8b7ff42668813afcc87b91fa4aecc37310b325d8
[libav.git] / libavcodec / vp3.c
1 /*
2 * Copyright (C) 2003-2004 the ffmpeg project
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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 libavcodec/vp3.c
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 "avcodec.h"
37 #include "dsputil.h"
38 #include "get_bits.h"
39
40 #include "vp3data.h"
41 #include "xiph.h"
42
43 #define FRAGMENT_PIXELS 8
44
45 static av_cold int vp3_decode_end(AVCodecContext *avctx);
46
47 typedef struct Coeff {
48 struct Coeff *next;
49 DCTELEM coeff;
50 uint8_t index;
51 } Coeff;
52
53 //FIXME split things out into their own arrays
54 typedef struct Vp3Fragment {
55 Coeff *next_coeff;
56 uint8_t coding_method;
57 int8_t motion_x;
58 int8_t motion_y;
59 uint8_t qpi;
60 } Vp3Fragment;
61
62 #define SB_NOT_CODED 0
63 #define SB_PARTIALLY_CODED 1
64 #define SB_FULLY_CODED 2
65
66 #define MODE_INTER_NO_MV 0
67 #define MODE_INTRA 1
68 #define MODE_INTER_PLUS_MV 2
69 #define MODE_INTER_LAST_MV 3
70 #define MODE_INTER_PRIOR_LAST 4
71 #define MODE_USING_GOLDEN 5
72 #define MODE_GOLDEN_MV 6
73 #define MODE_INTER_FOURMV 7
74 #define CODING_MODE_COUNT 8
75
76 /* special internal mode */
77 #define MODE_COPY 8
78
79 /* There are 6 preset schemes, plus a free-form scheme */
80 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
81 {
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
120 #define MIN_DEQUANT_VAL 2
121
122 typedef struct Vp3DecodeContext {
123 AVCodecContext *avctx;
124 int theora, theora_tables;
125 int version;
126 int width, height;
127 AVFrame golden_frame;
128 AVFrame last_frame;
129 AVFrame current_frame;
130 int keyframe;
131 DSPContext dsp;
132 int flipped_image;
133 int last_slice_end;
134
135 int qps[3];
136 int nqps;
137 int last_qps[3];
138
139 int superblock_count;
140 int y_superblock_width;
141 int y_superblock_height;
142 int c_superblock_width;
143 int c_superblock_height;
144 int u_superblock_start;
145 int v_superblock_start;
146 unsigned char *superblock_coding;
147
148 int macroblock_count;
149 int macroblock_width;
150 int macroblock_height;
151
152 int fragment_count;
153 int fragment_width;
154 int fragment_height;
155
156 Vp3Fragment *all_fragments;
157 uint8_t *coeff_counts;
158 Coeff *coeffs;
159 Coeff *next_coeff;
160 int fragment_start[3];
161 int data_offset[3];
162
163 ScanTable scantable;
164
165 /* tables */
166 uint16_t coded_dc_scale_factor[64];
167 uint32_t coded_ac_scale_factor[64];
168 uint8_t base_matrix[384][64];
169 uint8_t qr_count[2][3];
170 uint8_t qr_size [2][3][64];
171 uint16_t qr_base[2][3][64];
172
173 /* this is a list of indexes into the all_fragments array indicating
174 * which of the fragments are coded */
175 int *coded_fragment_list;
176 int coded_fragment_list_index;
177
178 /* track which fragments have already been decoded; called 'fast'
179 * because this data structure avoids having to iterate through every
180 * fragment in coded_fragment_list; once a fragment has been fully
181 * decoded, it is removed from this list */
182 int *fast_fragment_list;
183 int fragment_list_y_head;
184 int fragment_list_c_head;
185
186 VLC dc_vlc[16];
187 VLC ac_vlc_1[16];
188 VLC ac_vlc_2[16];
189 VLC ac_vlc_3[16];
190 VLC ac_vlc_4[16];
191
192 VLC superblock_run_length_vlc;
193 VLC fragment_run_length_vlc;
194 VLC mode_code_vlc;
195 VLC motion_vector_vlc;
196
197 /* these arrays need to be on 16-byte boundaries since SSE2 operations
198 * index into them */
199 DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
200
201 /* This table contains superblock_count * 16 entries. Each set of 16
202 * numbers corresponds to the fragment indexes 0..15 of the superblock.
203 * An entry will be -1 to indicate that no entry corresponds to that
204 * index. */
205 int *superblock_fragments;
206
207 /* This is an array that indicates how a particular macroblock
208 * is coded. */
209 unsigned char *macroblock_coding;
210
211 int first_coded_y_fragment;
212 int first_coded_c_fragment;
213 int last_coded_y_fragment;
214 int last_coded_c_fragment;
215
216 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
217 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
218
219 /* Huffman decode */
220 int hti;
221 unsigned int hbits;
222 int entries;
223 int huff_code_size;
224 uint16_t huffman_table[80][32][2];
225
226 uint8_t filter_limit_values[64];
227 DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
228 } Vp3DecodeContext;
229
230 /************************************************************************
231 * VP3 specific functions
232 ************************************************************************/
233
234 /*
235 * This function sets up all of the various blocks mappings:
236 * superblocks <-> fragments, macroblocks <-> fragments,
237 * superblocks <-> macroblocks
238 *
239 * Returns 0 is successful; returns 1 if *anything* went wrong.
240 */
241 static int init_block_mapping(Vp3DecodeContext *s)
242 {
243 int i, j;
244 signed int hilbert_walk_mb[4];
245
246 int current_fragment = 0;
247 int current_width = 0;
248 int current_height = 0;
249 int right_edge = 0;
250 int bottom_edge = 0;
251 int superblock_row_inc = 0;
252 int mapping_index = 0;
253
254 int current_macroblock;
255 int c_fragment;
256
257 static const signed char travel_width[16] = {
258 1, 1, 0, -1,
259 0, 0, 1, 0,
260 1, 0, 1, 0,
261 0, -1, 0, 1
262 };
263
264 static const signed char travel_height[16] = {
265 0, 0, 1, 0,
266 1, 1, 0, -1,
267 0, 1, 0, -1,
268 -1, 0, -1, 0
269 };
270
271 hilbert_walk_mb[0] = 1;
272 hilbert_walk_mb[1] = s->macroblock_width;
273 hilbert_walk_mb[2] = 1;
274 hilbert_walk_mb[3] = -s->macroblock_width;
275
276 /* iterate through each superblock (all planes) and map the fragments */
277 for (i = 0; i < s->superblock_count; i++) {
278 /* time to re-assign the limits? */
279 if (i == 0) {
280
281 /* start of Y superblocks */
282 right_edge = s->fragment_width;
283 bottom_edge = s->fragment_height;
284 current_width = -1;
285 current_height = 0;
286 superblock_row_inc = 3 * s->fragment_width -
287 (s->y_superblock_width * 4 - s->fragment_width);
288
289 /* the first operation for this variable is to advance by 1 */
290 current_fragment = -1;
291
292 } else if (i == s->u_superblock_start) {
293
294 /* start of U superblocks */
295 right_edge = s->fragment_width / 2;
296 bottom_edge = s->fragment_height / 2;
297 current_width = -1;
298 current_height = 0;
299 superblock_row_inc = 3 * (s->fragment_width / 2) -
300 (s->c_superblock_width * 4 - s->fragment_width / 2);
301
302 /* the first operation for this variable is to advance by 1 */
303 current_fragment = s->fragment_start[1] - 1;
304
305 } else if (i == s->v_superblock_start) {
306
307 /* start of V superblocks */
308 right_edge = s->fragment_width / 2;
309 bottom_edge = s->fragment_height / 2;
310 current_width = -1;
311 current_height = 0;
312 superblock_row_inc = 3 * (s->fragment_width / 2) -
313 (s->c_superblock_width * 4 - s->fragment_width / 2);
314
315 /* the first operation for this variable is to advance by 1 */
316 current_fragment = s->fragment_start[2] - 1;
317
318 }
319
320 if (current_width >= right_edge - 1) {
321 /* reset width and move to next superblock row */
322 current_width = -1;
323 current_height += 4;
324
325 /* fragment is now at the start of a new superblock row */
326 current_fragment += superblock_row_inc;
327 }
328
329 /* iterate through all 16 fragments in a superblock */
330 for (j = 0; j < 16; j++) {
331 current_fragment += travel_width[j] + right_edge * travel_height[j];
332 current_width += travel_width[j];
333 current_height += travel_height[j];
334
335 /* check if the fragment is in bounds */
336 if ((current_width < right_edge) &&
337 (current_height < bottom_edge)) {
338 s->superblock_fragments[mapping_index] = current_fragment;
339 } else {
340 s->superblock_fragments[mapping_index] = -1;
341 }
342
343 mapping_index++;
344 }
345 }
346
347 return 0; /* successful path out */
348 }
349
350 /*
351 * This function wipes out all of the fragment data.
352 */
353 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
354 {
355 int i;
356
357 /* zero out all of the fragment information */
358 s->coded_fragment_list_index = 0;
359 for (i = 0; i < s->fragment_count; i++) {
360 s->coeff_counts[i] = 0;
361 s->all_fragments[i].motion_x = 127;
362 s->all_fragments[i].motion_y = 127;
363 s->all_fragments[i].next_coeff= NULL;
364 s->all_fragments[i].qpi = 0;
365 s->coeffs[i].index=
366 s->coeffs[i].coeff=0;
367 s->coeffs[i].next= NULL;
368 }
369 }
370
371 /*
372 * This function sets up the dequantization tables used for a particular
373 * frame.
374 */
375 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
376 {
377 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
378 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
379 int i, plane, inter, qri, bmi, bmj, qistart;
380
381 for(inter=0; inter<2; inter++){
382 for(plane=0; plane<3; plane++){
383 int sum=0;
384 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
385 sum+= s->qr_size[inter][plane][qri];
386 if(s->qps[qpi] <= sum)
387 break;
388 }
389 qistart= sum - s->qr_size[inter][plane][qri];
390 bmi= s->qr_base[inter][plane][qri ];
391 bmj= s->qr_base[inter][plane][qri+1];
392 for(i=0; i<64; i++){
393 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
394 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
395 + s->qr_size[inter][plane][qri])
396 / (2*s->qr_size[inter][plane][qri]);
397
398 int qmin= 8<<(inter + !i);
399 int qscale= i ? ac_scale_factor : dc_scale_factor;
400
401 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
402 }
403 // all DC coefficients use the same quant so as not to interfere with DC prediction
404 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
405 }
406 }
407
408 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
409 }
410
411 /*
412 * This function initializes the loop filter boundary limits if the frame's
413 * quality index is different from the previous frame's.
414 *
415 * The filter_limit_values may not be larger than 127.
416 */
417 static void init_loop_filter(Vp3DecodeContext *s)
418 {
419 int *bounding_values= s->bounding_values_array+127;
420 int filter_limit;
421 int x;
422 int value;
423
424 filter_limit = s->filter_limit_values[s->qps[0]];
425
426 /* set up the bounding values */
427 memset(s->bounding_values_array, 0, 256 * sizeof(int));
428 for (x = 0; x < filter_limit; x++) {
429 bounding_values[-x] = -x;
430 bounding_values[x] = x;
431 }
432 for (x = value = filter_limit; x < 128 && value; x++, value--) {
433 bounding_values[ x] = value;
434 bounding_values[-x] = -value;
435 }
436 if (value)
437 bounding_values[128] = value;
438 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
439 }
440
441 /*
442 * This function unpacks all of the superblock/macroblock/fragment coding
443 * information from the bitstream.
444 */
445 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
446 {
447 int bit = 0;
448 int current_superblock = 0;
449 int current_run = 0;
450 int decode_fully_flags = 0;
451 int decode_partial_blocks = 0;
452 int first_c_fragment_seen;
453
454 int i, j;
455 int current_fragment;
456
457 if (s->keyframe) {
458 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
459
460 } else {
461
462 /* unpack the list of partially-coded superblocks */
463 bit = get_bits1(gb);
464 /* toggle the bit because as soon as the first run length is
465 * fetched the bit will be toggled again */
466 bit ^= 1;
467 while (current_superblock < s->superblock_count) {
468 if (current_run-- == 0) {
469 bit ^= 1;
470 current_run = get_vlc2(gb,
471 s->superblock_run_length_vlc.table, 6, 2);
472 if (current_run == 33)
473 current_run += get_bits(gb, 12);
474
475 /* if any of the superblocks are not partially coded, flag
476 * a boolean to decode the list of fully-coded superblocks */
477 if (bit == 0) {
478 decode_fully_flags = 1;
479 } else {
480
481 /* make a note of the fact that there are partially coded
482 * superblocks */
483 decode_partial_blocks = 1;
484 }
485 }
486 s->superblock_coding[current_superblock++] = bit;
487 }
488
489 /* unpack the list of fully coded superblocks if any of the blocks were
490 * not marked as partially coded in the previous step */
491 if (decode_fully_flags) {
492
493 current_superblock = 0;
494 current_run = 0;
495 bit = get_bits1(gb);
496 /* toggle the bit because as soon as the first run length is
497 * fetched the bit will be toggled again */
498 bit ^= 1;
499 while (current_superblock < s->superblock_count) {
500
501 /* skip any superblocks already marked as partially coded */
502 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
503
504 if (current_run-- == 0) {
505 bit ^= 1;
506 current_run = get_vlc2(gb,
507 s->superblock_run_length_vlc.table, 6, 2);
508 if (current_run == 33)
509 current_run += get_bits(gb, 12);
510 }
511 s->superblock_coding[current_superblock] = 2*bit;
512 }
513 current_superblock++;
514 }
515 }
516
517 /* if there were partial blocks, initialize bitstream for
518 * unpacking fragment codings */
519 if (decode_partial_blocks) {
520
521 current_run = 0;
522 bit = get_bits1(gb);
523 /* toggle the bit because as soon as the first run length is
524 * fetched the bit will be toggled again */
525 bit ^= 1;
526 }
527 }
528
529 /* figure out which fragments are coded; iterate through each
530 * superblock (all planes) */
531 s->coded_fragment_list_index = 0;
532 s->next_coeff= s->coeffs + s->fragment_count;
533 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
534 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
535 first_c_fragment_seen = 0;
536 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
537 for (i = 0; i < s->superblock_count; i++) {
538
539 /* iterate through all 16 fragments in a superblock */
540 for (j = 0; j < 16; j++) {
541
542 /* if the fragment is in bounds, check its coding status */
543 current_fragment = s->superblock_fragments[i * 16 + j];
544 if (current_fragment >= s->fragment_count) {
545 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
546 current_fragment, s->fragment_count);
547 return 1;
548 }
549 if (current_fragment != -1) {
550 if (s->superblock_coding[i] == SB_NOT_CODED) {
551
552 /* copy all the fragments from the prior frame */
553 s->all_fragments[current_fragment].coding_method =
554 MODE_COPY;
555
556 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
557
558 /* fragment may or may not be coded; this is the case
559 * that cares about the fragment coding runs */
560 if (current_run-- == 0) {
561 bit ^= 1;
562 current_run = get_vlc2(gb,
563 s->fragment_run_length_vlc.table, 5, 2);
564 }
565
566 if (bit) {
567 /* default mode; actual mode will be decoded in
568 * the next phase */
569 s->all_fragments[current_fragment].coding_method =
570 MODE_INTER_NO_MV;
571 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
572 s->coded_fragment_list[s->coded_fragment_list_index] =
573 current_fragment;
574 if ((current_fragment >= s->fragment_start[1]) &&
575 (s->last_coded_y_fragment == -1) &&
576 (!first_c_fragment_seen)) {
577 s->first_coded_c_fragment = s->coded_fragment_list_index;
578 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
579 first_c_fragment_seen = 1;
580 }
581 s->coded_fragment_list_index++;
582 } else {
583 /* not coded; copy this fragment from the prior frame */
584 s->all_fragments[current_fragment].coding_method =
585 MODE_COPY;
586 }
587
588 } else {
589
590 /* fragments are fully coded in this superblock; actual
591 * coding will be determined in next step */
592 s->all_fragments[current_fragment].coding_method =
593 MODE_INTER_NO_MV;
594 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
595 s->coded_fragment_list[s->coded_fragment_list_index] =
596 current_fragment;
597 if ((current_fragment >= s->fragment_start[1]) &&
598 (s->last_coded_y_fragment == -1) &&
599 (!first_c_fragment_seen)) {
600 s->first_coded_c_fragment = s->coded_fragment_list_index;
601 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
602 first_c_fragment_seen = 1;
603 }
604 s->coded_fragment_list_index++;
605 }
606 }
607 }
608 }
609
610 if (!first_c_fragment_seen)
611 /* only Y fragments coded in this frame */
612 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
613 else
614 /* end the list of coded C fragments */
615 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
616
617 for (i = 0; i < s->fragment_count - 1; i++) {
618 s->fast_fragment_list[i] = i + 1;
619 }
620 s->fast_fragment_list[s->fragment_count - 1] = -1;
621
622 if (s->last_coded_y_fragment == -1)
623 s->fragment_list_y_head = -1;
624 else {
625 s->fragment_list_y_head = s->first_coded_y_fragment;
626 s->fast_fragment_list[s->last_coded_y_fragment] = -1;
627 }
628
629 if (s->last_coded_c_fragment == -1)
630 s->fragment_list_c_head = -1;
631 else {
632 s->fragment_list_c_head = s->first_coded_c_fragment;
633 s->fast_fragment_list[s->last_coded_c_fragment] = -1;
634 }
635
636 return 0;
637 }
638
639 /*
640 * This function unpacks all the coding mode data for individual macroblocks
641 * from the bitstream.
642 */
643 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
644 {
645 int i, j, k, sb_x, sb_y;
646 int scheme;
647 int current_macroblock;
648 int current_fragment;
649 int coding_mode;
650 int custom_mode_alphabet[CODING_MODE_COUNT];
651
652 if (s->keyframe) {
653 for (i = 0; i < s->fragment_count; i++)
654 s->all_fragments[i].coding_method = MODE_INTRA;
655
656 } else {
657
658 /* fetch the mode coding scheme for this frame */
659 scheme = get_bits(gb, 3);
660
661 /* is it a custom coding scheme? */
662 if (scheme == 0) {
663 for (i = 0; i < 8; i++)
664 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
665 for (i = 0; i < 8; i++)
666 custom_mode_alphabet[get_bits(gb, 3)] = i;
667 }
668
669 /* iterate through all of the macroblocks that contain 1 or more
670 * coded fragments */
671 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
672 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
673
674 for (j = 0; j < 4; j++) {
675 int mb_x = 2*sb_x + (j>>1);
676 int mb_y = 2*sb_y + (((j>>1)+j)&1);
677 int frags_coded = 0;
678 current_macroblock = mb_y * s->macroblock_width + mb_x;
679
680 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
681 continue;
682
683 #define BLOCK_X (2*mb_x + (k&1))
684 #define BLOCK_Y (2*mb_y + (k>>1))
685 /* coding modes are only stored if the macroblock has at least one
686 * luma block coded, otherwise it must be INTER_NO_MV */
687 for (k = 0; k < 4; k++) {
688 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
689 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
690 break;
691 }
692 if (k == 4) {
693 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
694 continue;
695 }
696
697 /* mode 7 means get 3 bits for each coding mode */
698 if (scheme == 7)
699 coding_mode = get_bits(gb, 3);
700 else if(scheme == 0)
701 coding_mode = custom_mode_alphabet
702 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
703 else
704 coding_mode = ModeAlphabet[scheme-1]
705 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
706
707 s->macroblock_coding[current_macroblock] = coding_mode;
708 for (k = 0; k < 4; k++) {
709 current_fragment =
710 BLOCK_Y*s->fragment_width + BLOCK_X;
711 if (s->all_fragments[current_fragment].coding_method !=
712 MODE_COPY)
713 s->all_fragments[current_fragment].coding_method =
714 coding_mode;
715 }
716 for (k = 0; k < 2; k++) {
717 current_fragment = s->fragment_start[k+1] +
718 mb_y*(s->fragment_width>>1) + mb_x;
719 if (s->all_fragments[current_fragment].coding_method !=
720 MODE_COPY)
721 s->all_fragments[current_fragment].coding_method =
722 coding_mode;
723 }
724 }
725 }
726 }
727 }
728
729 return 0;
730 }
731
732 /*
733 * This function unpacks all the motion vectors for the individual
734 * macroblocks from the bitstream.
735 */
736 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
737 {
738 int j, k, l, sb_x, sb_y;
739 int coding_mode;
740 int motion_x[6];
741 int motion_y[6];
742 int last_motion_x = 0;
743 int last_motion_y = 0;
744 int prior_last_motion_x = 0;
745 int prior_last_motion_y = 0;
746 int current_macroblock;
747 int current_fragment;
748
749 if (s->keyframe)
750 return 0;
751
752 memset(motion_x, 0, 6 * sizeof(int));
753 memset(motion_y, 0, 6 * sizeof(int));
754
755 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
756 coding_mode = get_bits1(gb);
757
758 /* iterate through all of the macroblocks that contain 1 or more
759 * coded fragments */
760 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
761 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
762
763 for (j = 0; j < 4; j++) {
764 int mb_x = 2*sb_x + (j>>1);
765 int mb_y = 2*sb_y + (((j>>1)+j)&1);
766 current_macroblock = mb_y * s->macroblock_width + mb_x;
767
768 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
769 (s->macroblock_coding[current_macroblock] == MODE_COPY))
770 continue;
771
772 switch (s->macroblock_coding[current_macroblock]) {
773
774 case MODE_INTER_PLUS_MV:
775 case MODE_GOLDEN_MV:
776 /* all 6 fragments use the same motion vector */
777 if (coding_mode == 0) {
778 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
779 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
780 } else {
781 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
782 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
783 }
784
785 /* vector maintenance, only on MODE_INTER_PLUS_MV */
786 if (s->macroblock_coding[current_macroblock] ==
787 MODE_INTER_PLUS_MV) {
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 }
793 break;
794
795 case MODE_INTER_FOURMV:
796 /* vector maintenance */
797 prior_last_motion_x = last_motion_x;
798 prior_last_motion_y = last_motion_y;
799
800 /* fetch 4 vectors from the bitstream, one for each
801 * Y fragment, then average for the C fragment vectors */
802 motion_x[4] = motion_y[4] = 0;
803 for (k = 0; k < 4; k++) {
804 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
805 for (l = 0; l < s->coded_fragment_list_index; l++)
806 if (s->coded_fragment_list[l] == current_fragment)
807 break;
808 if (l < s->coded_fragment_list_index) {
809 if (coding_mode == 0) {
810 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
811 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
812 } else {
813 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
814 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
815 }
816 last_motion_x = motion_x[k];
817 last_motion_y = motion_y[k];
818 } else {
819 motion_x[k] = 0;
820 motion_y[k] = 0;
821 }
822 motion_x[4] += motion_x[k];
823 motion_y[4] += motion_y[k];
824 }
825
826 motion_x[5]=
827 motion_x[4]= RSHIFT(motion_x[4], 2);
828 motion_y[5]=
829 motion_y[4]= RSHIFT(motion_y[4], 2);
830 break;
831
832 case MODE_INTER_LAST_MV:
833 /* all 6 fragments use the last motion vector */
834 motion_x[0] = last_motion_x;
835 motion_y[0] = last_motion_y;
836
837 /* no vector maintenance (last vector remains the
838 * last vector) */
839 break;
840
841 case MODE_INTER_PRIOR_LAST:
842 /* all 6 fragments use the motion vector prior to the
843 * last motion vector */
844 motion_x[0] = prior_last_motion_x;
845 motion_y[0] = prior_last_motion_y;
846
847 /* vector maintenance */
848 prior_last_motion_x = last_motion_x;
849 prior_last_motion_y = last_motion_y;
850 last_motion_x = motion_x[0];
851 last_motion_y = motion_y[0];
852 break;
853
854 default:
855 /* covers intra, inter without MV, golden without MV */
856 motion_x[0] = 0;
857 motion_y[0] = 0;
858
859 /* no vector maintenance */
860 break;
861 }
862
863 /* assign the motion vectors to the correct fragments */
864 for (k = 0; k < 4; k++) {
865 current_fragment =
866 BLOCK_Y*s->fragment_width + BLOCK_X;
867 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
868 s->all_fragments[current_fragment].motion_x = motion_x[k];
869 s->all_fragments[current_fragment].motion_y = motion_y[k];
870 } else {
871 s->all_fragments[current_fragment].motion_x = motion_x[0];
872 s->all_fragments[current_fragment].motion_y = motion_y[0];
873 }
874 }
875 for (k = 0; k < 2; k++) {
876 current_fragment = s->fragment_start[k+1] +
877 mb_y*(s->fragment_width>>1) + mb_x;
878 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
879 s->all_fragments[current_fragment].motion_x = motion_x[k+4];
880 s->all_fragments[current_fragment].motion_y = motion_y[k+4];
881 } else {
882 s->all_fragments[current_fragment].motion_x = motion_x[0];
883 s->all_fragments[current_fragment].motion_y = motion_y[0];
884 }
885 }
886 }
887 }
888 }
889
890 return 0;
891 }
892
893 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
894 {
895 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
896 int num_blocks = s->coded_fragment_list_index;
897
898 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
899 i = blocks_decoded = num_blocks_at_qpi = 0;
900
901 bit = get_bits1(gb);
902
903 do {
904 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
905 if (run_length == 34)
906 run_length += get_bits(gb, 12);
907 blocks_decoded += run_length;
908
909 if (!bit)
910 num_blocks_at_qpi += run_length;
911
912 for (j = 0; j < run_length; i++) {
913 if (i >= s->coded_fragment_list_index)
914 return -1;
915
916 if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
917 s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
918 j++;
919 }
920 }
921
922 if (run_length == 4129)
923 bit = get_bits1(gb);
924 else
925 bit ^= 1;
926 } while (blocks_decoded < num_blocks);
927
928 num_blocks -= num_blocks_at_qpi;
929 }
930
931 return 0;
932 }
933
934 /*
935 * This function is called by unpack_dct_coeffs() to extract the VLCs from
936 * the bitstream. The VLCs encode tokens which are used to unpack DCT
937 * data. This function unpacks all the VLCs for either the Y plane or both
938 * C planes, and is called for DC coefficients or different AC coefficient
939 * levels (since different coefficient types require different VLC tables.
940 *
941 * This function returns a residual eob run. E.g, if a particular token gave
942 * instructions to EOB the next 5 fragments and there were only 2 fragments
943 * left in the current fragment range, 3 would be returned so that it could
944 * be passed into the next call to this same function.
945 */
946 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
947 VLC *table, int coeff_index,
948 int y_plane,
949 int eob_run)
950 {
951 int i;
952 int token;
953 int zero_run = 0;
954 DCTELEM coeff = 0;
955 Vp3Fragment *fragment;
956 int bits_to_get;
957 int next_fragment;
958 int previous_fragment;
959 int fragment_num;
960 int *list_head;
961
962 /* local references to structure members to avoid repeated deferences */
963 uint8_t *perm= s->scantable.permutated;
964 int *coded_fragment_list = s->coded_fragment_list;
965 Vp3Fragment *all_fragments = s->all_fragments;
966 uint8_t *coeff_counts = s->coeff_counts;
967 VLC_TYPE (*vlc_table)[2] = table->table;
968 int *fast_fragment_list = s->fast_fragment_list;
969
970 if (y_plane) {
971 next_fragment = s->fragment_list_y_head;
972 list_head = &s->fragment_list_y_head;
973 } else {
974 next_fragment = s->fragment_list_c_head;
975 list_head = &s->fragment_list_c_head;
976 }
977
978 i = next_fragment;
979 previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */
980 while (i != -1) {
981 fragment_num = coded_fragment_list[i];
982
983 if (coeff_counts[fragment_num] > coeff_index) {
984 previous_fragment = i;
985 i = fast_fragment_list[i];
986 continue;
987 }
988 fragment = &all_fragments[fragment_num];
989
990 if (!eob_run) {
991 /* decode a VLC into a token */
992 token = get_vlc2(gb, vlc_table, 5, 3);
993 /* use the token to get a zero run, a coefficient, and an eob run */
994 if (token <= 6) {
995 eob_run = eob_run_base[token];
996 if (eob_run_get_bits[token])
997 eob_run += get_bits(gb, eob_run_get_bits[token]);
998 coeff = zero_run = 0;
999 } else {
1000 bits_to_get = coeff_get_bits[token];
1001 if (bits_to_get)
1002 bits_to_get = get_bits(gb, bits_to_get);
1003 coeff = coeff_tables[token][bits_to_get];
1004
1005 zero_run = zero_run_base[token];
1006 if (zero_run_get_bits[token])
1007 zero_run += get_bits(gb, zero_run_get_bits[token]);
1008 }
1009 }
1010
1011 if (!eob_run) {
1012 coeff_counts[fragment_num] += zero_run;
1013 if (coeff_counts[fragment_num] < 64){
1014 fragment->next_coeff->coeff= coeff;
1015 fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1016 fragment->next_coeff->next= s->next_coeff;
1017 s->next_coeff->next=NULL;
1018 fragment->next_coeff= s->next_coeff++;
1019 }
1020 /* previous fragment is now this fragment */
1021 previous_fragment = i;
1022 } else {
1023 coeff_counts[fragment_num] |= 128;
1024 eob_run--;
1025 /* remove this fragment from the list */
1026 if (previous_fragment != -1)
1027 fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1028 else
1029 *list_head = fast_fragment_list[i];
1030 /* previous fragment remains unchanged */
1031 }
1032
1033 i = fast_fragment_list[i];
1034 }
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 /* fetch the DC table indexes */
1059 dc_y_table = get_bits(gb, 4);
1060 dc_c_table = get_bits(gb, 4);
1061
1062 /* unpack the Y plane DC coefficients */
1063 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1064 1, residual_eob_run);
1065
1066 /* reverse prediction of the Y-plane DC coefficients */
1067 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1068
1069 /* unpack the C plane DC coefficients */
1070 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1071 0, residual_eob_run);
1072
1073 /* reverse prediction of the C-plane DC coefficients */
1074 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1075 {
1076 reverse_dc_prediction(s, s->fragment_start[1],
1077 s->fragment_width / 2, s->fragment_height / 2);
1078 reverse_dc_prediction(s, s->fragment_start[2],
1079 s->fragment_width / 2, s->fragment_height / 2);
1080 }
1081
1082 /* fetch the AC table indexes */
1083 ac_y_table = get_bits(gb, 4);
1084 ac_c_table = get_bits(gb, 4);
1085
1086 /* build tables of AC VLC tables */
1087 for (i = 1; i <= 5; i++) {
1088 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1089 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1090 }
1091 for (i = 6; i <= 14; i++) {
1092 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1093 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1094 }
1095 for (i = 15; i <= 27; i++) {
1096 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1097 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1098 }
1099 for (i = 28; i <= 63; i++) {
1100 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1101 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1102 }
1103
1104 /* decode all AC coefficents */
1105 for (i = 1; i <= 63; i++) {
1106 if (s->fragment_list_y_head != -1)
1107 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1108 1, residual_eob_run);
1109
1110 if (s->fragment_list_c_head != -1)
1111 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1112 0, residual_eob_run);
1113 }
1114
1115 return 0;
1116 }
1117
1118 /*
1119 * This function reverses the DC prediction for each coded fragment in
1120 * the frame. Much of this function is adapted directly from the original
1121 * VP3 source code.
1122 */
1123 #define COMPATIBLE_FRAME(x) \
1124 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1125 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1126
1127 static void reverse_dc_prediction(Vp3DecodeContext *s,
1128 int first_fragment,
1129 int fragment_width,
1130 int fragment_height)
1131 {
1132
1133 #define PUL 8
1134 #define PU 4
1135 #define PUR 2
1136 #define PL 1
1137
1138 int x, y;
1139 int i = first_fragment;
1140
1141 int predicted_dc;
1142
1143 /* DC values for the left, up-left, up, and up-right fragments */
1144 int vl, vul, vu, vur;
1145
1146 /* indexes for the left, up-left, up, and up-right fragments */
1147 int l, ul, u, ur;
1148
1149 /*
1150 * The 6 fields mean:
1151 * 0: up-left multiplier
1152 * 1: up multiplier
1153 * 2: up-right multiplier
1154 * 3: left multiplier
1155 */
1156 static const int predictor_transform[16][4] = {
1157 { 0, 0, 0, 0},
1158 { 0, 0, 0,128}, // PL
1159 { 0, 0,128, 0}, // PUR
1160 { 0, 0, 53, 75}, // PUR|PL
1161 { 0,128, 0, 0}, // PU
1162 { 0, 64, 0, 64}, // PU|PL
1163 { 0,128, 0, 0}, // PU|PUR
1164 { 0, 0, 53, 75}, // PU|PUR|PL
1165 {128, 0, 0, 0}, // PUL
1166 { 0, 0, 0,128}, // PUL|PL
1167 { 64, 0, 64, 0}, // PUL|PUR
1168 { 0, 0, 53, 75}, // PUL|PUR|PL
1169 { 0,128, 0, 0}, // PUL|PU
1170 {-104,116, 0,116}, // PUL|PU|PL
1171 { 24, 80, 24, 0}, // PUL|PU|PUR
1172 {-104,116, 0,116} // PUL|PU|PUR|PL
1173 };
1174
1175 /* This table shows which types of blocks can use other blocks for
1176 * prediction. For example, INTRA is the only mode in this table to
1177 * have a frame number of 0. That means INTRA blocks can only predict
1178 * from other INTRA blocks. There are 2 golden frame coding types;
1179 * blocks encoding in these modes can only predict from other blocks
1180 * that were encoded with these 1 of these 2 modes. */
1181 static const unsigned char compatible_frame[9] = {
1182 1, /* MODE_INTER_NO_MV */
1183 0, /* MODE_INTRA */
1184 1, /* MODE_INTER_PLUS_MV */
1185 1, /* MODE_INTER_LAST_MV */
1186 1, /* MODE_INTER_PRIOR_MV */
1187 2, /* MODE_USING_GOLDEN */
1188 2, /* MODE_GOLDEN_MV */
1189 1, /* MODE_INTER_FOUR_MV */
1190 3 /* MODE_COPY */
1191 };
1192 int current_frame_type;
1193
1194 /* there is a last DC predictor for each of the 3 frame types */
1195 short last_dc[3];
1196
1197 int transform = 0;
1198
1199 vul = vu = vur = vl = 0;
1200 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1201
1202 /* for each fragment row... */
1203 for (y = 0; y < fragment_height; y++) {
1204
1205 /* for each fragment in a row... */
1206 for (x = 0; x < fragment_width; x++, i++) {
1207
1208 /* reverse prediction if this block was coded */
1209 if (s->all_fragments[i].coding_method != MODE_COPY) {
1210
1211 current_frame_type =
1212 compatible_frame[s->all_fragments[i].coding_method];
1213
1214 transform= 0;
1215 if(x){
1216 l= i-1;
1217 vl = DC_COEFF(l);
1218 if(COMPATIBLE_FRAME(l))
1219 transform |= PL;
1220 }
1221 if(y){
1222 u= i-fragment_width;
1223 vu = DC_COEFF(u);
1224 if(COMPATIBLE_FRAME(u))
1225 transform |= PU;
1226 if(x){
1227 ul= i-fragment_width-1;
1228 vul = DC_COEFF(ul);
1229 if(COMPATIBLE_FRAME(ul))
1230 transform |= PUL;
1231 }
1232 if(x + 1 < fragment_width){
1233 ur= i-fragment_width+1;
1234 vur = DC_COEFF(ur);
1235 if(COMPATIBLE_FRAME(ur))
1236 transform |= PUR;
1237 }
1238 }
1239
1240 if (transform == 0) {
1241
1242 /* if there were no fragments to predict from, use last
1243 * DC saved */
1244 predicted_dc = last_dc[current_frame_type];
1245 } else {
1246
1247 /* apply the appropriate predictor transform */
1248 predicted_dc =
1249 (predictor_transform[transform][0] * vul) +
1250 (predictor_transform[transform][1] * vu) +
1251 (predictor_transform[transform][2] * vur) +
1252 (predictor_transform[transform][3] * vl);
1253
1254 predicted_dc /= 128;
1255
1256 /* check for outranging on the [ul u l] and
1257 * [ul u ur l] predictors */
1258 if ((transform == 15) || (transform == 13)) {
1259 if (FFABS(predicted_dc - vu) > 128)
1260 predicted_dc = vu;
1261 else if (FFABS(predicted_dc - vl) > 128)
1262 predicted_dc = vl;
1263 else if (FFABS(predicted_dc - vul) > 128)
1264 predicted_dc = vul;
1265 }
1266 }
1267
1268 /* at long last, apply the predictor */
1269 if(s->coeffs[i].index){
1270 *s->next_coeff= s->coeffs[i];
1271 s->coeffs[i].index=0;
1272 s->coeffs[i].coeff=0;
1273 s->coeffs[i].next= s->next_coeff++;
1274 }
1275 s->coeffs[i].coeff += predicted_dc;
1276 /* save the DC */
1277 last_dc[current_frame_type] = DC_COEFF(i);
1278 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1279 s->coeff_counts[i]= 129;
1280 // s->all_fragments[i].next_coeff= s->next_coeff;
1281 s->coeffs[i].next= s->next_coeff;
1282 (s->next_coeff++)->next=NULL;
1283 }
1284 }
1285 }
1286 }
1287 }
1288
1289 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1290 {
1291 int x, y;
1292 int *bounding_values= s->bounding_values_array+127;
1293
1294 int width = s->fragment_width >> !!plane;
1295 int height = s->fragment_height >> !!plane;
1296 int fragment = s->fragment_start [plane] + ystart * width;
1297 int stride = s->current_frame.linesize[plane];
1298 uint8_t *plane_data = s->current_frame.data [plane];
1299 if (!s->flipped_image) stride = -stride;
1300 plane_data += s->data_offset[plane] + 8*ystart*stride;
1301
1302 for (y = ystart; y < yend; y++) {
1303
1304 for (x = 0; x < width; x++) {
1305 /* This code basically just deblocks on the edges of coded blocks.
1306 * However, it has to be much more complicated because of the
1307 * braindamaged deblock ordering used in VP3/Theora. Order matters
1308 * because some pixels get filtered twice. */
1309 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1310 {
1311 /* do not perform left edge filter for left columns frags */
1312 if (x > 0) {
1313 s->dsp.vp3_h_loop_filter(
1314 plane_data + 8*x,
1315 stride, bounding_values);
1316 }
1317
1318 /* do not perform top edge filter for top row fragments */
1319 if (y > 0) {
1320 s->dsp.vp3_v_loop_filter(
1321 plane_data + 8*x,
1322 stride, bounding_values);
1323 }
1324
1325 /* do not perform right edge filter for right column
1326 * fragments or if right fragment neighbor is also coded
1327 * in this frame (it will be filtered in next iteration) */
1328 if ((x < width - 1) &&
1329 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1330 s->dsp.vp3_h_loop_filter(
1331 plane_data + 8*x + 8,
1332 stride, bounding_values);
1333 }
1334
1335 /* do not perform bottom edge filter for bottom row
1336 * fragments or if bottom fragment neighbor is also coded
1337 * in this frame (it will be filtered in the next row) */
1338 if ((y < height - 1) &&
1339 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1340 s->dsp.vp3_v_loop_filter(
1341 plane_data + 8*x + 8*stride,
1342 stride, bounding_values);
1343 }
1344 }
1345
1346 fragment++;
1347 }
1348 plane_data += 8*stride;
1349 }
1350 }
1351
1352 /**
1353 * called when all pixels up to row y are complete
1354 */
1355 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1356 {
1357 int h, cy;
1358 int offset[4];
1359
1360 if(s->avctx->draw_horiz_band==NULL)
1361 return;
1362
1363 h= y - s->last_slice_end;
1364 y -= h;
1365
1366 if (!s->flipped_image) {
1367 if (y == 0)
1368 h -= s->height - s->avctx->height; // account for non-mod16
1369 y = s->height - y - h;
1370 }
1371
1372 cy = y >> 1;
1373 offset[0] = s->current_frame.linesize[0]*y;
1374 offset[1] = s->current_frame.linesize[1]*cy;
1375 offset[2] = s->current_frame.linesize[2]*cy;
1376 offset[3] = 0;
1377
1378 emms_c();
1379 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1380 s->last_slice_end= y + h;
1381 }
1382
1383 /*
1384 * Perform the final rendering for a particular slice of data.
1385 * The slice number ranges from 0..(macroblock_height - 1).
1386 */
1387 static void render_slice(Vp3DecodeContext *s, int slice)
1388 {
1389 int x;
1390 int16_t *dequantizer;
1391 DECLARE_ALIGNED_16(DCTELEM, block)[64];
1392 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1393 int motion_halfpel_index;
1394 uint8_t *motion_source;
1395 int plane;
1396
1397 if (slice >= s->macroblock_height)
1398 return;
1399
1400 for (plane = 0; plane < 3; plane++) {
1401 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1402 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1403 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1404 int stride = s->current_frame.linesize[plane];
1405 int plane_width = s->width >> !!plane;
1406 int plane_height = s->height >> !!plane;
1407 int y = slice * FRAGMENT_PIXELS << !plane ;
1408 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1409 int i = s->fragment_start[plane] + (y>>3)*(s->fragment_width>>!!plane);
1410
1411 if (!s->flipped_image) stride = -stride;
1412
1413
1414 if(FFABS(stride) > 2048)
1415 return; //various tables are fixed size
1416
1417 /* for each fragment row in the slice (both of them)... */
1418 for (; y < slice_height; y += 8) {
1419
1420 /* for each fragment in a row... */
1421 for (x = 0; x < plane_width; x += 8, i++) {
1422 int first_pixel = y*stride + x;
1423
1424 if ((i < 0) || (i >= s->fragment_count)) {
1425 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1426 return;
1427 }
1428
1429 /* transform if this block was coded */
1430 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1431 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1432
1433 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1434 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1435 motion_source= golden_plane;
1436 else
1437 motion_source= last_plane;
1438
1439 motion_source += first_pixel;
1440 motion_halfpel_index = 0;
1441
1442 /* sort out the motion vector if this fragment is coded
1443 * using a motion vector method */
1444 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1445 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1446 int src_x, src_y;
1447 motion_x = s->all_fragments[i].motion_x;
1448 motion_y = s->all_fragments[i].motion_y;
1449 if(plane){
1450 motion_x= (motion_x>>1) | (motion_x&1);
1451 motion_y= (motion_y>>1) | (motion_y&1);
1452 }
1453
1454 src_x= (motion_x>>1) + x;
1455 src_y= (motion_y>>1) + y;
1456 if ((motion_x == 127) || (motion_y == 127))
1457 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1458
1459 motion_halfpel_index = motion_x & 0x01;
1460 motion_source += (motion_x >> 1);
1461
1462 motion_halfpel_index |= (motion_y & 0x01) << 1;
1463 motion_source += ((motion_y >> 1) * stride);
1464
1465 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1466 uint8_t *temp= s->edge_emu_buffer;
1467 if(stride<0) temp -= 9*stride;
1468 else temp += 9*stride;
1469
1470 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1471 motion_source= temp;
1472 }
1473 }
1474
1475
1476 /* first, take care of copying a block from either the
1477 * previous or the golden frame */
1478 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1479 /* Note, it is possible to implement all MC cases with
1480 put_no_rnd_pixels_l2 which would look more like the
1481 VP3 source but this would be slower as
1482 put_no_rnd_pixels_tab is better optimzed */
1483 if(motion_halfpel_index != 3){
1484 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1485 output_plane + first_pixel,
1486 motion_source, stride, 8);
1487 }else{
1488 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1489 s->dsp.put_no_rnd_pixels_l2[1](
1490 output_plane + first_pixel,
1491 motion_source - d,
1492 motion_source + stride + 1 + d,
1493 stride, 8);
1494 }
1495 dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1496 }else{
1497 dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1498 }
1499
1500 /* dequantize the DCT coefficients */
1501 if(s->avctx->idct_algo==FF_IDCT_VP3){
1502 Coeff *coeff= s->coeffs + i;
1503 s->dsp.clear_block(block);
1504 while(coeff->next){
1505 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1506 coeff= coeff->next;
1507 }
1508 }else{
1509 Coeff *coeff= s->coeffs + i;
1510 s->dsp.clear_block(block);
1511 while(coeff->next){
1512 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1513 coeff= coeff->next;
1514 }
1515 }
1516
1517 /* invert DCT and place (or add) in final output */
1518
1519 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1520 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1521 block[0] += 128<<3;
1522 s->dsp.idct_put(
1523 output_plane + first_pixel,
1524 stride,
1525 block);
1526 } else {
1527 s->dsp.idct_add(
1528 output_plane + first_pixel,
1529 stride,
1530 block);
1531 }
1532 } else {
1533
1534 /* copy directly from the previous frame */
1535 s->dsp.put_pixels_tab[1][0](
1536 output_plane + first_pixel,
1537 last_plane + first_pixel,
1538 stride, 8);
1539
1540 }
1541 }
1542 // Filter the previous block row. We can't filter the current row yet
1543 // since it needs pixels from the next row
1544 if (y > 0)
1545 apply_loop_filter(s, plane, (y>>3)-1, (y>>3));
1546 }
1547 }
1548
1549 /* this looks like a good place for slice dispatch... */
1550 /* algorithm:
1551 * if (slice == s->macroblock_height - 1)
1552 * dispatch (both last slice & 2nd-to-last slice);
1553 * else if (slice > 0)
1554 * dispatch (slice - 1);
1555 */
1556
1557 // now that we've filtered the last rows, they're safe to display
1558 if (slice)
1559 vp3_draw_horiz_band(s, 16*slice);
1560 }
1561
1562 /*
1563 * This is the ffmpeg/libavcodec API init function.
1564 */
1565 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1566 {
1567 Vp3DecodeContext *s = avctx->priv_data;
1568 int i, inter, plane;
1569 int c_width;
1570 int c_height;
1571 int y_superblock_count;
1572 int c_superblock_count;
1573
1574 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1575 s->version = 0;
1576 else
1577 s->version = 1;
1578
1579 s->avctx = avctx;
1580 s->width = FFALIGN(avctx->width, 16);
1581 s->height = FFALIGN(avctx->height, 16);
1582 avctx->pix_fmt = PIX_FMT_YUV420P;
1583 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1584 if(avctx->idct_algo==FF_IDCT_AUTO)
1585 avctx->idct_algo=FF_IDCT_VP3;
1586 dsputil_init(&s->dsp, avctx);
1587
1588 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1589
1590 /* initialize to an impossible value which will force a recalculation
1591 * in the first frame decode */
1592 for (i = 0; i < 3; i++)
1593 s->qps[i] = -1;
1594
1595 s->y_superblock_width = (s->width + 31) / 32;
1596 s->y_superblock_height = (s->height + 31) / 32;
1597 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1598
1599 /* work out the dimensions for the C planes */
1600 c_width = s->width / 2;
1601 c_height = s->height / 2;
1602 s->c_superblock_width = (c_width + 31) / 32;
1603 s->c_superblock_height = (c_height + 31) / 32;
1604 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1605
1606 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1607 s->u_superblock_start = y_superblock_count;
1608 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1609 s->superblock_coding = av_malloc(s->superblock_count);
1610
1611 s->macroblock_width = (s->width + 15) / 16;
1612 s->macroblock_height = (s->height + 15) / 16;
1613 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1614
1615 s->fragment_width = s->width / FRAGMENT_PIXELS;
1616 s->fragment_height = s->height / FRAGMENT_PIXELS;
1617
1618 /* fragment count covers all 8x8 blocks for all 3 planes */
1619 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1620 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1621 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1622
1623 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1624 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1625 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1626 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1627 s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1628 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1629 !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1630 vp3_decode_end(avctx);
1631 return -1;
1632 }
1633
1634 if (!s->theora_tables)
1635 {
1636 for (i = 0; i < 64; i++) {
1637 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1638 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1639 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1640 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1641 s->base_matrix[2][i] = vp31_inter_dequant[i];
1642 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1643 }
1644
1645 for(inter=0; inter<2; inter++){
1646 for(plane=0; plane<3; plane++){
1647 s->qr_count[inter][plane]= 1;
1648 s->qr_size [inter][plane][0]= 63;
1649 s->qr_base [inter][plane][0]=
1650 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1651 }
1652 }
1653
1654 /* init VLC tables */
1655 for (i = 0; i < 16; i++) {
1656
1657 /* DC histograms */
1658 init_vlc(&s->dc_vlc[i], 5, 32,
1659 &dc_bias[i][0][1], 4, 2,
1660 &dc_bias[i][0][0], 4, 2, 0);
1661
1662 /* group 1 AC histograms */
1663 init_vlc(&s->ac_vlc_1[i], 5, 32,
1664 &ac_bias_0[i][0][1], 4, 2,
1665 &ac_bias_0[i][0][0], 4, 2, 0);
1666
1667 /* group 2 AC histograms */
1668 init_vlc(&s->ac_vlc_2[i], 5, 32,
1669 &ac_bias_1[i][0][1], 4, 2,
1670 &ac_bias_1[i][0][0], 4, 2, 0);
1671
1672 /* group 3 AC histograms */
1673 init_vlc(&s->ac_vlc_3[i], 5, 32,
1674 &ac_bias_2[i][0][1], 4, 2,
1675 &ac_bias_2[i][0][0], 4, 2, 0);
1676
1677 /* group 4 AC histograms */
1678 init_vlc(&s->ac_vlc_4[i], 5, 32,
1679 &ac_bias_3[i][0][1], 4, 2,
1680 &ac_bias_3[i][0][0], 4, 2, 0);
1681 }
1682 } else {
1683 for (i = 0; i < 16; i++) {
1684
1685 /* DC histograms */
1686 if (init_vlc(&s->dc_vlc[i], 5, 32,
1687 &s->huffman_table[i][0][1], 4, 2,
1688 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1689 goto vlc_fail;
1690
1691 /* group 1 AC histograms */
1692 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1693 &s->huffman_table[i+16][0][1], 4, 2,
1694 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1695 goto vlc_fail;
1696
1697 /* group 2 AC histograms */
1698 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1699 &s->huffman_table[i+16*2][0][1], 4, 2,
1700 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1701 goto vlc_fail;
1702
1703 /* group 3 AC histograms */
1704 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1705 &s->huffman_table[i+16*3][0][1], 4, 2,
1706 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1707 goto vlc_fail;
1708
1709 /* group 4 AC histograms */
1710 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1711 &s->huffman_table[i+16*4][0][1], 4, 2,
1712 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1713 goto vlc_fail;
1714 }
1715 }
1716
1717 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1718 &superblock_run_length_vlc_table[0][1], 4, 2,
1719 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1720
1721 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1722 &fragment_run_length_vlc_table[0][1], 4, 2,
1723 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1724
1725 init_vlc(&s->mode_code_vlc, 3, 8,
1726 &mode_code_vlc_table[0][1], 2, 1,
1727 &mode_code_vlc_table[0][0], 2, 1, 0);
1728
1729 init_vlc(&s->motion_vector_vlc, 6, 63,
1730 &motion_vector_vlc_table[0][1], 2, 1,
1731 &motion_vector_vlc_table[0][0], 2, 1, 0);
1732
1733 /* work out the block mapping tables */
1734 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1735 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1736 if (!s->superblock_fragments || !s->macroblock_coding) {
1737 vp3_decode_end(avctx);
1738 return -1;
1739 }
1740 init_block_mapping(s);
1741
1742 for (i = 0; i < 3; i++) {
1743 s->current_frame.data[i] = NULL;
1744 s->last_frame.data[i] = NULL;
1745 s->golden_frame.data[i] = NULL;
1746 }
1747
1748 return 0;
1749
1750 vlc_fail:
1751 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1752 return -1;
1753 }
1754
1755 /*
1756 * This is the ffmpeg/libavcodec API frame decode function.
1757 */
1758 static int vp3_decode_frame(AVCodecContext *avctx,
1759 void *data, int *data_size,
1760 AVPacket *avpkt)
1761 {
1762 const uint8_t *buf = avpkt->data;
1763 int buf_size = avpkt->size;
1764 Vp3DecodeContext *s = avctx->priv_data;
1765 GetBitContext gb;
1766 static int counter = 0;
1767 int i;
1768
1769 init_get_bits(&gb, buf, buf_size * 8);
1770
1771 if (s->theora && get_bits1(&gb))
1772 {
1773 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1774 return -1;
1775 }
1776
1777 s->keyframe = !get_bits1(&gb);
1778 if (!s->theora)
1779 skip_bits(&gb, 1);
1780 for (i = 0; i < 3; i++)
1781 s->last_qps[i] = s->qps[i];
1782
1783 s->nqps=0;
1784 do{
1785 s->qps[s->nqps++]= get_bits(&gb, 6);
1786 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1787 for (i = s->nqps; i < 3; i++)
1788 s->qps[i] = -1;
1789
1790 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1791 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1792 s->keyframe?"key":"", counter, s->qps[0]);
1793 counter++;
1794
1795 if (s->qps[0] != s->last_qps[0])
1796 init_loop_filter(s);
1797
1798 for (i = 0; i < s->nqps; i++)
1799 // reinit all dequantizers if the first one changed, because
1800 // the DC of the first quantizer must be used for all matrices
1801 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1802 init_dequantizer(s, i);
1803
1804 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1805 return buf_size;
1806
1807 if (s->keyframe) {
1808 if (!s->theora)
1809 {
1810 skip_bits(&gb, 4); /* width code */
1811 skip_bits(&gb, 4); /* height code */
1812 if (s->version)
1813 {
1814 s->version = get_bits(&gb, 5);
1815 if (counter == 1)
1816 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1817 }
1818 }
1819 if (s->version || s->theora)
1820 {
1821 if (get_bits1(&gb))
1822 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1823 skip_bits(&gb, 2); /* reserved? */
1824 }
1825
1826 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1827 if (s->golden_frame.data[0])
1828 avctx->release_buffer(avctx, &s->golden_frame);
1829 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1830 } else {
1831 if (s->golden_frame.data[0])
1832 avctx->release_buffer(avctx, &s->golden_frame);
1833 if (s->last_frame.data[0])
1834 avctx->release_buffer(avctx, &s->last_frame);
1835 }
1836
1837 s->golden_frame.reference = 3;
1838 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1839 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1840 return -1;
1841 }
1842
1843 /* golden frame is also the current frame */
1844 s->current_frame= s->golden_frame;
1845 } else {
1846 /* allocate a new current frame */
1847 s->current_frame.reference = 3;
1848 if (!s->golden_frame.data[0]) {
1849 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1850 return -1;
1851 }
1852 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1853 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1854 return -1;
1855 }
1856 }
1857
1858 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1859 s->current_frame.qstride= 0;
1860
1861 init_frame(s, &gb);
1862
1863 if (unpack_superblocks(s, &gb)){
1864 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1865 return -1;
1866 }
1867 if (unpack_modes(s, &gb)){
1868 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1869 return -1;
1870 }
1871 if (unpack_vectors(s, &gb)){
1872 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1873 return -1;
1874 }
1875 if (unpack_block_qpis(s, &gb)){
1876 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1877 return -1;
1878 }
1879 if (unpack_dct_coeffs(s, &gb)){
1880 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1881 return -1;
1882 }
1883
1884 for (i = 0; i < 3; i++) {
1885 if (s->flipped_image)
1886 s->data_offset[i] = 0;
1887 else
1888 s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1889 }
1890
1891 s->last_slice_end = 0;
1892 for (i = 0; i < s->macroblock_height; i++)
1893 render_slice(s, i);
1894
1895 // filter the last row
1896 for (i = 0; i < 3; i++) {
1897 int row = (s->height >> (3+!!i)) - 1;
1898 apply_loop_filter(s, i, row, row+1);
1899 }
1900 vp3_draw_horiz_band(s, s->height);
1901
1902 *data_size=sizeof(AVFrame);
1903 *(AVFrame*)data= s->current_frame;
1904
1905 /* release the last frame, if it is allocated and if it is not the
1906 * golden frame */
1907 if ((s->last_frame.data[0]) &&
1908 (s->last_frame.data[0] != s->golden_frame.data[0]))
1909 avctx->release_buffer(avctx, &s->last_frame);
1910
1911 /* shuffle frames (last = current) */
1912 s->last_frame= s->current_frame;
1913 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1914
1915 return buf_size;
1916 }
1917
1918 /*
1919 * This is the ffmpeg/libavcodec API module cleanup function.
1920 */
1921 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1922 {
1923 Vp3DecodeContext *s = avctx->priv_data;
1924 int i;
1925
1926 av_free(s->superblock_coding);
1927 av_free(s->all_fragments);
1928 av_free(s->coeff_counts);
1929 av_free(s->coeffs);
1930 av_free(s->coded_fragment_list);
1931 av_free(s->fast_fragment_list);
1932 av_free(s->superblock_fragments);
1933 av_free(s->macroblock_coding);
1934
1935 for (i = 0; i < 16; i++) {
1936 free_vlc(&s->dc_vlc[i]);
1937 free_vlc(&s->ac_vlc_1[i]);
1938 free_vlc(&s->ac_vlc_2[i]);
1939 free_vlc(&s->ac_vlc_3[i]);
1940 free_vlc(&s->ac_vlc_4[i]);
1941 }
1942
1943 free_vlc(&s->superblock_run_length_vlc);
1944 free_vlc(&s->fragment_run_length_vlc);
1945 free_vlc(&s->mode_code_vlc);
1946 free_vlc(&s->motion_vector_vlc);
1947
1948 /* release all frames */
1949 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1950 avctx->release_buffer(avctx, &s->golden_frame);
1951 if (s->last_frame.data[0])
1952 avctx->release_buffer(avctx, &s->last_frame);
1953 /* no need to release the current_frame since it will always be pointing
1954 * to the same frame as either the golden or last frame */
1955
1956 return 0;
1957 }
1958
1959 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1960 {
1961 Vp3DecodeContext *s = avctx->priv_data;
1962
1963 if (get_bits1(gb)) {
1964 int token;
1965 if (s->entries >= 32) { /* overflow */
1966 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1967 return -1;
1968 }
1969 token = get_bits(gb, 5);
1970 //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size);
1971 s->huffman_table[s->hti][token][0] = s->hbits;
1972 s->huffman_table[s->hti][token][1] = s->huff_code_size;
1973 s->entries++;
1974 }
1975 else {
1976 if (s->huff_code_size >= 32) {/* overflow */
1977 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1978 return -1;
1979 }
1980 s->huff_code_size++;
1981 s->hbits <<= 1;
1982 if (read_huffman_tree(avctx, gb))
1983 return -1;
1984 s->hbits |= 1;
1985 if (read_huffman_tree(avctx, gb))
1986 return -1;
1987 s->hbits >>= 1;
1988 s->huff_code_size--;
1989 }
1990 return 0;
1991 }
1992
1993 #if CONFIG_THEORA_DECODER
1994 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1995 {
1996 Vp3DecodeContext *s = avctx->priv_data;
1997 int visible_width, visible_height, colorspace;
1998
1999 s->theora = get_bits_long(gb, 24);
2000 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2001
2002 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2003 /* but previous versions have the image flipped relative to vp3 */
2004 if (s->theora < 0x030200)
2005 {
2006 s->flipped_image = 1;
2007 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2008 }
2009
2010 visible_width = s->width = get_bits(gb, 16) << 4;
2011 visible_height = s->height = get_bits(gb, 16) << 4;
2012
2013 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2014 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2015 s->width= s->height= 0;
2016 return -1;
2017 }
2018
2019 if (s->theora >= 0x030200) {
2020 visible_width = get_bits_long(gb, 24);
2021 visible_height = get_bits_long(gb, 24);
2022
2023 skip_bits(gb, 8); /* offset x */
2024 skip_bits(gb, 8); /* offset y */
2025 }
2026
2027 skip_bits(gb, 32); /* fps numerator */
2028 skip_bits(gb, 32); /* fps denumerator */
2029 skip_bits(gb, 24); /* aspect numerator */
2030 skip_bits(gb, 24); /* aspect denumerator */
2031
2032 if (s->theora < 0x030200)
2033 skip_bits(gb, 5); /* keyframe frequency force */
2034 colorspace = get_bits(gb, 8);
2035 skip_bits(gb, 24); /* bitrate */
2036
2037 skip_bits(gb, 6); /* quality hint */
2038
2039 if (s->theora >= 0x030200)
2040 {
2041 skip_bits(gb, 5); /* keyframe frequency force */
2042 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2043 skip_bits(gb, 3); /* reserved */
2044 }
2045
2046 // align_get_bits(gb);
2047
2048 if ( visible_width <= s->width && visible_width > s->width-16
2049 && visible_height <= s->height && visible_height > s->height-16)
2050 avcodec_set_dimensions(avctx, visible_width, visible_height);
2051 else
2052 avcodec_set_dimensions(avctx, s->width, s->height);
2053
2054 if (colorspace == 1) {
2055 avctx->color_primaries = AVCOL_PRI_BT470M;
2056 } else if (colorspace == 2) {
2057 avctx->color_primaries = AVCOL_PRI_BT470BG;
2058 }
2059 if (colorspace == 1 || colorspace == 2) {
2060 avctx->colorspace = AVCOL_SPC_BT470BG;
2061 avctx->color_trc = AVCOL_TRC_BT709;
2062 }
2063
2064 return 0;
2065 }
2066
2067 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2068 {
2069 Vp3DecodeContext *s = avctx->priv_data;
2070 int i, n, matrices, inter, plane;
2071
2072 if (s->theora >= 0x030200) {
2073 n = get_bits(gb, 3);
2074 /* loop filter limit values table */
2075 for (i = 0; i < 64; i++) {
2076 s->filter_limit_values[i] = get_bits(gb, n);
2077 if (s->filter_limit_values[i] > 127) {
2078 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2079 s->filter_limit_values[i] = 127;
2080 }
2081 }
2082 }
2083
2084 if (s->theora >= 0x030200)
2085 n = get_bits(gb, 4) + 1;
2086 else
2087 n = 16;
2088 /* quality threshold table */
2089 for (i = 0; i < 64; i++)
2090 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2091
2092 if (s->theora >= 0x030200)
2093 n = get_bits(gb, 4) + 1;
2094 else
2095 n = 16;
2096 /* dc scale factor table */
2097 for (i = 0; i < 64; i++)
2098 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2099
2100 if (s->theora >= 0x030200)
2101 matrices = get_bits(gb, 9) + 1;
2102 else
2103 matrices = 3;
2104
2105 if(matrices > 384){
2106 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2107 return -1;
2108 }
2109
2110 for(n=0; n<matrices; n++){
2111 for (i = 0; i < 64; i++)
2112 s->base_matrix[n][i]= get_bits(gb, 8);
2113 }
2114
2115 for (inter = 0; inter <= 1; inter++) {
2116 for (plane = 0; plane <= 2; plane++) {
2117 int newqr= 1;
2118 if (inter || plane > 0)
2119 newqr = get_bits1(gb);
2120 if (!newqr) {
2121 int qtj, plj;
2122 if(inter && get_bits1(gb)){
2123 qtj = 0;
2124 plj = plane;
2125 }else{
2126 qtj= (3*inter + plane - 1) / 3;
2127 plj= (plane + 2) % 3;
2128 }
2129 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2130 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2131 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2132 } else {
2133 int qri= 0;
2134 int qi = 0;
2135
2136 for(;;){
2137 i= get_bits(gb, av_log2(matrices-1)+1);
2138 if(i>= matrices){
2139 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2140 return -1;
2141 }
2142 s->qr_base[inter][plane][qri]= i;
2143 if(qi >= 63)
2144 break;
2145 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2146 s->qr_size[inter][plane][qri++]= i;
2147 qi += i;
2148 }
2149
2150 if (qi > 63) {
2151 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2152 return -1;
2153 }
2154 s->qr_count[inter][plane]= qri;
2155 }
2156 }
2157 }
2158
2159 /* Huffman tables */
2160 for (s->hti = 0; s->hti < 80; s->hti++) {
2161 s->entries = 0;
2162 s->huff_code_size = 1;
2163 if (!get_bits1(gb)) {
2164 s->hbits = 0;
2165 if(read_huffman_tree(avctx, gb))
2166 return -1;
2167 s->hbits = 1;
2168 if(read_huffman_tree(avctx, gb))
2169 return -1;
2170 }
2171 }
2172
2173 s->theora_tables = 1;
2174
2175 return 0;
2176 }
2177
2178 static av_cold int theora_decode_init(AVCodecContext *avctx)
2179 {
2180 Vp3DecodeContext *s = avctx->priv_data;
2181 GetBitContext gb;
2182 int ptype;
2183 uint8_t *header_start[3];
2184 int header_len[3];
2185 int i;
2186
2187 s->theora = 1;
2188
2189 if (!avctx->extradata_size)
2190 {
2191 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2192 return -1;
2193 }
2194
2195 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2196 42, header_start, header_len) < 0) {
2197 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2198 return -1;
2199 }
2200
2201 for(i=0;i<3;i++) {
2202 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2203
2204 ptype = get_bits(&gb, 8);
2205
2206 if (!(ptype & 0x80))
2207 {
2208 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2209 // return -1;
2210 }
2211
2212 // FIXME: Check for this as well.
2213 skip_bits_long(&gb, 6*8); /* "theora" */
2214
2215 switch(ptype)
2216 {
2217 case 0x80:
2218 theora_decode_header(avctx, &gb);
2219 break;
2220 case 0x81:
2221 // FIXME: is this needed? it breaks sometimes
2222 // theora_decode_comments(avctx, gb);
2223 break;
2224 case 0x82:
2225 if (theora_decode_tables(avctx, &gb))
2226 return -1;
2227 break;
2228 default:
2229 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2230 break;
2231 }
2232 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2233 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2234 if (s->theora < 0x030200)
2235 break;
2236 }
2237
2238 return vp3_decode_init(avctx);
2239 }
2240
2241 AVCodec theora_decoder = {
2242 "theora",
2243 CODEC_TYPE_VIDEO,
2244 CODEC_ID_THEORA,
2245 sizeof(Vp3DecodeContext),
2246 theora_decode_init,
2247 NULL,
2248 vp3_decode_end,
2249 vp3_decode_frame,
2250 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2251 NULL,
2252 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2253 };
2254 #endif
2255
2256 AVCodec vp3_decoder = {
2257 "vp3",
2258 CODEC_TYPE_VIDEO,
2259 CODEC_ID_VP3,
2260 sizeof(Vp3DecodeContext),
2261 vp3_decode_init,
2262 NULL,
2263 vp3_decode_end,
2264 vp3_decode_frame,
2265 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2266 NULL,
2267 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2268 };