?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
aarch64: Add assembly support for -fsanitize=hwaddress tagged globals.
[libav.git] / libavcodec / vp3.c
blobcb8925ba4c5a7dd831ccd0f0ca142fddcdf825ca
1 /*
2 * Copyright (C) 2003-2004 The FFmpeg project
3 *
4 * This file is part of Libav.
5 *
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 /**
22 * @file
23 * On2 VP3 Video Decoder
24 *
25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26 * For more information about the VP3 coding process, visit:
27 * http://wiki.multimedia.cx/index.php?title=On2_VP3
28 *
29 * Theora decoder by Alex Beregszaszi
30 */
31
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <string.h>
35
36 #include "libavutil/imgutils.h"
37
38 #include "avcodec.h"
39 #include "get_bits.h"
40 #include "hpeldsp.h"
41 #include "internal.h"
42 #include "mathops.h"
43 #include "thread.h"
44 #include "videodsp.h"
45 #include "vp3data.h"
46 #include "vp3dsp.h"
47 #include "xiph.h"
48
49 #define FRAGMENT_PIXELS 8
50
51 // FIXME split things out into their own arrays
52 typedef struct Vp3Fragment {
53 int16_t dc;
54 uint8_t coding_method;
55 uint8_t qpi;
56 } Vp3Fragment;
57
58 #define SB_NOT_CODED 0
59 #define SB_PARTIALLY_CODED 1
60 #define SB_FULLY_CODED 2
61
62 // This is the maximum length of a single long bit run that can be encoded
63 // for superblock coding or block qps. Theora special-cases this to read a
64 // bit instead of flipping the current bit to allow for runs longer than 4129.
65 #define MAXIMUM_LONG_BIT_RUN 4129
66
67 #define MODE_INTER_NO_MV 0
68 #define MODE_INTRA 1
69 #define MODE_INTER_PLUS_MV 2
70 #define MODE_INTER_LAST_MV 3
71 #define MODE_INTER_PRIOR_LAST 4
72 #define MODE_USING_GOLDEN 5
73 #define MODE_GOLDEN_MV 6
74 #define MODE_INTER_FOURMV 7
75 #define CODING_MODE_COUNT 8
76
77 /* special internal mode */
78 #define MODE_COPY 8
79
80 /* There are 6 preset schemes, plus a free-form scheme */
81 static const int ModeAlphabet[6][CODING_MODE_COUNT] = {
82 /* scheme 1: Last motion vector dominates */
83 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
84 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
85 MODE_INTRA, MODE_USING_GOLDEN,
86 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
87
88 /* scheme 2 */
89 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
90 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
91 MODE_INTRA, MODE_USING_GOLDEN,
92 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
93
94 /* scheme 3 */
95 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
96 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
97 MODE_INTRA, MODE_USING_GOLDEN,
98 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
99
100 /* scheme 4 */
101 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
102 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
103 MODE_INTRA, MODE_USING_GOLDEN,
104 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
105
106 /* scheme 5: No motion vector dominates */
107 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
108 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
109 MODE_INTRA, MODE_USING_GOLDEN,
110 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
111
112 /* scheme 6 */
113 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
114 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
115 MODE_INTER_PLUS_MV, MODE_INTRA,
116 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
117 };
118
119 static const uint8_t hilbert_offset[16][2] = {
120 { 0, 0 }, { 1, 0 }, { 1, 1 }, { 0, 1 },
121 { 0, 2 }, { 0, 3 }, { 1, 3 }, { 1, 2 },
122 { 2, 2 }, { 2, 3 }, { 3, 3 }, { 3, 2 },
123 { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 }
124 };
125
126 #define MIN_DEQUANT_VAL 2
127
128 typedef struct Vp3DecodeContext {
129 AVCodecContext *avctx;
130 int theora, theora_tables;
131 int version;
132 int width, height;
133 int chroma_x_shift, chroma_y_shift;
134 ThreadFrame golden_frame;
135 ThreadFrame last_frame;
136 ThreadFrame current_frame;
137 int keyframe;
138 uint8_t idct_permutation[64];
139 uint8_t idct_scantable[64];
140 HpelDSPContext hdsp;
141 VideoDSPContext vdsp;
142 VP3DSPContext vp3dsp;
143 DECLARE_ALIGNED(16, int16_t, block)[64];
144 int flipped_image;
145 int last_slice_end;
146 int skip_loop_filter;
147
148 int qps[3];
149 int nqps;
150 int last_qps[3];
151
152 int superblock_count;
153 int y_superblock_width;
154 int y_superblock_height;
155 int y_superblock_count;
156 int c_superblock_width;
157 int c_superblock_height;
158 int c_superblock_count;
159 int u_superblock_start;
160 int v_superblock_start;
161 unsigned char *superblock_coding;
162
163 int macroblock_count;
164 int macroblock_width;
165 int macroblock_height;
166
167 int fragment_count;
168 int fragment_width[2];
169 int fragment_height[2];
170
171 Vp3Fragment *all_fragments;
172 int fragment_start[3];
173 int data_offset[3];
174 uint8_t offset_x;
175 uint8_t offset_y;
176
177 int8_t (*motion_val[2])[2];
178
179 /* tables */
180 uint16_t coded_dc_scale_factor[64];
181 uint32_t coded_ac_scale_factor[64];
182 uint8_t base_matrix[384][64];
183 uint8_t qr_count[2][3];
184 uint8_t qr_size[2][3][64];
185 uint16_t qr_base[2][3][64];
186
187 /**
188 * This is a list of all tokens in bitstream order. Reordering takes place
189 * by pulling from each level during IDCT. As a consequence, IDCT must be
190 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
191 * otherwise. The 32 different tokens with up to 12 bits of extradata are
192 * collapsed into 3 types, packed as follows:
193 * (from the low to high bits)
194 *
195 * 2 bits: type (0,1,2)
196 * 0: EOB run, 14 bits for run length (12 needed)
197 * 1: zero run, 7 bits for run length
198 * 7 bits for the next coefficient (3 needed)
199 * 2: coefficient, 14 bits (11 needed)
200 *
201 * Coefficients are signed, so are packed in the highest bits for automatic
202 * sign extension.
203 */
204 int16_t *dct_tokens[3][64];
205 int16_t *dct_tokens_base;
206 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
207 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
208 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
209
210 /**
211 * number of blocks that contain DCT coefficients at
212 * the given level or higher
213 */
214 int num_coded_frags[3][64];
215 int total_num_coded_frags;
216
217 /* this is a list of indexes into the all_fragments array indicating
218 * which of the fragments are coded */
219 int *coded_fragment_list[3];
220
221 VLC dc_vlc[16];
222 VLC ac_vlc_1[16];
223 VLC ac_vlc_2[16];
224 VLC ac_vlc_3[16];
225 VLC ac_vlc_4[16];
226
227 VLC superblock_run_length_vlc;
228 VLC fragment_run_length_vlc;
229 VLC mode_code_vlc;
230 VLC motion_vector_vlc;
231
232 /* these arrays need to be on 16-byte boundaries since SSE2 operations
233 * index into them */
234 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
235
236 /* This table contains superblock_count * 16 entries. Each set of 16
237 * numbers corresponds to the fragment indexes 0..15 of the superblock.
238 * An entry will be -1 to indicate that no entry corresponds to that
239 * index. */
240 int *superblock_fragments;
241
242 /* This is an array that indicates how a particular macroblock
243 * is coded. */
244 unsigned char *macroblock_coding;
245
246 uint8_t *edge_emu_buffer;
247
248 /* Huffman decode */
249 int hti;
250 unsigned int hbits;
251 int entries;
252 int huff_code_size;
253 uint32_t huffman_table[80][32][2];
254
255 uint8_t filter_limit_values[64];
256 DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
257 } Vp3DecodeContext;
258
259 /************************************************************************
260 * VP3 specific functions
261 ************************************************************************/
262
263 static void vp3_decode_flush(AVCodecContext *avctx)
264 {
265 Vp3DecodeContext *s = avctx->priv_data;
266
267 if (s->golden_frame.f)
268 ff_thread_release_buffer(avctx, &s->golden_frame);
269 if (s->last_frame.f)
270 ff_thread_release_buffer(avctx, &s->last_frame);
271 if (s->current_frame.f)
272 ff_thread_release_buffer(avctx, &s->current_frame);
273 }
274
275 static av_cold int vp3_decode_end(AVCodecContext *avctx)
276 {
277 Vp3DecodeContext *s = avctx->priv_data;
278 int i;
279
280 av_freep(&s->superblock_coding);
281 av_freep(&s->all_fragments);
282 av_freep(&s->coded_fragment_list[0]);
283 av_freep(&s->dct_tokens_base);
284 av_freep(&s->superblock_fragments);
285 av_freep(&s->macroblock_coding);
286 av_freep(&s->motion_val[0]);
287 av_freep(&s->motion_val[1]);
288 av_freep(&s->edge_emu_buffer);
289
290 /* release all frames */
291 vp3_decode_flush(avctx);
292 av_frame_free(&s->current_frame.f);
293 av_frame_free(&s->last_frame.f);
294 av_frame_free(&s->golden_frame.f);
295
296 if (avctx->internal->is_copy)
297 return 0;
298
299 for (i = 0; i < 16; i++) {
300 ff_free_vlc(&s->dc_vlc[i]);
301 ff_free_vlc(&s->ac_vlc_1[i]);
302 ff_free_vlc(&s->ac_vlc_2[i]);
303 ff_free_vlc(&s->ac_vlc_3[i]);
304 ff_free_vlc(&s->ac_vlc_4[i]);
305 }
306
307 ff_free_vlc(&s->superblock_run_length_vlc);
308 ff_free_vlc(&s->fragment_run_length_vlc);
309 ff_free_vlc(&s->mode_code_vlc);
310 ff_free_vlc(&s->motion_vector_vlc);
311
312 return 0;
313 }
314
315 /*
316 * This function sets up all of the various blocks mappings:
317 * superblocks <-> fragments, macroblocks <-> fragments,
318 * superblocks <-> macroblocks
319 *
320 * @return 0 is successful; returns 1 if *anything* went wrong.
321 */
322 static int init_block_mapping(Vp3DecodeContext *s)
323 {
324 int sb_x, sb_y, plane;
325 int x, y, i, j = 0;
326
327 for (plane = 0; plane < 3; plane++) {
328 int sb_width = plane ? s->c_superblock_width
329 : s->y_superblock_width;
330 int sb_height = plane ? s->c_superblock_height
331 : s->y_superblock_height;
332 int frag_width = s->fragment_width[!!plane];
333 int frag_height = s->fragment_height[!!plane];
334
335 for (sb_y = 0; sb_y < sb_height; sb_y++)
336 for (sb_x = 0; sb_x < sb_width; sb_x++)
337 for (i = 0; i < 16; i++) {
338 x = 4 * sb_x + hilbert_offset[i][0];
339 y = 4 * sb_y + hilbert_offset[i][1];
340
341 if (x < frag_width && y < frag_height)
342 s->superblock_fragments[j++] = s->fragment_start[plane] +
343 y * frag_width + x;
344 else
345 s->superblock_fragments[j++] = -1;
346 }
347 }
348
349 return 0; /* successful path out */
350 }
351
352 /*
353 * This function sets up the dequantization tables used for a particular
354 * frame.
355 */
356 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
357 {
358 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
359 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
360 int i, plane, inter, qri, bmi, bmj, qistart;
361
362 for (inter = 0; inter < 2; inter++) {
363 for (plane = 0; plane < 3; plane++) {
364 int sum = 0;
365 for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
366 sum += s->qr_size[inter][plane][qri];
367 if (s->qps[qpi] <= sum)
368 break;
369 }
370 qistart = sum - s->qr_size[inter][plane][qri];
371 bmi = s->qr_base[inter][plane][qri];
372 bmj = s->qr_base[inter][plane][qri + 1];
373 for (i = 0; i < 64; i++) {
374 int coeff = (2 * (sum - s->qps[qpi]) * s->base_matrix[bmi][i] -
375 2 * (qistart - s->qps[qpi]) * s->base_matrix[bmj][i] +
376 s->qr_size[inter][plane][qri]) /
377 (2 * s->qr_size[inter][plane][qri]);
378
379 int qmin = 8 << (inter + !i);
380 int qscale = i ? ac_scale_factor : dc_scale_factor;
381
382 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
383 av_clip((qscale * coeff) / 100 * 4, qmin, 4096);
384 }
385 /* all DC coefficients use the same quant so as not to interfere
386 * with DC prediction */
387 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
388 }
389 }
390 }
391
392 /*
393 * This function initializes the loop filter boundary limits if the frame's
394 * quality index is different from the previous frame's.
395 *
396 * The filter_limit_values may not be larger than 127.
397 */
398 static void init_loop_filter(Vp3DecodeContext *s)
399 {
400 int *bounding_values = s->bounding_values_array + 127;
401 int filter_limit;
402 int x;
403 int value;
404
405 filter_limit = s->filter_limit_values[s->qps[0]];
406 assert(filter_limit < 128);
407
408 /* set up the bounding values */
409 memset(s->bounding_values_array, 0, 256 * sizeof(int));
410 for (x = 0; x < filter_limit; x++) {
411 bounding_values[-x] = -x;
412 bounding_values[x] = x;
413 }
414 for (x = value = filter_limit; x < 128 && value; x++, value--) {
415 bounding_values[ x] = value;
416 bounding_values[-x] = -value;
417 }
418 if (value)
419 bounding_values[128] = value;
420 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
421 }
422
423 /*
424 * This function unpacks all of the superblock/macroblock/fragment coding
425 * information from the bitstream.
426 */
427 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
428 {
429 int superblock_starts[3] = {
430 0, s->u_superblock_start, s->v_superblock_start
431 };
432 int bit = 0;
433 int current_superblock = 0;
434 int current_run = 0;
435 int num_partial_superblocks = 0;
436
437 int i, j;
438 int current_fragment;
439 int plane;
440
441 if (s->keyframe) {
442 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
443 } else {
444 /* unpack the list of partially-coded superblocks */
445 bit = get_bits1(gb) ^ 1;
446 current_run = 0;
447
448 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
449 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
450 bit = get_bits1(gb);
451 else
452 bit ^= 1;
453
454 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
455 6, 2) + 1;
456 if (current_run == 34)
457 current_run += get_bits(gb, 12);
458
459 if (current_superblock + current_run > s->superblock_count) {
460 av_log(s->avctx, AV_LOG_ERROR,
461 "Invalid partially coded superblock run length\n");
462 return -1;
463 }
464
465 memset(s->superblock_coding + current_superblock, bit, current_run);
466
467 current_superblock += current_run;
468 if (bit)
469 num_partial_superblocks += current_run;
470 }
471
472 /* unpack the list of fully coded superblocks if any of the blocks were
473 * not marked as partially coded in the previous step */
474 if (num_partial_superblocks < s->superblock_count) {
475 int superblocks_decoded = 0;
476
477 current_superblock = 0;
478 bit = get_bits1(gb) ^ 1;
479 current_run = 0;
480
481 while (superblocks_decoded < s->superblock_count - num_partial_superblocks &&
482 get_bits_left(gb) > 0) {
483 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
484 bit = get_bits1(gb);
485 else
486 bit ^= 1;
487
488 current_run = get_vlc2(gb, s->superblock_run_length_vlc.table,
489 6, 2) + 1;
490 if (current_run == 34)
491 current_run += get_bits(gb, 12);
492
493 for (j = 0; j < current_run; current_superblock++) {
494 if (current_superblock >= s->superblock_count) {
495 av_log(s->avctx, AV_LOG_ERROR,
496 "Invalid fully coded superblock run length\n");
497 return -1;
498 }
499
500 /* skip any superblocks already marked as partially coded */
501 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
502 s->superblock_coding[current_superblock] = 2 * bit;
503 j++;
504 }
505 }
506 superblocks_decoded += current_run;
507 }
508 }
509
510 /* if there were partial blocks, initialize bitstream for
511 * unpacking fragment codings */
512 if (num_partial_superblocks) {
513 current_run = 0;
514 bit = get_bits1(gb);
515 /* toggle the bit because as soon as the first run length is
516 * fetched the bit will be toggled again */
517 bit ^= 1;
518 }
519 }
520
521 /* figure out which fragments are coded; iterate through each
522 * superblock (all planes) */
523 s->total_num_coded_frags = 0;
524 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
525
526 for (plane = 0; plane < 3; plane++) {
527 int sb_start = superblock_starts[plane];
528 int sb_end = sb_start + (plane ? s->c_superblock_count
529 : s->y_superblock_count);
530 int num_coded_frags = 0;
531
532 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
533 /* iterate through all 16 fragments in a superblock */
534 for (j = 0; j < 16; j++) {
535 /* if the fragment is in bounds, check its coding status */
536 current_fragment = s->superblock_fragments[i * 16 + j];
537 if (current_fragment != -1) {
538 int coded = s->superblock_coding[i];
539
540 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
541 /* fragment may or may not be coded; this is the case
542 * that cares about the fragment coding runs */
543 if (current_run-- == 0) {
544 bit ^= 1;
545 current_run = get_vlc2(gb, s->fragment_run_length_vlc.table, 5, 2);
546 }
547 coded = bit;
548 }
549
550 if (coded) {
551 /* default mode; actual mode will be decoded in
552 * the next phase */
553 s->all_fragments[current_fragment].coding_method =
554 MODE_INTER_NO_MV;
555 s->coded_fragment_list[plane][num_coded_frags++] =
556 current_fragment;
557 } else {
558 /* not coded; copy this fragment from the prior frame */
559 s->all_fragments[current_fragment].coding_method =
560 MODE_COPY;
561 }
562 }
563 }
564 }
565 s->total_num_coded_frags += num_coded_frags;
566 for (i = 0; i < 64; i++)
567 s->num_coded_frags[plane][i] = num_coded_frags;
568 if (plane < 2)
569 s->coded_fragment_list[plane + 1] = s->coded_fragment_list[plane] +
570 num_coded_frags;
571 }
572 return 0;
573 }
574
575 /*
576 * This function unpacks all the coding mode data for individual macroblocks
577 * from the bitstream.
578 */
579 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
580 {
581 int i, j, k, sb_x, sb_y;
582 int scheme;
583 int current_macroblock;
584 int current_fragment;
585 int coding_mode;
586 int custom_mode_alphabet[CODING_MODE_COUNT];
587 const int *alphabet;
588 Vp3Fragment *frag;
589
590 if (s->keyframe) {
591 for (i = 0; i < s->fragment_count; i++)
592 s->all_fragments[i].coding_method = MODE_INTRA;
593 } else {
594 /* fetch the mode coding scheme for this frame */
595 scheme = get_bits(gb, 3);
596
597 /* is it a custom coding scheme? */
598 if (scheme == 0) {
599 for (i = 0; i < 8; i++)
600 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
601 for (i = 0; i < 8; i++)
602 custom_mode_alphabet[get_bits(gb, 3)] = i;
603 alphabet = custom_mode_alphabet;
604 } else
605 alphabet = ModeAlphabet[scheme - 1];
606
607 /* iterate through all of the macroblocks that contain 1 or more
608 * coded fragments */
609 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
610 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
611 if (get_bits_left(gb) <= 0)
612 return -1;
613
614 for (j = 0; j < 4; j++) {
615 int mb_x = 2 * sb_x + (j >> 1);
616 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
617 current_macroblock = mb_y * s->macroblock_width + mb_x;
618
619 if (mb_x >= s->macroblock_width ||
620 mb_y >= s->macroblock_height)
621 continue;
622
623 #define BLOCK_X (2 * mb_x + (k & 1))
624 #define BLOCK_Y (2 * mb_y + (k >> 1))
625 /* coding modes are only stored if the macroblock has
626 * at least one luma block coded, otherwise it must be
627 * INTER_NO_MV */
628 for (k = 0; k < 4; k++) {
629 current_fragment = BLOCK_Y *
630 s->fragment_width[0] + BLOCK_X;
631 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
632 break;
633 }
634 if (k == 4) {
635 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
636 continue;
637 }
638
639 /* mode 7 means get 3 bits for each coding mode */
640 if (scheme == 7)
641 coding_mode = get_bits(gb, 3);
642 else
643 coding_mode = alphabet[get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
644
645 s->macroblock_coding[current_macroblock] = coding_mode;
646 for (k = 0; k < 4; k++) {
647 frag = s->all_fragments + BLOCK_Y * s->fragment_width[0] + BLOCK_X;
648 if (frag->coding_method != MODE_COPY)
649 frag->coding_method = coding_mode;
650 }
651
652 #define SET_CHROMA_MODES \
653 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
654 frag[s->fragment_start[1]].coding_method = coding_mode; \
655 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
656 frag[s->fragment_start[2]].coding_method = coding_mode;
657
658 if (s->chroma_y_shift) {
659 frag = s->all_fragments + mb_y *
660 s->fragment_width[1] + mb_x;
661 SET_CHROMA_MODES
662 } else if (s->chroma_x_shift) {
663 frag = s->all_fragments +
664 2 * mb_y * s->fragment_width[1] + mb_x;
665 for (k = 0; k < 2; k++) {
666 SET_CHROMA_MODES
667 frag += s->fragment_width[1];
668 }
669 } else {
670 for (k = 0; k < 4; k++) {
671 frag = s->all_fragments +
672 BLOCK_Y * s->fragment_width[1] + BLOCK_X;
673 SET_CHROMA_MODES
674 }
675 }
676 }
677 }
678 }
679 }
680
681 return 0;
682 }
683
684 /*
685 * This function unpacks all the motion vectors for the individual
686 * macroblocks from the bitstream.
687 */
688 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
689 {
690 int j, k, sb_x, sb_y;
691 int coding_mode;
692 int motion_x[4];
693 int motion_y[4];
694 int last_motion_x = 0;
695 int last_motion_y = 0;
696 int prior_last_motion_x = 0;
697 int prior_last_motion_y = 0;
698 int current_macroblock;
699 int current_fragment;
700 int frag;
701
702 if (s->keyframe)
703 return 0;
704
705 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
706 coding_mode = get_bits1(gb);
707
708 /* iterate through all of the macroblocks that contain 1 or more
709 * coded fragments */
710 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
711 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
712 if (get_bits_left(gb) <= 0)
713 return -1;
714
715 for (j = 0; j < 4; j++) {
716 int mb_x = 2 * sb_x + (j >> 1);
717 int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
718 current_macroblock = mb_y * s->macroblock_width + mb_x;
719
720 if (mb_x >= s->macroblock_width ||
721 mb_y >= s->macroblock_height ||
722 s->macroblock_coding[current_macroblock] == MODE_COPY)
723 continue;
724
725 switch (s->macroblock_coding[current_macroblock]) {
726 case MODE_INTER_PLUS_MV:
727 case MODE_GOLDEN_MV:
728 /* all 6 fragments use the same motion vector */
729 if (coding_mode == 0) {
730 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
731 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
732 } else {
733 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
734 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
735 }
736
737 /* vector maintenance, only on MODE_INTER_PLUS_MV */
738 if (s->macroblock_coding[current_macroblock] == MODE_INTER_PLUS_MV) {
739 prior_last_motion_x = last_motion_x;
740 prior_last_motion_y = last_motion_y;
741 last_motion_x = motion_x[0];
742 last_motion_y = motion_y[0];
743 }
744 break;
745
746 case MODE_INTER_FOURMV:
747 /* vector maintenance */
748 prior_last_motion_x = last_motion_x;
749 prior_last_motion_y = last_motion_y;
750
751 /* fetch 4 vectors from the bitstream, one for each
752 * Y fragment, then average for the C fragment vectors */
753 for (k = 0; k < 4; k++) {
754 current_fragment = BLOCK_Y * s->fragment_width[0] + BLOCK_X;
755 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
756 if (coding_mode == 0) {
757 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
758 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
759 } else {
760 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
761 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
762 }
763 last_motion_x = motion_x[k];
764 last_motion_y = motion_y[k];
765 } else {
766 motion_x[k] = 0;
767 motion_y[k] = 0;
768 }
769 }
770 break;
771
772 case MODE_INTER_LAST_MV:
773 /* all 6 fragments use the last motion vector */
774 motion_x[0] = last_motion_x;
775 motion_y[0] = last_motion_y;
776
777 /* no vector maintenance (last vector remains the
778 * last vector) */
779 break;
780
781 case MODE_INTER_PRIOR_LAST:
782 /* all 6 fragments use the motion vector prior to the
783 * last motion vector */
784 motion_x[0] = prior_last_motion_x;
785 motion_y[0] = prior_last_motion_y;
786
787 /* vector maintenance */
788 prior_last_motion_x = last_motion_x;
789 prior_last_motion_y = last_motion_y;
790 last_motion_x = motion_x[0];
791 last_motion_y = motion_y[0];
792 break;
793
794 default:
795 /* covers intra, inter without MV, golden without MV */
796 motion_x[0] = 0;
797 motion_y[0] = 0;
798
799 /* no vector maintenance */
800 break;
801 }
802
803 /* assign the motion vectors to the correct fragments */
804 for (k = 0; k < 4; k++) {
805 current_fragment =
806 BLOCK_Y * s->fragment_width[0] + BLOCK_X;
807 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
808 s->motion_val[0][current_fragment][0] = motion_x[k];
809 s->motion_val[0][current_fragment][1] = motion_y[k];
810 } else {
811 s->motion_val[0][current_fragment][0] = motion_x[0];
812 s->motion_val[0][current_fragment][1] = motion_y[0];
813 }
814 }
815
816 if (s->chroma_y_shift) {
817 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
818 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] +
819 motion_x[2] + motion_x[3], 2);
820 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
821 motion_y[2] + motion_y[3], 2);
822 }
823 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
824 motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
825 frag = mb_y * s->fragment_width[1] + mb_x;
826 s->motion_val[1][frag][0] = motion_x[0];
827 s->motion_val[1][frag][1] = motion_y[0];
828 } else if (s->chroma_x_shift) {
829 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
830 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
831 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
832 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
833 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
834 } else {
835 motion_x[1] = motion_x[0];
836 motion_y[1] = motion_y[0];
837 }
838 motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
839 motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
840
841 frag = 2 * mb_y * s->fragment_width[1] + mb_x;
842 for (k = 0; k < 2; k++) {
843 s->motion_val[1][frag][0] = motion_x[k];
844 s->motion_val[1][frag][1] = motion_y[k];
845 frag += s->fragment_width[1];
846 }
847 } else {
848 for (k = 0; k < 4; k++) {
849 frag = BLOCK_Y * s->fragment_width[1] + BLOCK_X;
850 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
851 s->motion_val[1][frag][0] = motion_x[k];
852 s->motion_val[1][frag][1] = motion_y[k];
853 } else {
854 s->motion_val[1][frag][0] = motion_x[0];
855 s->motion_val[1][frag][1] = motion_y[0];
856 }
857 }
858 }
859 }
860 }
861 }
862
863 return 0;
864 }
865
866 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
867 {
868 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
869 int num_blocks = s->total_num_coded_frags;
870
871 for (qpi = 0; qpi < s->nqps - 1 && num_blocks > 0; qpi++) {
872 i = blocks_decoded = num_blocks_at_qpi = 0;
873
874 bit = get_bits1(gb) ^ 1;
875 run_length = 0;
876
877 do {
878 if (run_length == MAXIMUM_LONG_BIT_RUN)
879 bit = get_bits1(gb);
880 else
881 bit ^= 1;
882
883 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
884 if (run_length == 34)
885 run_length += get_bits(gb, 12);
886 blocks_decoded += run_length;
887
888 if (!bit)
889 num_blocks_at_qpi += run_length;
890
891 for (j = 0; j < run_length; i++) {
892 if (i >= s->total_num_coded_frags)
893 return -1;
894
895 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
896 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
897 j++;
898 }
899 }
900 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
901
902 num_blocks -= num_blocks_at_qpi;
903 }
904
905 return 0;
906 }
907
908 /*
909 * This function is called by unpack_dct_coeffs() to extract the VLCs from
910 * the bitstream. The VLCs encode tokens which are used to unpack DCT
911 * data. This function unpacks all the VLCs for either the Y plane or both
912 * C planes, and is called for DC coefficients or different AC coefficient
913 * levels (since different coefficient types require different VLC tables.
914 *
915 * This function returns a residual eob run. E.g, if a particular token gave
916 * instructions to EOB the next 5 fragments and there were only 2 fragments
917 * left in the current fragment range, 3 would be returned so that it could
918 * be passed into the next call to this same function.
919 */
920 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
921 VLC *table, int coeff_index,
922 int plane,
923 int eob_run)
924 {
925 int i, j = 0;
926 int token;
927 int zero_run = 0;
928 int16_t coeff = 0;
929 int bits_to_get;
930 int blocks_ended;
931 int coeff_i = 0;
932 int num_coeffs = s->num_coded_frags[plane][coeff_index];
933 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
934
935 /* local references to structure members to avoid repeated dereferences */
936 int *coded_fragment_list = s->coded_fragment_list[plane];
937 Vp3Fragment *all_fragments = s->all_fragments;
938 VLC_TYPE(*vlc_table)[2] = table->table;
939
940 if (num_coeffs < 0)
941 av_log(s->avctx, AV_LOG_ERROR,
942 "Invalid number of coefficients at level %d\n", coeff_index);
943
944 if (eob_run > num_coeffs) {
945 coeff_i =
946 blocks_ended = num_coeffs;
947 eob_run -= num_coeffs;
948 } else {
949 coeff_i =
950 blocks_ended = eob_run;
951 eob_run = 0;
952 }
953
954 // insert fake EOB token to cover the split between planes or zzi
955 if (blocks_ended)
956 dct_tokens[j++] = blocks_ended << 2;
957
958 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
959 /* decode a VLC into a token */
960 token = get_vlc2(gb, vlc_table, 11, 3);
961 /* use the token to get a zero run, a coefficient, and an eob run */
962 if ((unsigned) token <= 6U) {
963 eob_run = eob_run_base[token];
964 if (eob_run_get_bits[token])
965 eob_run += get_bits(gb, eob_run_get_bits[token]);
966
967 // record only the number of blocks ended in this plane,
968 // any spill will be recorded in the next plane.
969 if (eob_run > num_coeffs - coeff_i) {
970 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
971 blocks_ended += num_coeffs - coeff_i;
972 eob_run -= num_coeffs - coeff_i;
973 coeff_i = num_coeffs;
974 } else {
975 dct_tokens[j++] = TOKEN_EOB(eob_run);
976 blocks_ended += eob_run;
977 coeff_i += eob_run;
978 eob_run = 0;
979 }
980 } else if (token >= 0) {
981 bits_to_get = coeff_get_bits[token];
982 if (bits_to_get)
983 bits_to_get = get_bits(gb, bits_to_get);
984 coeff = coeff_tables[token][bits_to_get];
985
986 zero_run = zero_run_base[token];
987 if (zero_run_get_bits[token])
988 zero_run += get_bits(gb, zero_run_get_bits[token]);
989
990 if (zero_run) {
991 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
992 } else {
993 // Save DC into the fragment structure. DC prediction is
994 // done in raster order, so the actual DC can't be in with
995 // other tokens. We still need the token in dct_tokens[]
996 // however, or else the structure collapses on itself.
997 if (!coeff_index)
998 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
999
1000 dct_tokens[j++] = TOKEN_COEFF(coeff);
1001 }
1002
1003 if (coeff_index + zero_run > 64) {
1004 av_log(s->avctx, AV_LOG_DEBUG,
1005 "Invalid zero run of %d with %d coeffs left\n",
1006 zero_run, 64 - coeff_index);
1007 zero_run = 64 - coeff_index;
1008 }
1009
1010 // zero runs code multiple coefficients,
1011 // so don't try to decode coeffs for those higher levels
1012 for (i = coeff_index + 1; i <= coeff_index + zero_run; i++)
1013 s->num_coded_frags[plane][i]--;
1014 coeff_i++;
1015 } else {
1016 av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
1017 return -1;
1018 }
1019 }
1020
1021 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1022 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1023
1024 // decrement the number of blocks that have higher coefficients for each
1025 // EOB run at this level
1026 if (blocks_ended)
1027 for (i = coeff_index + 1; i < 64; i++)
1028 s->num_coded_frags[plane][i] -= blocks_ended;
1029
1030 // setup the next buffer
1031 if (plane < 2)
1032 s->dct_tokens[plane + 1][coeff_index] = dct_tokens + j;
1033 else if (coeff_index < 63)
1034 s->dct_tokens[0][coeff_index + 1] = dct_tokens + j;
1035
1036 return eob_run;
1037 }
1038
1039 static void reverse_dc_prediction(Vp3DecodeContext *s,
1040 int first_fragment,
1041 int fragment_width,
1042 int fragment_height);
1043 /*
1044 * This function unpacks all of the DCT coefficient data from the
1045 * bitstream.
1046 */
1047 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1048 {
1049 int i;
1050 int dc_y_table;
1051 int dc_c_table;
1052 int ac_y_table;
1053 int ac_c_table;
1054 int residual_eob_run = 0;
1055 VLC *y_tables[64];
1056 VLC *c_tables[64];
1057
1058 s->dct_tokens[0][0] = s->dct_tokens_base;
1059
1060 /* fetch the DC table indexes */
1061 dc_y_table = get_bits(gb, 4);
1062 dc_c_table = get_bits(gb, 4);
1063
1064 /* unpack the Y plane DC coefficients */
1065 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1066 0, residual_eob_run);
1067 if (residual_eob_run < 0)
1068 return residual_eob_run;
1069
1070 /* reverse prediction of the Y-plane DC coefficients */
1071 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1072
1073 /* unpack the C plane DC coefficients */
1074 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1075 1, residual_eob_run);
1076 if (residual_eob_run < 0)
1077 return residual_eob_run;
1078 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1079 2, residual_eob_run);
1080 if (residual_eob_run < 0)
1081 return residual_eob_run;
1082
1083 /* reverse prediction of the C-plane DC coefficients */
1084 if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1085 reverse_dc_prediction(s, s->fragment_start[1],
1086 s->fragment_width[1], s->fragment_height[1]);
1087 reverse_dc_prediction(s, s->fragment_start[2],
1088 s->fragment_width[1], s->fragment_height[1]);
1089 }
1090
1091 /* fetch the AC table indexes */
1092 ac_y_table = get_bits(gb, 4);
1093 ac_c_table = get_bits(gb, 4);
1094
1095 /* build tables of AC VLC tables */
1096 for (i = 1; i <= 5; i++) {
1097 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1098 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1099 }
1100 for (i = 6; i <= 14; i++) {
1101 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1102 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1103 }
1104 for (i = 15; i <= 27; i++) {
1105 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1106 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1107 }
1108 for (i = 28; i <= 63; i++) {
1109 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1110 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1111 }
1112
1113 /* decode all AC coefficients */
1114 for (i = 1; i <= 63; i++) {
1115 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1116 0, residual_eob_run);
1117 if (residual_eob_run < 0)
1118 return residual_eob_run;
1119
1120 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1121 1, residual_eob_run);
1122 if (residual_eob_run < 0)
1123 return residual_eob_run;
1124 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1125 2, residual_eob_run);
1126 if (residual_eob_run < 0)
1127 return residual_eob_run;
1128 }
1129
1130 return 0;
1131 }
1132
1133 /*
1134 * This function reverses the DC prediction for each coded fragment in
1135 * the frame. Much of this function is adapted directly from the original
1136 * VP3 source code.
1137 */
1138 #define COMPATIBLE_FRAME(x) \
1139 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1140 #define DC_COEFF(u) s->all_fragments[u].dc
1141
1142 static void reverse_dc_prediction(Vp3DecodeContext *s,
1143 int first_fragment,
1144 int fragment_width,
1145 int fragment_height)
1146 {
1147 #define PUL 8
1148 #define PU 4
1149 #define PUR 2
1150 #define PL 1
1151
1152 int x, y;
1153 int i = first_fragment;
1154
1155 int predicted_dc;
1156
1157 /* DC values for the left, up-left, up, and up-right fragments */
1158 int vl, vul, vu, vur;
1159
1160 /* indexes for the left, up-left, up, and up-right fragments */
1161 int l, ul, u, ur;
1162
1163 /*
1164 * The 6 fields mean:
1165 * 0: up-left multiplier
1166 * 1: up multiplier
1167 * 2: up-right multiplier
1168 * 3: left multiplier
1169 */
1170 static const int predictor_transform[16][4] = {
1171 { 0, 0, 0, 0 },
1172 { 0, 0, 0, 128 }, // PL
1173 { 0, 0, 128, 0 }, // PUR
1174 { 0, 0, 53, 75 }, // PUR|PL
1175 { 0, 128, 0, 0 }, // PU
1176 { 0, 64, 0, 64 }, // PU |PL
1177 { 0, 128, 0, 0 }, // PU |PUR
1178 { 0, 0, 53, 75 }, // PU |PUR|PL
1179 { 128, 0, 0, 0 }, // PUL
1180 { 0, 0, 0, 128 }, // PUL|PL
1181 { 64, 0, 64, 0 }, // PUL|PUR
1182 { 0, 0, 53, 75 }, // PUL|PUR|PL
1183 { 0, 128, 0, 0 }, // PUL|PU
1184 { -104, 116, 0, 116 }, // PUL|PU |PL
1185 { 24, 80, 24, 0 }, // PUL|PU |PUR
1186 { -104, 116, 0, 116 } // PUL|PU |PUR|PL
1187 };
1188
1189 /* This table shows which types of blocks can use other blocks for
1190 * prediction. For example, INTRA is the only mode in this table to
1191 * have a frame number of 0. That means INTRA blocks can only predict
1192 * from other INTRA blocks. There are 2 golden frame coding types;
1193 * blocks encoding in these modes can only predict from other blocks
1194 * that were encoded with these 1 of these 2 modes. */
1195 static const unsigned char compatible_frame[9] = {
1196 1, /* MODE_INTER_NO_MV */
1197 0, /* MODE_INTRA */
1198 1, /* MODE_INTER_PLUS_MV */
1199 1, /* MODE_INTER_LAST_MV */
1200 1, /* MODE_INTER_PRIOR_MV */
1201 2, /* MODE_USING_GOLDEN */
1202 2, /* MODE_GOLDEN_MV */
1203 1, /* MODE_INTER_FOUR_MV */
1204 3 /* MODE_COPY */
1205 };
1206 int current_frame_type;
1207
1208 /* there is a last DC predictor for each of the 3 frame types */
1209 short last_dc[3];
1210
1211 int transform = 0;
1212
1213 vul =
1214 vu =
1215 vur =
1216 vl = 0;
1217 last_dc[0] =
1218 last_dc[1] =
1219 last_dc[2] = 0;
1220
1221 /* for each fragment row... */
1222 for (y = 0; y < fragment_height; y++) {
1223 /* for each fragment in a row... */
1224 for (x = 0; x < fragment_width; x++, i++) {
1225
1226 /* reverse prediction if this block was coded */
1227 if (s->all_fragments[i].coding_method != MODE_COPY) {
1228 current_frame_type =
1229 compatible_frame[s->all_fragments[i].coding_method];
1230
1231 transform = 0;
1232 if (x) {
1233 l = i - 1;
1234 vl = DC_COEFF(l);
1235 if (COMPATIBLE_FRAME(l))
1236 transform |= PL;
1237 }
1238 if (y) {
1239 u = i - fragment_width;
1240 vu = DC_COEFF(u);
1241 if (COMPATIBLE_FRAME(u))
1242 transform |= PU;
1243 if (x) {
1244 ul = i - fragment_width - 1;
1245 vul = DC_COEFF(ul);
1246 if (COMPATIBLE_FRAME(ul))
1247 transform |= PUL;
1248 }
1249 if (x + 1 < fragment_width) {
1250 ur = i - fragment_width + 1;
1251 vur = DC_COEFF(ur);
1252 if (COMPATIBLE_FRAME(ur))
1253 transform |= PUR;
1254 }
1255 }
1256
1257 if (transform == 0) {
1258 /* if there were no fragments to predict from, use last
1259 * DC saved */
1260 predicted_dc = last_dc[current_frame_type];
1261 } else {
1262 /* apply the appropriate predictor transform */
1263 predicted_dc =
1264 (predictor_transform[transform][0] * vul) +
1265 (predictor_transform[transform][1] * vu) +
1266 (predictor_transform[transform][2] * vur) +
1267 (predictor_transform[transform][3] * vl);
1268
1269 predicted_dc /= 128;
1270
1271 /* check for outranging on the [ul u l] and
1272 * [ul u ur l] predictors */
1273 if ((transform == 15) || (transform == 13)) {
1274 if (FFABS(predicted_dc - vu) > 128)
1275 predicted_dc = vu;
1276 else if (FFABS(predicted_dc - vl) > 128)
1277 predicted_dc = vl;
1278 else if (FFABS(predicted_dc - vul) > 128)
1279 predicted_dc = vul;
1280 }
1281 }
1282
1283 /* at long last, apply the predictor */
1284 DC_COEFF(i) += predicted_dc;
1285 /* save the DC */
1286 last_dc[current_frame_type] = DC_COEFF(i);
1287 }
1288 }
1289 }
1290 }
1291
1292 static void apply_loop_filter(Vp3DecodeContext *s, int plane,
1293 int ystart, int yend)
1294 {
1295 int x, y;
1296 int *bounding_values = s->bounding_values_array + 127;
1297
1298 int width = s->fragment_width[!!plane];
1299 int height = s->fragment_height[!!plane];
1300 int fragment = s->fragment_start[plane] + ystart * width;
1301 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1302 uint8_t *plane_data = s->current_frame.f->data[plane];
1303 if (!s->flipped_image)
1304 stride = -stride;
1305 plane_data += s->data_offset[plane] + 8 * ystart * stride;
1306
1307 for (y = ystart; y < yend; y++) {
1308 for (x = 0; x < width; x++) {
1309 /* This code basically just deblocks on the edges of coded blocks.
1310 * However, it has to be much more complicated because of the
1311 * brain damaged deblock ordering used in VP3/Theora. Order matters
1312 * because some pixels get filtered twice. */
1313 if (s->all_fragments[fragment].coding_method != MODE_COPY) {
1314 /* do not perform left edge filter for left columns frags */
1315 if (x > 0) {
1316 s->vp3dsp.h_loop_filter(
1317 plane_data + 8 * x,
1318 stride, bounding_values);
1319 }
1320
1321 /* do not perform top edge filter for top row fragments */
1322 if (y > 0) {
1323 s->vp3dsp.v_loop_filter(
1324 plane_data + 8 * x,
1325 stride, bounding_values);
1326 }
1327
1328 /* do not perform right edge filter for right column
1329 * fragments or if right fragment neighbor is also coded
1330 * in this frame (it will be filtered in next iteration) */
1331 if ((x < width - 1) &&
1332 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1333 s->vp3dsp.h_loop_filter(
1334 plane_data + 8 * x + 8,
1335 stride, bounding_values);
1336 }
1337
1338 /* do not perform bottom edge filter for bottom row
1339 * fragments or if bottom fragment neighbor is also coded
1340 * in this frame (it will be filtered in the next row) */
1341 if ((y < height - 1) &&
1342 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1343 s->vp3dsp.v_loop_filter(
1344 plane_data + 8 * x + 8 * stride,
1345 stride, bounding_values);
1346 }
1347 }
1348
1349 fragment++;
1350 }
1351 plane_data += 8 * stride;
1352 }
1353 }
1354
1355 /**
1356 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1357 * for the next block in coding order
1358 */
1359 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1360 int plane, int inter, int16_t block[64])
1361 {
1362 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1363 uint8_t *perm = s->idct_scantable;
1364 int i = 0;
1365
1366 do {
1367 int token = *s->dct_tokens[plane][i];
1368 switch (token & 3) {
1369 case 0: // EOB
1370 if (--token < 4) // 0-3 are token types so the EOB run must now be 0
1371 s->dct_tokens[plane][i]++;
1372 else
1373 *s->dct_tokens[plane][i] = token & ~3;
1374 goto end;
1375 case 1: // zero run
1376 s->dct_tokens[plane][i]++;
1377 i += (token >> 2) & 0x7f;
1378 if (i > 63) {
1379 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1380 return i;
1381 }
1382 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1383 i++;
1384 break;
1385 case 2: // coeff
1386 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1387 s->dct_tokens[plane][i++]++;
1388 break;
1389 default: // shouldn't happen
1390 return i;
1391 }
1392 } while (i < 64);
1393 // return value is expected to be a valid level
1394 i--;
1395 end:
1396 // the actual DC+prediction is in the fragment structure
1397 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1398 return i;
1399 }
1400
1401 /**
1402 * called when all pixels up to row y are complete
1403 */
1404 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1405 {
1406 int h, cy, i;
1407 int offset[AV_NUM_DATA_POINTERS];
1408
1409 if (HAVE_THREADS && s->avctx->active_thread_type & FF_THREAD_FRAME) {
1410 int y_flipped = s->flipped_image ? s->height - y : y;
1411
1412 /* At the end of the frame, report INT_MAX instead of the height of
1413 * the frame. This makes the other threads' ff_thread_await_progress()
1414 * calls cheaper, because they don't have to clip their values. */
1415 ff_thread_report_progress(&s->current_frame,
1416 y_flipped == s->height ? INT_MAX
1417 : y_flipped - 1,
1418 0);
1419 }
1420
1421 if (!s->avctx->draw_horiz_band)
1422 return;
1423
1424 h = y - s->last_slice_end;
1425 s->last_slice_end = y;
1426 y -= h;
1427
1428 if (!s->flipped_image)
1429 y = s->height - y - h;
1430
1431 cy = y >> s->chroma_y_shift;
1432 offset[0] = s->current_frame.f->linesize[0] * y;
1433 offset[1] = s->current_frame.f->linesize[1] * cy;
1434 offset[2] = s->current_frame.f->linesize[2] * cy;
1435 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1436 offset[i] = 0;
1437
1438 emms_c();
1439 s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
1440 }
1441
1442 /**
1443 * Wait for the reference frame of the current fragment.
1444 * The progress value is in luma pixel rows.
1445 */
1446 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
1447 int motion_y, int y)
1448 {
1449 ThreadFrame *ref_frame;
1450 int ref_row;
1451 int border = motion_y & 1;
1452
1453 if (fragment->coding_method == MODE_USING_GOLDEN ||
1454 fragment->coding_method == MODE_GOLDEN_MV)
1455 ref_frame = &s->golden_frame;
1456 else
1457 ref_frame = &s->last_frame;
1458
1459 ref_row = y + (motion_y >> 1);
1460 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1461
1462 ff_thread_await_progress(ref_frame, ref_row, 0);
1463 }
1464
1465 /*
1466 * Perform the final rendering for a particular slice of data.
1467 * The slice number ranges from 0..(c_superblock_height - 1).
1468 */
1469 static void render_slice(Vp3DecodeContext *s, int slice)
1470 {
1471 int x, y, i, j, fragment;
1472 int16_t *block = s->block;
1473 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1474 int motion_halfpel_index;
1475 uint8_t *motion_source;
1476 int plane, first_pixel;
1477
1478 if (slice >= s->c_superblock_height)
1479 return;
1480
1481 for (plane = 0; plane < 3; plane++) {
1482 uint8_t *output_plane = s->current_frame.f->data[plane] +
1483 s->data_offset[plane];
1484 uint8_t *last_plane = s->last_frame.f->data[plane] +
1485 s->data_offset[plane];
1486 uint8_t *golden_plane = s->golden_frame.f->data[plane] +
1487 s->data_offset[plane];
1488 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1489 int plane_width = s->width >> (plane && s->chroma_x_shift);
1490 int plane_height = s->height >> (plane && s->chroma_y_shift);
1491 int8_t(*motion_val)[2] = s->motion_val[!!plane];
1492
1493 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1494 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1495 int slice_width = plane ? s->c_superblock_width
1496 : s->y_superblock_width;
1497
1498 int fragment_width = s->fragment_width[!!plane];
1499 int fragment_height = s->fragment_height[!!plane];
1500 int fragment_start = s->fragment_start[plane];
1501
1502 int do_await = !plane && HAVE_THREADS &&
1503 (s->avctx->active_thread_type & FF_THREAD_FRAME);
1504
1505 if (!s->flipped_image)
1506 stride = -stride;
1507 if (CONFIG_GRAY && plane && (s->avctx->flags & AV_CODEC_FLAG_GRAY))
1508 continue;
1509
1510 /* for each superblock row in the slice (both of them)... */
1511 for (; sb_y < slice_height; sb_y++) {
1512 /* for each superblock in a row... */
1513 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1514 /* for each block in a superblock... */
1515 for (j = 0; j < 16; j++) {
1516 x = 4 * sb_x + hilbert_offset[j][0];
1517 y = 4 * sb_y + hilbert_offset[j][1];
1518 fragment = y * fragment_width + x;
1519
1520 i = fragment_start + fragment;
1521
1522 // bounds check
1523 if (x >= fragment_width || y >= fragment_height)
1524 continue;
1525
1526 first_pixel = 8 * y * stride + 8 * x;
1527
1528 if (do_await &&
1529 s->all_fragments[i].coding_method != MODE_INTRA)
1530 await_reference_row(s, &s->all_fragments[i],
1531 motion_val[fragment][1],
1532 (16 * y) >> s->chroma_y_shift);
1533
1534 /* transform if this block was coded */
1535 if (s->all_fragments[i].coding_method != MODE_COPY) {
1536 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1537 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1538 motion_source = golden_plane;
1539 else
1540 motion_source = last_plane;
1541
1542 motion_source += first_pixel;
1543 motion_halfpel_index = 0;
1544
1545 /* sort out the motion vector if this fragment is coded
1546 * using a motion vector method */
1547 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1548 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1549 int src_x, src_y;
1550 motion_x = motion_val[fragment][0];
1551 motion_y = motion_val[fragment][1];
1552
1553 src_x = (motion_x >> 1) + 8 * x;
1554 src_y = (motion_y >> 1) + 8 * y;
1555
1556 motion_halfpel_index = motion_x & 0x01;
1557 motion_source += (motion_x >> 1);
1558
1559 motion_halfpel_index |= (motion_y & 0x01) << 1;
1560 motion_source += ((motion_y >> 1) * stride);
1561
1562 if (src_x < 0 || src_y < 0 ||
1563 src_x + 9 >= plane_width ||
1564 src_y + 9 >= plane_height) {
1565 uint8_t *temp = s->edge_emu_buffer;
1566 if (stride < 0)
1567 temp -= 8 * stride;
1568
1569 s->vdsp.emulated_edge_mc(temp, motion_source,
1570 stride, stride,
1571 9, 9, src_x, src_y,
1572 plane_width,
1573 plane_height);
1574 motion_source = temp;
1575 }
1576 }
1577
1578 /* first, take care of copying a block from either the
1579 * previous or the golden frame */
1580 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1581 /* Note, it is possible to implement all MC cases
1582 * with put_no_rnd_pixels_l2 which would look more
1583 * like the VP3 source but this would be slower as
1584 * put_no_rnd_pixels_tab is better optimized */
1585 if (motion_halfpel_index != 3) {
1586 s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1587 output_plane + first_pixel,
1588 motion_source, stride, 8);
1589 } else {
1590 /* d is 0 if motion_x and _y have the same sign,
1591 * else -1 */
1592 int d = (motion_x ^ motion_y) >> 31;
1593 s->vp3dsp.put_no_rnd_pixels_l2(output_plane + first_pixel,
1594 motion_source - d,
1595 motion_source + stride + 1 + d,
1596 stride, 8);
1597 }
1598 }
1599
1600 /* invert DCT and place (or add) in final output */
1601
1602 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1603 int index;
1604 index = vp3_dequant(s, s->all_fragments + i,
1605 plane, 0, block);
1606 if (index > 63)
1607 continue;
1608 s->vp3dsp.idct_put(output_plane + first_pixel,
1609 stride,
1610 block);
1611 } else {
1612 int index = vp3_dequant(s, s->all_fragments + i,
1613 plane, 1, block);
1614 if (index > 63)
1615 continue;
1616 if (index > 0) {
1617 s->vp3dsp.idct_add(output_plane + first_pixel,
1618 stride,
1619 block);
1620 } else {
1621 s->vp3dsp.idct_dc_add(output_plane + first_pixel,
1622 stride, block);
1623 }
1624 }
1625 } else {
1626 /* copy directly from the previous frame */
1627 s->hdsp.put_pixels_tab[1][0](
1628 output_plane + first_pixel,
1629 last_plane + first_pixel,
1630 stride, 8);
1631 }
1632 }
1633 }
1634
1635 // Filter up to the last row in the superblock row
1636 if (!s->skip_loop_filter)
1637 apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
1638 FFMIN(4 * sb_y + 3, fragment_height - 1));
1639 }
1640 }
1641
1642 /* this looks like a good place for slice dispatch... */
1643 /* algorithm:
1644 * if (slice == s->macroblock_height - 1)
1645 * dispatch (both last slice & 2nd-to-last slice);
1646 * else if (slice > 0)
1647 * dispatch (slice - 1);
1648 */
1649
1650 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) - 16,
1651 s->height - 16));
1652 }
1653
1654 /// Allocate tables for per-frame data in Vp3DecodeContext
1655 static av_cold int allocate_tables(AVCodecContext *avctx)
1656 {
1657 Vp3DecodeContext *s = avctx->priv_data;
1658 int y_fragment_count, c_fragment_count;
1659
1660 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1661 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1662
1663 s->superblock_coding = av_malloc(s->superblock_count);
1664 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1665
1666 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1667
1668 s->dct_tokens_base = av_malloc(64 * s->fragment_count *
1669 sizeof(*s->dct_tokens_base));
1670 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1671 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1672
1673 /* work out the block mapping tables */
1674 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1675 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1676
1677 if (!s->superblock_coding || !s->all_fragments ||
1678 !s->dct_tokens_base || !s->coded_fragment_list[0] ||
1679 !s->superblock_fragments || !s->macroblock_coding ||
1680 !s->motion_val[0] || !s->motion_val[1]) {
1681 vp3_decode_end(avctx);
1682 return -1;
1683 }
1684
1685 init_block_mapping(s);
1686
1687 return 0;
1688 }
1689
1690 static av_cold int init_frames(Vp3DecodeContext *s)
1691 {
1692 s->current_frame.f = av_frame_alloc();
1693 s->last_frame.f = av_frame_alloc();
1694 s->golden_frame.f = av_frame_alloc();
1695
1696 if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) {
1697 av_frame_free(&s->current_frame.f);
1698 av_frame_free(&s->last_frame.f);
1699 av_frame_free(&s->golden_frame.f);
1700 return AVERROR(ENOMEM);
1701 }
1702
1703 return 0;
1704 }
1705
1706 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1707 {
1708 Vp3DecodeContext *s = avctx->priv_data;
1709 int i, inter, plane, ret;
1710 int c_width;
1711 int c_height;
1712 int y_fragment_count, c_fragment_count;
1713
1714 ret = init_frames(s);
1715 if (ret < 0)
1716 return ret;
1717
1718 avctx->internal->allocate_progress = 1;
1719
1720 if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
1721 s->version = 0;
1722 else
1723 s->version = 1;
1724
1725 s->avctx = avctx;
1726 s->width = FFALIGN(avctx->coded_width, 16);
1727 s->height = FFALIGN(avctx->coded_height, 16);
1728 if (avctx->pix_fmt == AV_PIX_FMT_NONE)
1729 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1730 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1731 ff_hpeldsp_init(&s->hdsp, avctx->flags | AV_CODEC_FLAG_BITEXACT);
1732 ff_videodsp_init(&s->vdsp, 8);
1733 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
1734
1735 for (i = 0; i < 64; i++) {
1736 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
1737 s->idct_permutation[i] = TRANSPOSE(i);
1738 s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]);
1739 #undef TRANSPOSE
1740 }
1741
1742 /* initialize to an impossible value which will force a recalculation
1743 * in the first frame decode */
1744 for (i = 0; i < 3; i++)
1745 s->qps[i] = -1;
1746
1747 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift,
1748 &s->chroma_y_shift);
1749
1750 s->y_superblock_width = (s->width + 31) / 32;
1751 s->y_superblock_height = (s->height + 31) / 32;
1752 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1753
1754 /* work out the dimensions for the C planes */
1755 c_width = s->width >> s->chroma_x_shift;
1756 c_height = s->height >> s->chroma_y_shift;
1757 s->c_superblock_width = (c_width + 31) / 32;
1758 s->c_superblock_height = (c_height + 31) / 32;
1759 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1760
1761 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1762 s->u_superblock_start = s->y_superblock_count;
1763 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1764
1765 s->macroblock_width = (s->width + 15) / 16;
1766 s->macroblock_height = (s->height + 15) / 16;
1767 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1768
1769 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1770 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1771 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1772 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1773
1774 /* fragment count covers all 8x8 blocks for all 3 planes */
1775 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1776 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1777 s->fragment_count = y_fragment_count + 2 * c_fragment_count;
1778 s->fragment_start[1] = y_fragment_count;
1779 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1780
1781 if (!s->theora_tables) {
1782 for (i = 0; i < 64; i++) {
1783 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1784 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1785 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1786 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1787 s->base_matrix[2][i] = vp31_inter_dequant[i];
1788 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1789 }
1790
1791 for (inter = 0; inter < 2; inter++) {
1792 for (plane = 0; plane < 3; plane++) {
1793 s->qr_count[inter][plane] = 1;
1794 s->qr_size[inter][plane][0] = 63;
1795 s->qr_base[inter][plane][0] =
1796 s->qr_base[inter][plane][1] = 2 * inter + (!!plane) * !inter;
1797 }
1798 }
1799
1800 /* init VLC tables */
1801 for (i = 0; i < 16; i++) {
1802 /* DC histograms */
1803 init_vlc(&s->dc_vlc[i], 11, 32,
1804 &dc_bias[i][0][1], 4, 2,
1805 &dc_bias[i][0][0], 4, 2, 0);
1806
1807 /* group 1 AC histograms */
1808 init_vlc(&s->ac_vlc_1[i], 11, 32,
1809 &ac_bias_0[i][0][1], 4, 2,
1810 &ac_bias_0[i][0][0], 4, 2, 0);
1811
1812 /* group 2 AC histograms */
1813 init_vlc(&s->ac_vlc_2[i], 11, 32,
1814 &ac_bias_1[i][0][1], 4, 2,
1815 &ac_bias_1[i][0][0], 4, 2, 0);
1816
1817 /* group 3 AC histograms */
1818 init_vlc(&s->ac_vlc_3[i], 11, 32,
1819 &ac_bias_2[i][0][1], 4, 2,
1820 &ac_bias_2[i][0][0], 4, 2, 0);
1821
1822 /* group 4 AC histograms */
1823 init_vlc(&s->ac_vlc_4[i], 11, 32,
1824 &ac_bias_3[i][0][1], 4, 2,
1825 &ac_bias_3[i][0][0], 4, 2, 0);
1826 }
1827 } else {
1828 for (i = 0; i < 16; i++) {
1829 /* DC histograms */
1830 if (init_vlc(&s->dc_vlc[i], 11, 32,
1831 &s->huffman_table[i][0][1], 8, 4,
1832 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1833 goto vlc_fail;
1834
1835 /* group 1 AC histograms */
1836 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1837 &s->huffman_table[i + 16][0][1], 8, 4,
1838 &s->huffman_table[i + 16][0][0], 8, 4, 0) < 0)
1839 goto vlc_fail;
1840
1841 /* group 2 AC histograms */
1842 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1843 &s->huffman_table[i + 16 * 2][0][1], 8, 4,
1844 &s->huffman_table[i + 16 * 2][0][0], 8, 4, 0) < 0)
1845 goto vlc_fail;
1846
1847 /* group 3 AC histograms */
1848 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1849 &s->huffman_table[i + 16 * 3][0][1], 8, 4,
1850 &s->huffman_table[i + 16 * 3][0][0], 8, 4, 0) < 0)
1851 goto vlc_fail;
1852
1853 /* group 4 AC histograms */
1854 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1855 &s->huffman_table[i + 16 * 4][0][1], 8, 4,
1856 &s->huffman_table[i + 16 * 4][0][0], 8, 4, 0) < 0)
1857 goto vlc_fail;
1858 }
1859 }
1860
1861 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1862 &superblock_run_length_vlc_table[0][1], 4, 2,
1863 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1864
1865 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1866 &fragment_run_length_vlc_table[0][1], 4, 2,
1867 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1868
1869 init_vlc(&s->mode_code_vlc, 3, 8,
1870 &mode_code_vlc_table[0][1], 2, 1,
1871 &mode_code_vlc_table[0][0], 2, 1, 0);
1872
1873 init_vlc(&s->motion_vector_vlc, 6, 63,
1874 &motion_vector_vlc_table[0][1], 2, 1,
1875 &motion_vector_vlc_table[0][0], 2, 1, 0);
1876
1877 return allocate_tables(avctx);
1878
1879 vlc_fail:
1880 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1881 return -1;
1882 }
1883
1884 /// Release and shuffle frames after decode finishes
1885 static int update_frames(AVCodecContext *avctx)
1886 {
1887 Vp3DecodeContext *s = avctx->priv_data;
1888 int ret = 0;
1889
1890 /* shuffle frames (last = current) */
1891 ff_thread_release_buffer(avctx, &s->last_frame);
1892 ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
1893 if (ret < 0)
1894 goto fail;
1895
1896 if (s->keyframe) {
1897 ff_thread_release_buffer(avctx, &s->golden_frame);
1898 ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
1899 }
1900
1901 fail:
1902 ff_thread_release_buffer(avctx, &s->current_frame);
1903 return ret;
1904 }
1905
1906 static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
1907 {
1908 ff_thread_release_buffer(s->avctx, dst);
1909 if (src->f->data[0])
1910 return ff_thread_ref_frame(dst, src);
1911 return 0;
1912 }
1913
1914 static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
1915 {
1916 int ret;
1917 if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
1918 (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 ||
1919 (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0)
1920 return ret;
1921 return 0;
1922 }
1923
1924 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1925 {
1926 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1927 int qps_changed = 0, i, err;
1928
1929 #define copy_fields(to, from, start_field, end_field) \
1930 memcpy(&to->start_field, &from->start_field, \
1931 (char *) &to->end_field - (char *) &to->start_field)
1932
1933 if (!s1->current_frame.f->data[0] ||
1934 s->width != s1->width || s->height != s1->height) {
1935 if (s != s1)
1936 ref_frames(s, s1);
1937 return -1;
1938 }
1939
1940 if (s != s1) {
1941 // init tables if the first frame hasn't been decoded
1942 if (!s->current_frame.f->data[0]) {
1943 int y_fragment_count, c_fragment_count;
1944 s->avctx = dst;
1945 err = allocate_tables(dst);
1946 if (err)
1947 return err;
1948 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1949 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1950 memcpy(s->motion_val[0], s1->motion_val[0],
1951 y_fragment_count * sizeof(*s->motion_val[0]));
1952 memcpy(s->motion_val[1], s1->motion_val[1],
1953 c_fragment_count * sizeof(*s->motion_val[1]));
1954 }
1955
1956 // copy previous frame data
1957 if ((err = ref_frames(s, s1)) < 0)
1958 return err;
1959
1960 s->keyframe = s1->keyframe;
1961
1962 // copy qscale data if necessary
1963 for (i = 0; i < 3; i++) {
1964 if (s->qps[i] != s1->qps[1]) {
1965 qps_changed = 1;
1966 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1967 }
1968 }
1969
1970 if (s->qps[0] != s1->qps[0])
1971 memcpy(&s->bounding_values_array, &s1->bounding_values_array,
1972 sizeof(s->bounding_values_array));
1973
1974 if (qps_changed)
1975 copy_fields(s, s1, qps, superblock_count);
1976 #undef copy_fields
1977 }
1978
1979 return update_frames(dst);
1980 }
1981
1982 static int vp3_decode_frame(AVCodecContext *avctx,
1983 void *data, int *got_frame,
1984 AVPacket *avpkt)
1985 {
1986 AVFrame *frame = data;
1987 const uint8_t *buf = avpkt->data;
1988 int buf_size = avpkt->size;
1989 Vp3DecodeContext *s = avctx->priv_data;
1990 GetBitContext gb;
1991 int i, ret;
1992
1993 init_get_bits(&gb, buf, buf_size * 8);
1994
1995 if (s->theora && get_bits1(&gb)) {
1996 av_log(avctx, AV_LOG_ERROR,
1997 "Header packet passed to frame decoder, skipping\n");
1998 return -1;
1999 }
2000
2001 s->keyframe = !get_bits1(&gb);
2002 if (!s->theora)
2003 skip_bits(&gb, 1);
2004 for (i = 0; i < 3; i++)
2005 s->last_qps[i] = s->qps[i];
2006
2007 s->nqps = 0;
2008 do {
2009 s->qps[s->nqps++] = get_bits(&gb, 6);
2010 } while (s->theora >= 0x030200 && s->nqps < 3 && get_bits1(&gb));
2011 for (i = s->nqps; i < 3; i++)
2012 s->qps[i] = -1;
2013
2014 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2015 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2016 s->keyframe ? "key" : "", avctx->frame_number + 1, s->qps[0]);
2017
2018 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
2019 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL
2020 : AVDISCARD_NONKEY);
2021
2022 if (s->qps[0] != s->last_qps[0])
2023 init_loop_filter(s);
2024
2025 for (i = 0; i < s->nqps; i++)
2026 // reinit all dequantizers if the first one changed, because
2027 // the DC of the first quantizer must be used for all matrices
2028 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
2029 init_dequantizer(s, i);
2030
2031 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2032 return buf_size;
2033
2034 s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
2035 : AV_PICTURE_TYPE_P;
2036 if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) {
2037 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2038 goto error;
2039 }
2040
2041 if (!s->edge_emu_buffer)
2042 s->edge_emu_buffer = av_malloc(9 * FFABS(s->current_frame.f->linesize[0]));
2043
2044 if (s->keyframe) {
2045 if (!s->theora) {
2046 skip_bits(&gb, 4); /* width code */
2047 skip_bits(&gb, 4); /* height code */
2048 if (s->version) {
2049 s->version = get_bits(&gb, 5);
2050 if (avctx->frame_number == 0)
2051 av_log(s->avctx, AV_LOG_DEBUG,
2052 "VP version: %d\n", s->version);
2053 }
2054 }
2055 if (s->version || s->theora) {
2056 if (get_bits1(&gb))
2057 av_log(s->avctx, AV_LOG_ERROR,
2058 "Warning, unsupported keyframe coding type?!\n");
2059 skip_bits(&gb, 2); /* reserved? */
2060 }
2061 } else {
2062 if (!s->golden_frame.f->data[0]) {
2063 av_log(s->avctx, AV_LOG_WARNING,
2064 "vp3: first frame not a keyframe\n");
2065
2066 s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2067 if (ff_thread_get_buffer(avctx, &s->golden_frame,
2068 AV_GET_BUFFER_FLAG_REF) < 0) {
2069 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2070 goto error;
2071 }
2072 ff_thread_release_buffer(avctx, &s->last_frame);
2073 if ((ret = ff_thread_ref_frame(&s->last_frame,
2074 &s->golden_frame)) < 0)
2075 goto error;
2076 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2077 }
2078 }
2079
2080 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2081 ff_thread_finish_setup(avctx);
2082
2083 if (unpack_superblocks(s, &gb)) {
2084 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2085 goto error;
2086 }
2087 if (unpack_modes(s, &gb)) {
2088 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2089 goto error;
2090 }
2091 if (unpack_vectors(s, &gb)) {
2092 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2093 goto error;
2094 }
2095 if (unpack_block_qpis(s, &gb)) {
2096 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2097 goto error;
2098 }
2099 if (unpack_dct_coeffs(s, &gb)) {
2100 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2101 goto error;
2102 }
2103
2104 for (i = 0; i < 3; i++) {
2105 int height = s->height >> (i && s->chroma_y_shift);
2106 if (s->flipped_image)
2107 s->data_offset[i] = 0;
2108 else
2109 s->data_offset[i] = (height - 1) * s->current_frame.f->linesize[i];
2110 }
2111
2112 s->last_slice_end = 0;
2113 for (i = 0; i < s->c_superblock_height; i++)
2114 render_slice(s, i);
2115
2116 // filter the last row
2117 for (i = 0; i < 3; i++) {
2118 int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
2119 apply_loop_filter(s, i, row, row + 1);
2120 }
2121 vp3_draw_horiz_band(s, s->height);
2122
2123 /* output frame, offset as needed */
2124 if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2125 return ret;
2126
2127 frame->crop_left = s->offset_x;
2128 frame->crop_right = avctx->coded_width - avctx->width - s->offset_x;
2129 frame->crop_top = s->offset_y;
2130 frame->crop_bottom = avctx->coded_height - avctx->height - s->offset_y;
2131
2132 *got_frame = 1;
2133
2134 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME)) {
2135 ret = update_frames(avctx);
2136 if (ret < 0)
2137 return ret;
2138 }
2139
2140 return buf_size;
2141
2142 error:
2143 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2144
2145 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_FRAME))
2146 av_frame_unref(s->current_frame.f);
2147
2148 return -1;
2149 }
2150
2151 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2152 {
2153 Vp3DecodeContext *s = avctx->priv_data;
2154
2155 if (get_bits1(gb)) {
2156 int token;
2157 if (s->entries >= 32) { /* overflow */
2158 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2159 return -1;
2160 }
2161 token = get_bits(gb, 5);
2162 ff_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2163 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2164 s->huffman_table[s->hti][token][0] = s->hbits;
2165 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2166 s->entries++;
2167 } else {
2168 if (s->huff_code_size >= 32) { /* overflow */
2169 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2170 return -1;
2171 }
2172 s->huff_code_size++;
2173 s->hbits <<= 1;
2174 if (read_huffman_tree(avctx, gb))
2175 return -1;
2176 s->hbits |= 1;
2177 if (read_huffman_tree(avctx, gb))
2178 return -1;
2179 s->hbits >>= 1;
2180 s->huff_code_size--;
2181 }
2182 return 0;
2183 }
2184
2185 static int vp3_init_thread_copy(AVCodecContext *avctx)
2186 {
2187 Vp3DecodeContext *s = avctx->priv_data;
2188
2189 s->superblock_coding = NULL;
2190 s->all_fragments = NULL;
2191 s->coded_fragment_list[0] = NULL;
2192 s->dct_tokens_base = NULL;
2193 s->superblock_fragments = NULL;
2194 s->macroblock_coding = NULL;
2195 s->motion_val[0] = NULL;
2196 s->motion_val[1] = NULL;
2197 s->edge_emu_buffer = NULL;
2198
2199 return init_frames(s);
2200 }
2201
2202 #if CONFIG_THEORA_DECODER
2203 static const enum AVPixelFormat theora_pix_fmts[4] = {
2204 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2205 };
2206
2207 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2208 {
2209 Vp3DecodeContext *s = avctx->priv_data;
2210 int visible_width, visible_height, colorspace;
2211 uint8_t offset_x = 0, offset_y = 0;
2212 int ret;
2213 AVRational fps, aspect;
2214
2215 s->theora = get_bits_long(gb, 24);
2216 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2217
2218 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3
2219 * but previous versions have the image flipped relative to vp3 */
2220 if (s->theora < 0x030200) {
2221 s->flipped_image = 1;
2222 av_log(avctx, AV_LOG_DEBUG,
2223 "Old (<alpha3) Theora bitstream, flipped image\n");
2224 }
2225
2226 visible_width =
2227 s->width = get_bits(gb, 16) << 4;
2228 visible_height =
2229 s->height = get_bits(gb, 16) << 4;
2230
2231 if (s->theora >= 0x030200) {
2232 visible_width = get_bits_long(gb, 24);
2233 visible_height = get_bits_long(gb, 24);
2234
2235 offset_x = get_bits(gb, 8); /* offset x */
2236 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2237 }
2238
2239 /* sanity check */
2240 if (av_image_check_size(visible_width, visible_height, 0, avctx) < 0 ||
2241 visible_width + offset_x > s->width ||
2242 visible_height + offset_y > s->height) {
2243 av_log(s, AV_LOG_ERROR,
2244 "Invalid frame dimensions - w:%d h:%d x:%d y:%d (%dx%d).\n",
2245 visible_width, visible_height, offset_x, offset_y,
2246 s->width, s->height);
2247 return AVERROR_INVALIDDATA;
2248 }
2249
2250 fps.num = get_bits_long(gb, 32);
2251 fps.den = get_bits_long(gb, 32);
2252 if (fps.num && fps.den) {
2253 if (fps.num < 0 || fps.den < 0) {
2254 av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2255 return AVERROR_INVALIDDATA;
2256 }
2257 av_reduce(&avctx->framerate.den, &avctx->framerate.num,
2258 fps.den, fps.num, 1 << 30);
2259 }
2260
2261 aspect.num = get_bits_long(gb, 24);
2262 aspect.den = get_bits_long(gb, 24);
2263 if (aspect.num && aspect.den) {
2264 av_reduce(&avctx->sample_aspect_ratio.num,
2265 &avctx->sample_aspect_ratio.den,
2266 aspect.num, aspect.den, 1 << 30);
2267 ff_set_sar(avctx, avctx->sample_aspect_ratio);
2268 }
2269
2270 if (s->theora < 0x030200)
2271 skip_bits(gb, 5); /* keyframe frequency force */
2272 colorspace = get_bits(gb, 8);
2273 skip_bits(gb, 24); /* bitrate */
2274
2275 skip_bits(gb, 6); /* quality hint */
2276
2277 if (s->theora >= 0x030200) {
2278 skip_bits(gb, 5); /* keyframe frequency force */
2279 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2280 skip_bits(gb, 3); /* reserved */
2281 }
2282
2283 ret = ff_set_dimensions(avctx, s->width, s->height);
2284 if (ret < 0)
2285 return ret;
2286 if (!(avctx->flags2 & AV_CODEC_FLAG2_IGNORE_CROP) &&
2287 (visible_width != s->width || visible_height != s->height)) {
2288 avctx->width = visible_width;
2289 avctx->height = visible_height;
2290 // translate offsets from theora axis ([0,0] lower left)
2291 // to normal axis ([0,0] upper left)
2292 s->offset_x = offset_x;
2293 s->offset_y = s->height - visible_height - offset_y;
2294 }
2295
2296 if (colorspace == 1)
2297 avctx->color_primaries = AVCOL_PRI_BT470M;
2298 else if (colorspace == 2)
2299 avctx->color_primaries = AVCOL_PRI_BT470BG;
2300
2301 if (colorspace == 1 || colorspace == 2) {
2302 avctx->colorspace = AVCOL_SPC_BT470BG;
2303 avctx->color_trc = AVCOL_TRC_BT709;
2304 }
2305
2306 return 0;
2307 }
2308
2309 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2310 {
2311 Vp3DecodeContext *s = avctx->priv_data;
2312 int i, n, matrices, inter, plane;
2313
2314 if (s->theora >= 0x030200) {
2315 n = get_bits(gb, 3);
2316 /* loop filter limit values table */
2317 if (n)
2318 for (i = 0; i < 64; i++)
2319 s->filter_limit_values[i] = get_bits(gb, n);
2320 }
2321
2322 if (s->theora >= 0x030200)
2323 n = get_bits(gb, 4) + 1;
2324 else
2325 n = 16;
2326 /* quality threshold table */
2327 for (i = 0; i < 64; i++)
2328 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2329
2330 if (s->theora >= 0x030200)
2331 n = get_bits(gb, 4) + 1;
2332 else
2333 n = 16;
2334 /* dc scale factor table */
2335 for (i = 0; i < 64; i++)
2336 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2337
2338 if (s->theora >= 0x030200)
2339 matrices = get_bits(gb, 9) + 1;
2340 else
2341 matrices = 3;
2342
2343 if (matrices > 384) {
2344 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2345 return -1;
2346 }
2347
2348 for (n = 0; n < matrices; n++)
2349 for (i = 0; i < 64; i++)
2350 s->base_matrix[n][i] = get_bits(gb, 8);
2351
2352 for (inter = 0; inter <= 1; inter++) {
2353 for (plane = 0; plane <= 2; plane++) {
2354 int newqr = 1;
2355 if (inter || plane > 0)
2356 newqr = get_bits1(gb);
2357 if (!newqr) {
2358 int qtj, plj;
2359 if (inter && get_bits1(gb)) {
2360 qtj = 0;
2361 plj = plane;
2362 } else {
2363 qtj = (3 * inter + plane - 1) / 3;
2364 plj = (plane + 2) % 3;
2365 }
2366 s->qr_count[inter][plane] = s->qr_count[qtj][plj];
2367 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj],
2368 sizeof(s->qr_size[0][0]));
2369 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj],
2370 sizeof(s->qr_base[0][0]));
2371 } else {
2372 int qri = 0;
2373 int qi = 0;
2374
2375 for (;;) {
2376 i = get_bits(gb, av_log2(matrices - 1) + 1);
2377 if (i >= matrices) {
2378 av_log(avctx, AV_LOG_ERROR,
2379 "invalid base matrix index\n");
2380 return -1;
2381 }
2382 s->qr_base[inter][plane][qri] = i;
2383 if (qi >= 63)
2384 break;
2385 i = get_bits(gb, av_log2(63 - qi) + 1) + 1;
2386 s->qr_size[inter][plane][qri++] = i;
2387 qi += i;
2388 }
2389
2390 if (qi > 63) {
2391 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2392 return -1;
2393 }
2394 s->qr_count[inter][plane] = qri;
2395 }
2396 }
2397 }
2398
2399 /* Huffman tables */
2400 for (s->hti = 0; s->hti < 80; s->hti++) {
2401 s->entries = 0;
2402 s->huff_code_size = 1;
2403 if (!get_bits1(gb)) {
2404 s->hbits = 0;
2405 if (read_huffman_tree(avctx, gb))
2406 return -1;
2407 s->hbits = 1;
2408 if (read_huffman_tree(avctx, gb))
2409 return -1;
2410 }
2411 }
2412
2413 s->theora_tables = 1;
2414
2415 return 0;
2416 }
2417
2418 static av_cold int theora_decode_init(AVCodecContext *avctx)
2419 {
2420 Vp3DecodeContext *s = avctx->priv_data;
2421 GetBitContext gb;
2422 int ptype;
2423 uint8_t *header_start[3];
2424 int header_len[3];
2425 int i;
2426
2427 s->theora = 1;
2428
2429 if (!avctx->extradata_size) {
2430 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2431 return -1;
2432 }
2433
2434 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2435 42, header_start, header_len) < 0) {
2436 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2437 return -1;
2438 }
2439
2440 for (i = 0; i < 3; i++) {
2441 if (header_len[i] <= 0)
2442 continue;
2443 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2444
2445 ptype = get_bits(&gb, 8);
2446
2447 if (!(ptype & 0x80)) {
2448 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2449 // return -1;
2450 }
2451
2452 // FIXME: Check for this as well.
2453 skip_bits_long(&gb, 6 * 8); /* "theora" */
2454
2455 switch (ptype) {
2456 case 0x80:
2457 theora_decode_header(avctx, &gb);
2458 break;
2459 case 0x81:
2460 // FIXME: is this needed? it breaks sometimes
2461 // theora_decode_comments(avctx, gb);
2462 break;
2463 case 0x82:
2464 if (theora_decode_tables(avctx, &gb))
2465 return -1;
2466 break;
2467 default:
2468 av_log(avctx, AV_LOG_ERROR,
2469 "Unknown Theora config packet: %d\n", ptype & ~0x80);
2470 break;
2471 }
2472 if (ptype != 0x81 && 8 * header_len[i] != get_bits_count(&gb))
2473 av_log(avctx, AV_LOG_WARNING,
2474 "%d bits left in packet %X\n",
2475 8 * header_len[i] - get_bits_count(&gb), ptype);
2476 if (s->theora < 0x030200)
2477 break;
2478 }
2479
2480 return vp3_decode_init(avctx);
2481 }
2482
2483 AVCodec ff_theora_decoder = {
2484 .name = "theora",
2485 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2486 .type = AVMEDIA_TYPE_VIDEO,
2487 .id = AV_CODEC_ID_THEORA,
2488 .priv_data_size = sizeof(Vp3DecodeContext),
2489 .init = theora_decode_init,
2490 .close = vp3_decode_end,
2491 .decode = vp3_decode_frame,
2492 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
2493 AV_CODEC_CAP_FRAME_THREADS,
2494 .flush = vp3_decode_flush,
2495 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2496 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
2497 .caps_internal = FF_CODEC_CAP_EXPORTS_CROPPING,
2498 };
2499 #endif
2500
2501 AVCodec ff_vp3_decoder = {
2502 .name = "vp3",
2503 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2504 .type = AVMEDIA_TYPE_VIDEO,
2505 .id = AV_CODEC_ID_VP3,
2506 .priv_data_size = sizeof(Vp3DecodeContext),
2507 .init = vp3_decode_init,
2508 .close = vp3_decode_end,
2509 .decode = vp3_decode_frame,
2510 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DRAW_HORIZ_BAND |
2511 AV_CODEC_CAP_FRAME_THREADS,
2512 .flush = vp3_decode_flush,
2513 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2514 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
2515 };
Modern FFMPEG