(gb) < 206)
2929 return AVERROR_INVALIDDATA;
2930
2931 s->theora_header = 0;
2932 s->theora = get_bits(gb, 24);
2933 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2934 if (!s->theora) {
2935 s->theora = 1;
2936 avpriv_request_sample(s->avctx, "theora 0");
2937 }
2938
2939 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2940 * but previous versions have the image flipped relative to vp3 */
2941 if (s->theora < 0x030200) {
2942 s->flipped_image = 1;
2943 av_log(avctx, AV_LOG_DEBUG,
2944 "Old (<alpha3) Theora bitstream, flipped image\n");
2945 }
2946
2947 visible_width =
2948 s->width = get_bits(gb, 16) << 4;
2949 visible_height =
2950 s->height = get_bits(gb, 16) << 4;
2951
2952 if (s->theora >= 0x030200) {
2953 visible_width = get_bits(gb, 24);
2954 visible_height = get_bits(gb, 24);
2955
2956 offset_x = get_bits(gb, 8); /* offset x */
2957 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2958 }
2959
2960 /* sanity check */
2961 if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 ||
2962 visible_width + offset_x > s->width ||
2963 visible_height + offset_y > s->height) {
2964 av_log(avctx, AV_LOG_ERROR,
2965 "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n",
2966 visible_width, visible_height, offset_x, offset_y,
2967 s->width, s->height);
2968 return AVERROR_INVALIDDATA;
2969 }
2970
2971 fps.num = get_bits_long(gb, 32);
2972 fps.den = get_bits_long(gb, 32);
2973 if (fps.num && fps.den) {
2974 if (fps.num < 0 || fps.den < 0) {
2975 av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2976 return AVERROR_INVALIDDATA;
2977 }
2978 av_reduce(&avctx->framerate.den, &avctx->framerate.num,
2979 fps.den, fps.num, 1 << 30);
2980 }
2981
2982 aspect.num = get_bits(gb, 24);
2983 aspect.den = get_bits(gb, 24);
2984 if (aspect.num && aspect.den) {
2985 av_reduce(&avctx->sample_aspect_ratio.num,
2986 &avctx->sample_aspect_ratio.den,
2987 aspect.num, aspect.den, 1 << 30);
2988 ff_set_sar(avctx, avctx->sample_aspect_ratio);
2989 }
2990
2991 if (s->theora < 0x030200)
2992 skip_bits(gb, 5); /* keyframe frequency force */
2993 colorspace = get_bits(gb, 8);
2994 skip_bits(gb, 24); /* bitrate */
2995
2996 skip_bits(gb, 6); /* quality hint */
2997
2998 if (s->theora >= 0x030200) {
2999 skip_bits(gb, 5); /* keyframe frequency force */
3000 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
3001 if (avctx->pix_fmt == AV_PIX_FMT_NONE) {
3002 av_log(avctx, AV_LOG_ERROR, "Invalid pixel format\n");
3003 return AVERROR_INVALIDDATA;
3004 }
3005 skip_bits(gb, 3); /* reserved */
3006 } else
3007 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
3008
3009 ret = ff_set_dimensions(avctx, s->width, s->height);
3010 if (ret < 0)
3011 return ret;
3012 if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP)) {
3013 avctx->width = visible_width;
3014 avctx->height = visible_height;
3015 // translate offsets from theora axis ([0,0] lower left)
3016 // to normal axis ([0,0] upper left)
3017 s->offset_x = offset_x;
3018 s->offset_y = s->height - visible_height - offset_y;
3019 }
3020
3021 if (colorspace == 1)
3022 avctx->color_primaries = AVCOL_PRI_BT470M;
3023 else if (colorspace == 2)
3024 avctx->color_primaries = AVCOL_PRI_BT470BG;
3025
3026 if (colorspace == 1 || colorspace == 2) {
3027 avctx->colorspace = AVCOL_SPC_BT470BG;
3028 avctx->color_trc = AVCOL_TRC_BT709;
3029 }
3030
3031 s->theora_header = 1;
3032 return 0;
3033 }
3034
theora_decode_tables(AVCodecContext * avctx,GetBitContext * gb)3035 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
3036 {
3037 Vp3DecodeContext *s = avctx->priv_data;
3038 int i, n, matrices, inter, plane;
3039
3040 if (!s->theora_header)
3041 return AVERROR_INVALIDDATA;
3042
3043 if (s->theora >= 0x030200) {
3044 n = get_bits(gb, 3);
3045 /* loop filter limit values table */
3046 if (n)
3047 for (i = 0; i < 64; i++)
3048 s->filter_limit_values[i] = get_bits(gb, n);
3049 }
3050
3051 if (s->theora >= 0x030200)
3052 n = get_bits(gb, 4) + 1;
3053 else
3054 n = 16;
3055 /* quality threshold table */
3056 for (i = 0; i < 64; i++)
3057 s->coded_ac_scale_factor[i] = get_bits(gb, n);
3058
3059 if (s->theora >= 0x030200)
3060 n = get_bits(gb, 4) + 1;
3061 else
3062 n = 16;
3063 /* dc scale factor table */
3064 for (i = 0; i < 64; i++)
3065 s->coded_dc_scale_factor[0][i] =
3066 s->coded_dc_scale_factor[1][i] = get_bits(gb, n);
3067
3068 if (s->theora >= 0x030200)
3069 matrices = get_bits(gb, 9) + 1;
3070 else
3071 matrices = 3;
3072
3073 if (matrices > 384) {
3074 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
3075 return -1;
3076 }
3077
3078 for (n = 0; n < matrices; n++)
3079 for (i = 0; i < 64; i++)
3080 s->base_matrix[n][i] = get_bits(gb, 8);
3081
3082 for (inter = 0; inter <= 1; inter++) {
3083 for (plane = 0; plane <= 2; plane++) {
3084 int newqr = 1;
3085 if (inter || plane > 0)
3086 newqr = get_bits1(gb);
3087 if (!newqr) {
3088 int qtj, plj;
3089 if (inter && get_bits1(gb)) {
3090 qtj = 0;
3091 plj = plane;
3092 } else {
3093 qtj = (3 * inter + plane - 1) / 3;
3094 plj = (plane + 2) % 3;
3095 }
3096 s->qr_count[inter][plane] = s->qr_count[qtj][plj];
3097 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
3098 sizeof(s->qr_size[0][0]));
3099 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
3100 sizeof(s->qr_base[0][0]));
3101 } else {
3102 int qri = 0;
3103 int qi = 0;
3104
3105 for (;;) {
3106 i = get_bits(gb, av_log2(matrices - 1) + 1);
3107 if (i >= matrices) {
3108 av_log(avctx, AV_LOG_ERROR,
3109 "invalid base matrix index\n");
3110 return -1;
3111 }
3112 s->qr_base[inter][plane][qri] = i;
3113 if (qi >= 63)
3114 break;
3115 i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
3116 s->qr_size[inter][plane][qri++] = i;
3117 qi += i;
3118 }
3119
3120 if (qi > 63) {
3121 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
3122 return -1;
3123 }
3124 s->qr_count[inter][plane] = qri;
3125 }
3126 }
3127 }
3128
3129 /* Huffman tables */
3130 for (s->hti = 0; s->hti < 80; s->hti++) {
3131 s->entries = 0;
3132 s->huff_code_size = 1;
3133 if (!get_bits1(gb)) {
3134 s->hbits = 0;
3135 if (read_huffman_tree(avctx, gb))
3136 return -1;
3137 s->hbits = 1;
3138 if (read_huffman_tree(avctx, gb))
3139 return -1;
3140 }
3141 }
3142
3143 s->theora_tables = 1;
3144
3145 return 0;
3146 }
3147
theora_decode_init(AVCodecContext * avctx)3148 static av_cold int theora_decode_init(AVCodecContext *avctx)
3149 {
3150 Vp3DecodeContext *s = avctx->priv_data;
3151 GetBitContext gb;
3152 int ptype;
3153 const uint8_t *header_start[3];
3154 int header_len[3];
3155 int i;
3156 int ret;
3157
3158 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
3159
3160 s->theora = 1;
3161
3162 if (!avctx->extradata_size) {
3163 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
3164 return -1;
3165 }
3166
3167 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
3168 42, header_start, header_len) < 0) {
3169 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
3170 return -1;
3171 }
3172
3173 for (i = 0; i < 3; i++) {
3174 if (header_len[i] <= 0)
3175 continue;
3176 ret = init_get_bits8(&gb, header_start[i], header_len[i]);
3177 if (ret < 0)
3178 return ret;
3179
3180 ptype = get_bits(&gb, 8);
3181
3182 if (!(ptype & 0x80)) {
3183 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
3184 // return -1;
3185 }
3186
3187 // FIXME: Check for this as well.
3188 skip_bits_long(&gb, 6 * 8); /* "theora" */
3189
3190 switch (ptype) {
3191 case 0x80:
3192 if (theora_decode_header(avctx, &gb) < 0)
3193 return -1;
3194 break;
3195 case 0x81:
3196 // FIXME: is this needed? it breaks sometimes
3197 // theora_decode_comments(avctx, gb);
3198 break;
3199 case 0x82:
3200 if (theora_decode_tables(avctx, &gb))
3201 return -1;
3202 break;
3203 default:
3204 av_log(avctx, AV_LOG_ERROR,
3205 "Unknown Theora config packet: %d\n", ptype & ~0x80);
3206 break;
3207 }
3208 if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb))
3209 av_log(avctx, AV_LOG_WARNING,
3210 "%d bits left in packet %X\n",
3211 8 * header_len[i] - get_bits_count(&gb), ptype);
3212 if (s->theora < 0x030200)
3213 break;
3214 }
3215
3216 return vp3_decode_init(avctx);
3217 }
3218
3219 AVCodec ff_theora_decoder = {
3220 .name = "theora",
3221 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
3222 .type = AVMEDIA_TYPE_VIDEO,
3223 .id = AV_CODEC_ID_THEORA,
3224 .priv_data_size = sizeof(Vp3DecodeContext),
3225 .init = theora_decode_init,
3226 .close = vp3_decode_end,
3227 .decode = vp3_decode_frame,
3228 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3229 AV_CODEC_CAP_FRAME_THREADS,
3230 .flush = vp3_decode_flush,
3231 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3232 .caps_internal = FF_CODEC_CAP_EXPORTS_CROPPING | FF_CODEC_CAP_ALLOCATE_PROGRESS,
3233 };
3234 #endif
3235
3236 AVCodec ff_vp3_decoder = {
3237 .name = "vp3",
3238 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
3239 .type = AVMEDIA_TYPE_VIDEO,
3240 .id = AV_CODEC_ID_VP3,
3241 .priv_data_size = sizeof(Vp3DecodeContext),
3242 .init = vp3_decode_init,
3243 .close = vp3_decode_end,
3244 .decode = vp3_decode_frame,
3245 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3246 AV_CODEC_CAP_FRAME_THREADS,
3247 .flush = vp3_decode_flush,
3248 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3249 .caps_internal = FF_CODEC_CAP_ALLOCATE_PROGRESS,
3250 };
3251
3252 #if CONFIG_VP4_DECODER
3253 AVCodec ff_vp4_decoder = {
3254 .name = "vp4",
3255 .long_name = NULL_IF_CONFIG_SMALL("On2 VP4"),
3256 .type = AVMEDIA_TYPE_VIDEO,
3257 .id = AV_CODEC_ID_VP4,
3258 .priv_data_size = sizeof(Vp3DecodeContext),
3259 .init = vp3_decode_init,
3260 .close = vp3_decode_end,
3261 .decode = vp3_decode_frame,
3262 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
3263 AV_CODEC_CAP_FRAME_THREADS,
3264 .flush = vp3_decode_flush,
3265 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
3266 .caps_internal = FF_CODEC_CAP_ALLOCATE_PROGRESS,
3267 };
3268 #endif
3269