Eliminate macroblock -> fragment mapping array
[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 /* this is the macroblock that the fragment belongs to */
57 uint16_t macroblock;
58 uint8_t coding_method;
59 int8_t motion_x;
60 int8_t motion_y;
61 uint8_t qpi;
62 } Vp3Fragment;
63
64 #define SB_NOT_CODED 0
65 #define SB_PARTIALLY_CODED 1
66 #define SB_FULLY_CODED 2
67
68 #define MODE_INTER_NO_MV 0
69 #define MODE_INTRA 1
70 #define MODE_INTER_PLUS_MV 2
71 #define MODE_INTER_LAST_MV 3
72 #define MODE_INTER_PRIOR_LAST 4
73 #define MODE_USING_GOLDEN 5
74 #define MODE_GOLDEN_MV 6
75 #define MODE_INTER_FOURMV 7
76 #define CODING_MODE_COUNT 8
77
78 /* special internal mode */
79 #define MODE_COPY 8
80
81 /* There are 6 preset schemes, plus a free-form scheme */
82 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
83 {
84 /* scheme 1: Last motion vector dominates */
85 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
86 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
87 MODE_INTRA, MODE_USING_GOLDEN,
88 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
89
90 /* scheme 2 */
91 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
92 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
93 MODE_INTRA, MODE_USING_GOLDEN,
94 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
95
96 /* scheme 3 */
97 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
98 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
99 MODE_INTRA, MODE_USING_GOLDEN,
100 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
101
102 /* scheme 4 */
103 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
104 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
105 MODE_INTRA, MODE_USING_GOLDEN,
106 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
107
108 /* scheme 5: No motion vector dominates */
109 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
110 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
111 MODE_INTRA, MODE_USING_GOLDEN,
112 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
113
114 /* scheme 6 */
115 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
116 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
117 MODE_INTER_PLUS_MV, MODE_INTRA,
118 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
119
120 };
121
122 #define MIN_DEQUANT_VAL 2
123
124 typedef struct Vp3DecodeContext {
125 AVCodecContext *avctx;
126 int theora, theora_tables;
127 int version;
128 int width, height;
129 AVFrame golden_frame;
130 AVFrame last_frame;
131 AVFrame current_frame;
132 int keyframe;
133 DSPContext dsp;
134 int flipped_image;
135 int last_slice_end;
136
137 int qps[3];
138 int nqps;
139 int last_qps[3];
140
141 int superblock_count;
142 int y_superblock_width;
143 int y_superblock_height;
144 int c_superblock_width;
145 int c_superblock_height;
146 int u_superblock_start;
147 int v_superblock_start;
148 unsigned char *superblock_coding;
149
150 int macroblock_count;
151 int macroblock_width;
152 int macroblock_height;
153
154 int fragment_count;
155 int fragment_width;
156 int fragment_height;
157
158 Vp3Fragment *all_fragments;
159 uint8_t *coeff_counts;
160 Coeff *coeffs;
161 Coeff *next_coeff;
162 int fragment_start[3];
163 int data_offset[3];
164
165 ScanTable scantable;
166
167 /* tables */
168 uint16_t coded_dc_scale_factor[64];
169 uint32_t coded_ac_scale_factor[64];
170 uint8_t base_matrix[384][64];
171 uint8_t qr_count[2][3];
172 uint8_t qr_size [2][3][64];
173 uint16_t qr_base[2][3][64];
174
175 /* this is a list of indexes into the all_fragments array indicating
176 * which of the fragments are coded */
177 int *coded_fragment_list;
178 int coded_fragment_list_index;
179
180 /* track which fragments have already been decoded; called 'fast'
181 * because this data structure avoids having to iterate through every
182 * fragment in coded_fragment_list; once a fragment has been fully
183 * decoded, it is removed from this list */
184 int *fast_fragment_list;
185 int fragment_list_y_head;
186 int fragment_list_c_head;
187
188 VLC dc_vlc[16];
189 VLC ac_vlc_1[16];
190 VLC ac_vlc_2[16];
191 VLC ac_vlc_3[16];
192 VLC ac_vlc_4[16];
193
194 VLC superblock_run_length_vlc;
195 VLC fragment_run_length_vlc;
196 VLC mode_code_vlc;
197 VLC motion_vector_vlc;
198
199 /* these arrays need to be on 16-byte boundaries since SSE2 operations
200 * index into them */
201 DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
202
203 /* This table contains superblock_count * 16 entries. Each set of 16
204 * numbers corresponds to the fragment indexes 0..15 of the superblock.
205 * An entry will be -1 to indicate that no entry corresponds to that
206 * index. */
207 int *superblock_fragments;
208
209 /* This is an array that indicates how a particular macroblock
210 * is coded. */
211 unsigned char *macroblock_coding;
212
213 int first_coded_y_fragment;
214 int first_coded_c_fragment;
215 int last_coded_y_fragment;
216 int last_coded_c_fragment;
217
218 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
219 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
220
221 /* Huffman decode */
222 int hti;
223 unsigned int hbits;
224 int entries;
225 int huff_code_size;
226 uint16_t huffman_table[80][32][2];
227
228 uint8_t filter_limit_values[64];
229 DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
230 } Vp3DecodeContext;
231
232 /************************************************************************
233 * VP3 specific functions
234 ************************************************************************/
235
236 /*
237 * This function sets up all of the various blocks mappings:
238 * superblocks <-> fragments, macroblocks <-> fragments,
239 * superblocks <-> macroblocks
240 *
241 * Returns 0 is successful; returns 1 if *anything* went wrong.
242 */
243 static int init_block_mapping(Vp3DecodeContext *s)
244 {
245 int i, j;
246 signed int hilbert_walk_mb[4];
247
248 int current_fragment = 0;
249 int current_width = 0;
250 int current_height = 0;
251 int right_edge = 0;
252 int bottom_edge = 0;
253 int superblock_row_inc = 0;
254 int mapping_index = 0;
255
256 int current_macroblock;
257 int c_fragment;
258
259 static const signed char travel_width[16] = {
260 1, 1, 0, -1,
261 0, 0, 1, 0,
262 1, 0, 1, 0,
263 0, -1, 0, 1
264 };
265
266 static const signed char travel_height[16] = {
267 0, 0, 1, 0,
268 1, 1, 0, -1,
269 0, 1, 0, -1,
270 -1, 0, -1, 0
271 };
272
273 hilbert_walk_mb[0] = 1;
274 hilbert_walk_mb[1] = s->macroblock_width;
275 hilbert_walk_mb[2] = 1;
276 hilbert_walk_mb[3] = -s->macroblock_width;
277
278 /* iterate through each superblock (all planes) and map the fragments */
279 for (i = 0; i < s->superblock_count; i++) {
280 /* time to re-assign the limits? */
281 if (i == 0) {
282
283 /* start of Y superblocks */
284 right_edge = s->fragment_width;
285 bottom_edge = s->fragment_height;
286 current_width = -1;
287 current_height = 0;
288 superblock_row_inc = 3 * s->fragment_width -
289 (s->y_superblock_width * 4 - s->fragment_width);
290
291 /* the first operation for this variable is to advance by 1 */
292 current_fragment = -1;
293
294 } else if (i == s->u_superblock_start) {
295
296 /* start of U superblocks */
297 right_edge = s->fragment_width / 2;
298 bottom_edge = s->fragment_height / 2;
299 current_width = -1;
300 current_height = 0;
301 superblock_row_inc = 3 * (s->fragment_width / 2) -
302 (s->c_superblock_width * 4 - s->fragment_width / 2);
303
304 /* the first operation for this variable is to advance by 1 */
305 current_fragment = s->fragment_start[1] - 1;
306
307 } else if (i == s->v_superblock_start) {
308
309 /* start of V superblocks */
310 right_edge = s->fragment_width / 2;
311 bottom_edge = s->fragment_height / 2;
312 current_width = -1;
313 current_height = 0;
314 superblock_row_inc = 3 * (s->fragment_width / 2) -
315 (s->c_superblock_width * 4 - s->fragment_width / 2);
316
317 /* the first operation for this variable is to advance by 1 */
318 current_fragment = s->fragment_start[2] - 1;
319
320 }
321
322 if (current_width >= right_edge - 1) {
323 /* reset width and move to next superblock row */
324 current_width = -1;
325 current_height += 4;
326
327 /* fragment is now at the start of a new superblock row */
328 current_fragment += superblock_row_inc;
329 }
330
331 /* iterate through all 16 fragments in a superblock */
332 for (j = 0; j < 16; j++) {
333 current_fragment += travel_width[j] + right_edge * travel_height[j];
334 current_width += travel_width[j];
335 current_height += travel_height[j];
336
337 /* check if the fragment is in bounds */
338 if ((current_width < right_edge) &&
339 (current_height < bottom_edge)) {
340 s->superblock_fragments[mapping_index] = current_fragment;
341 } else {
342 s->superblock_fragments[mapping_index] = -1;
343 }
344
345 mapping_index++;
346 }
347 }
348
349 /* initialize the macroblock <-> fragment mapping */
350 current_fragment = 0;
351 current_macroblock = 0;
352 for (i = 0; i < s->fragment_height; i += 2) {
353
354 for (j = 0; j < s->fragment_width; j += 2) {
355
356 s->all_fragments[current_fragment].macroblock = current_macroblock;
357
358 if (j + 1 < s->fragment_width) {
359 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
360 }
361
362 if (i + 1 < s->fragment_height) {
363 s->all_fragments[current_fragment + s->fragment_width].macroblock =
364 current_macroblock;
365 }
366
367 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
368 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
369 current_macroblock;
370 }
371
372 /* C planes */
373 c_fragment = s->fragment_start[1] +
374 (i * s->fragment_width / 4) + (j / 2);
375 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
376
377 c_fragment = s->fragment_start[2] +
378 (i * s->fragment_width / 4) + (j / 2);
379 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
380
381 if (j + 2 <= s->fragment_width)
382 current_fragment += 2;
383 else
384 current_fragment++;
385 current_macroblock++;
386 }
387
388 current_fragment += s->fragment_width;
389 }
390
391 return 0; /* successful path out */
392 }
393
394 /*
395 * This function wipes out all of the fragment data.
396 */
397 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
398 {
399 int i;
400
401 /* zero out all of the fragment information */
402 s->coded_fragment_list_index = 0;
403 for (i = 0; i < s->fragment_count; i++) {
404 s->coeff_counts[i] = 0;
405 s->all_fragments[i].motion_x = 127;
406 s->all_fragments[i].motion_y = 127;
407 s->all_fragments[i].next_coeff= NULL;
408 s->all_fragments[i].qpi = 0;
409 s->coeffs[i].index=
410 s->coeffs[i].coeff=0;
411 s->coeffs[i].next= NULL;
412 }
413 }
414
415 /*
416 * This function sets up the dequantization tables used for a particular
417 * frame.
418 */
419 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
420 {
421 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
422 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
423 int i, plane, inter, qri, bmi, bmj, qistart;
424
425 for(inter=0; inter<2; inter++){
426 for(plane=0; plane<3; plane++){
427 int sum=0;
428 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
429 sum+= s->qr_size[inter][plane][qri];
430 if(s->qps[qpi] <= sum)
431 break;
432 }
433 qistart= sum - s->qr_size[inter][plane][qri];
434 bmi= s->qr_base[inter][plane][qri ];
435 bmj= s->qr_base[inter][plane][qri+1];
436 for(i=0; i<64; i++){
437 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
438 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
439 + s->qr_size[inter][plane][qri])
440 / (2*s->qr_size[inter][plane][qri]);
441
442 int qmin= 8<<(inter + !i);
443 int qscale= i ? ac_scale_factor : dc_scale_factor;
444
445 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
446 }
447 // all DC coefficients use the same quant so as not to interfere with DC prediction
448 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
449 }
450 }
451
452 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
453 }
454
455 /*
456 * This function initializes the loop filter boundary limits if the frame's
457 * quality index is different from the previous frame's.
458 *
459 * The filter_limit_values may not be larger than 127.
460 */
461 static void init_loop_filter(Vp3DecodeContext *s)
462 {
463 int *bounding_values= s->bounding_values_array+127;
464 int filter_limit;
465 int x;
466 int value;
467
468 filter_limit = s->filter_limit_values[s->qps[0]];
469
470 /* set up the bounding values */
471 memset(s->bounding_values_array, 0, 256 * sizeof(int));
472 for (x = 0; x < filter_limit; x++) {
473 bounding_values[-x] = -x;
474 bounding_values[x] = x;
475 }
476 for (x = value = filter_limit; x < 128 && value; x++, value--) {
477 bounding_values[ x] = value;
478 bounding_values[-x] = -value;
479 }
480 if (value)
481 bounding_values[128] = value;
482 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
483 }
484
485 /*
486 * This function unpacks all of the superblock/macroblock/fragment coding
487 * information from the bitstream.
488 */
489 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
490 {
491 int bit = 0;
492 int current_superblock = 0;
493 int current_run = 0;
494 int decode_fully_flags = 0;
495 int decode_partial_blocks = 0;
496 int first_c_fragment_seen;
497
498 int i, j;
499 int current_fragment;
500
501 if (s->keyframe) {
502 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
503
504 } else {
505
506 /* unpack the list of partially-coded superblocks */
507 bit = get_bits1(gb);
508 /* toggle the bit because as soon as the first run length is
509 * fetched the bit will be toggled again */
510 bit ^= 1;
511 while (current_superblock < s->superblock_count) {
512 if (current_run-- == 0) {
513 bit ^= 1;
514 current_run = get_vlc2(gb,
515 s->superblock_run_length_vlc.table, 6, 2);
516 if (current_run == 33)
517 current_run += get_bits(gb, 12);
518
519 /* if any of the superblocks are not partially coded, flag
520 * a boolean to decode the list of fully-coded superblocks */
521 if (bit == 0) {
522 decode_fully_flags = 1;
523 } else {
524
525 /* make a note of the fact that there are partially coded
526 * superblocks */
527 decode_partial_blocks = 1;
528 }
529 }
530 s->superblock_coding[current_superblock++] = bit;
531 }
532
533 /* unpack the list of fully coded superblocks if any of the blocks were
534 * not marked as partially coded in the previous step */
535 if (decode_fully_flags) {
536
537 current_superblock = 0;
538 current_run = 0;
539 bit = get_bits1(gb);
540 /* toggle the bit because as soon as the first run length is
541 * fetched the bit will be toggled again */
542 bit ^= 1;
543 while (current_superblock < s->superblock_count) {
544
545 /* skip any superblocks already marked as partially coded */
546 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
547
548 if (current_run-- == 0) {
549 bit ^= 1;
550 current_run = get_vlc2(gb,
551 s->superblock_run_length_vlc.table, 6, 2);
552 if (current_run == 33)
553 current_run += get_bits(gb, 12);
554 }
555 s->superblock_coding[current_superblock] = 2*bit;
556 }
557 current_superblock++;
558 }
559 }
560
561 /* if there were partial blocks, initialize bitstream for
562 * unpacking fragment codings */
563 if (decode_partial_blocks) {
564
565 current_run = 0;
566 bit = get_bits1(gb);
567 /* toggle the bit because as soon as the first run length is
568 * fetched the bit will be toggled again */
569 bit ^= 1;
570 }
571 }
572
573 /* figure out which fragments are coded; iterate through each
574 * superblock (all planes) */
575 s->coded_fragment_list_index = 0;
576 s->next_coeff= s->coeffs + s->fragment_count;
577 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
578 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
579 first_c_fragment_seen = 0;
580 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
581 for (i = 0; i < s->superblock_count; i++) {
582
583 /* iterate through all 16 fragments in a superblock */
584 for (j = 0; j < 16; j++) {
585
586 /* if the fragment is in bounds, check its coding status */
587 current_fragment = s->superblock_fragments[i * 16 + j];
588 if (current_fragment >= s->fragment_count) {
589 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
590 current_fragment, s->fragment_count);
591 return 1;
592 }
593 if (current_fragment != -1) {
594 if (s->superblock_coding[i] == SB_NOT_CODED) {
595
596 /* copy all the fragments from the prior frame */
597 s->all_fragments[current_fragment].coding_method =
598 MODE_COPY;
599
600 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
601
602 /* fragment may or may not be coded; this is the case
603 * that cares about the fragment coding runs */
604 if (current_run-- == 0) {
605 bit ^= 1;
606 current_run = get_vlc2(gb,
607 s->fragment_run_length_vlc.table, 5, 2);
608 }
609
610 if (bit) {
611 /* default mode; actual mode will be decoded in
612 * the next phase */
613 s->all_fragments[current_fragment].coding_method =
614 MODE_INTER_NO_MV;
615 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
616 s->coded_fragment_list[s->coded_fragment_list_index] =
617 current_fragment;
618 if ((current_fragment >= s->fragment_start[1]) &&
619 (s->last_coded_y_fragment == -1) &&
620 (!first_c_fragment_seen)) {
621 s->first_coded_c_fragment = s->coded_fragment_list_index;
622 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
623 first_c_fragment_seen = 1;
624 }
625 s->coded_fragment_list_index++;
626 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
627 } else {
628 /* not coded; copy this fragment from the prior frame */
629 s->all_fragments[current_fragment].coding_method =
630 MODE_COPY;
631 }
632
633 } else {
634
635 /* fragments are fully coded in this superblock; actual
636 * coding will be determined in next step */
637 s->all_fragments[current_fragment].coding_method =
638 MODE_INTER_NO_MV;
639 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
640 s->coded_fragment_list[s->coded_fragment_list_index] =
641 current_fragment;
642 if ((current_fragment >= s->fragment_start[1]) &&
643 (s->last_coded_y_fragment == -1) &&
644 (!first_c_fragment_seen)) {
645 s->first_coded_c_fragment = s->coded_fragment_list_index;
646 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
647 first_c_fragment_seen = 1;
648 }
649 s->coded_fragment_list_index++;
650 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
651 }
652 }
653 }
654 }
655
656 if (!first_c_fragment_seen)
657 /* only Y fragments coded in this frame */
658 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
659 else
660 /* end the list of coded C fragments */
661 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
662
663 for (i = 0; i < s->fragment_count - 1; i++) {
664 s->fast_fragment_list[i] = i + 1;
665 }
666 s->fast_fragment_list[s->fragment_count - 1] = -1;
667
668 if (s->last_coded_y_fragment == -1)
669 s->fragment_list_y_head = -1;
670 else {
671 s->fragment_list_y_head = s->first_coded_y_fragment;
672 s->fast_fragment_list[s->last_coded_y_fragment] = -1;
673 }
674
675 if (s->last_coded_c_fragment == -1)
676 s->fragment_list_c_head = -1;
677 else {
678 s->fragment_list_c_head = s->first_coded_c_fragment;
679 s->fast_fragment_list[s->last_coded_c_fragment] = -1;
680 }
681
682 return 0;
683 }
684
685 /*
686 * This function unpacks all the coding mode data for individual macroblocks
687 * from the bitstream.
688 */
689 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
690 {
691 int i, j, k, sb_x, sb_y;
692 int scheme;
693 int current_macroblock;
694 int current_fragment;
695 int coding_mode;
696 int custom_mode_alphabet[CODING_MODE_COUNT];
697
698 if (s->keyframe) {
699 for (i = 0; i < s->fragment_count; i++)
700 s->all_fragments[i].coding_method = MODE_INTRA;
701
702 } else {
703
704 /* fetch the mode coding scheme for this frame */
705 scheme = get_bits(gb, 3);
706
707 /* is it a custom coding scheme? */
708 if (scheme == 0) {
709 for (i = 0; i < 8; i++)
710 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
711 for (i = 0; i < 8; i++)
712 custom_mode_alphabet[get_bits(gb, 3)] = i;
713 }
714
715 /* iterate through all of the macroblocks that contain 1 or more
716 * coded fragments */
717 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
718 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
719
720 for (j = 0; j < 4; j++) {
721 int mb_x = 2*sb_x + (j>>1);
722 int mb_y = 2*sb_y + (((j>>1)+j)&1);
723 current_macroblock = mb_y * s->macroblock_width + mb_x;
724
725 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
726 (s->macroblock_coding[current_macroblock] == MODE_COPY))
727 continue;
728
729 #define BLOCK_X (2*mb_x + (k&1))
730 #define BLOCK_Y (2*mb_y + (k>>1))
731
732 /* mode 7 means get 3 bits for each coding mode */
733 if (scheme == 7)
734 coding_mode = get_bits(gb, 3);
735 else if(scheme == 0)
736 coding_mode = custom_mode_alphabet
737 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
738 else
739 coding_mode = ModeAlphabet[scheme-1]
740 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
741
742 s->macroblock_coding[current_macroblock] = coding_mode;
743 for (k = 0; k < 4; k++) {
744 current_fragment =
745 BLOCK_Y*s->fragment_width + BLOCK_X;
746 if (s->all_fragments[current_fragment].coding_method !=
747 MODE_COPY)
748 s->all_fragments[current_fragment].coding_method =
749 coding_mode;
750 }
751 for (k = 0; k < 2; k++) {
752 current_fragment = s->fragment_start[k+1] +
753 mb_y*(s->fragment_width>>1) + mb_x;
754 if (s->all_fragments[current_fragment].coding_method !=
755 MODE_COPY)
756 s->all_fragments[current_fragment].coding_method =
757 coding_mode;
758 }
759 }
760 }
761 }
762 }
763
764 return 0;
765 }
766
767 /*
768 * This function unpacks all the motion vectors for the individual
769 * macroblocks from the bitstream.
770 */
771 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
772 {
773 int j, k, l, sb_x, sb_y;
774 int coding_mode;
775 int motion_x[6];
776 int motion_y[6];
777 int last_motion_x = 0;
778 int last_motion_y = 0;
779 int prior_last_motion_x = 0;
780 int prior_last_motion_y = 0;
781 int current_macroblock;
782 int current_fragment;
783
784 if (s->keyframe)
785 return 0;
786
787 memset(motion_x, 0, 6 * sizeof(int));
788 memset(motion_y, 0, 6 * sizeof(int));
789
790 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
791 coding_mode = get_bits1(gb);
792
793 /* iterate through all of the macroblocks that contain 1 or more
794 * coded fragments */
795 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
796 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
797
798 for (j = 0; j < 4; j++) {
799 int mb_x = 2*sb_x + (j>>1);
800 int mb_y = 2*sb_y + (((j>>1)+j)&1);
801 current_macroblock = mb_y * s->macroblock_width + mb_x;
802
803 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
804 (s->macroblock_coding[current_macroblock] == MODE_COPY))
805 continue;
806
807 switch (s->macroblock_coding[current_macroblock]) {
808
809 case MODE_INTER_PLUS_MV:
810 case MODE_GOLDEN_MV:
811 /* all 6 fragments use the same motion vector */
812 if (coding_mode == 0) {
813 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
814 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
815 } else {
816 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
817 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
818 }
819
820 /* vector maintenance, only on MODE_INTER_PLUS_MV */
821 if (s->macroblock_coding[current_macroblock] ==
822 MODE_INTER_PLUS_MV) {
823 prior_last_motion_x = last_motion_x;
824 prior_last_motion_y = last_motion_y;
825 last_motion_x = motion_x[0];
826 last_motion_y = motion_y[0];
827 }
828 break;
829
830 case MODE_INTER_FOURMV:
831 /* vector maintenance */
832 prior_last_motion_x = last_motion_x;
833 prior_last_motion_y = last_motion_y;
834
835 /* fetch 4 vectors from the bitstream, one for each
836 * Y fragment, then average for the C fragment vectors */
837 motion_x[4] = motion_y[4] = 0;
838 for (k = 0; k < 4; k++) {
839 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
840 for (l = 0; l < s->coded_fragment_list_index; l++)
841 if (s->coded_fragment_list[l] == current_fragment)
842 break;
843 if (l < s->coded_fragment_list_index) {
844 if (coding_mode == 0) {
845 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
846 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
847 } else {
848 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
849 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
850 }
851 last_motion_x = motion_x[k];
852 last_motion_y = motion_y[k];
853 } else {
854 motion_x[k] = 0;
855 motion_y[k] = 0;
856 }
857 motion_x[4] += motion_x[k];
858 motion_y[4] += motion_y[k];
859 }
860
861 motion_x[5]=
862 motion_x[4]= RSHIFT(motion_x[4], 2);
863 motion_y[5]=
864 motion_y[4]= RSHIFT(motion_y[4], 2);
865 break;
866
867 case MODE_INTER_LAST_MV:
868 /* all 6 fragments use the last motion vector */
869 motion_x[0] = last_motion_x;
870 motion_y[0] = last_motion_y;
871
872 /* no vector maintenance (last vector remains the
873 * last vector) */
874 break;
875
876 case MODE_INTER_PRIOR_LAST:
877 /* all 6 fragments use the motion vector prior to the
878 * last motion vector */
879 motion_x[0] = prior_last_motion_x;
880 motion_y[0] = prior_last_motion_y;
881
882 /* vector maintenance */
883 prior_last_motion_x = last_motion_x;
884 prior_last_motion_y = last_motion_y;
885 last_motion_x = motion_x[0];
886 last_motion_y = motion_y[0];
887 break;
888
889 default:
890 /* covers intra, inter without MV, golden without MV */
891 motion_x[0] = 0;
892 motion_y[0] = 0;
893
894 /* no vector maintenance */
895 break;
896 }
897
898 /* assign the motion vectors to the correct fragments */
899 for (k = 0; k < 4; k++) {
900 current_fragment =
901 BLOCK_Y*s->fragment_width + BLOCK_X;
902 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
903 s->all_fragments[current_fragment].motion_x = motion_x[k];
904 s->all_fragments[current_fragment].motion_y = motion_y[k];
905 } else {
906 s->all_fragments[current_fragment].motion_x = motion_x[0];
907 s->all_fragments[current_fragment].motion_y = motion_y[0];
908 }
909 }
910 for (k = 0; k < 2; k++) {
911 current_fragment = s->fragment_start[k+1] +
912 mb_y*(s->fragment_width>>1) + mb_x;
913 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
914 s->all_fragments[current_fragment].motion_x = motion_x[k+4];
915 s->all_fragments[current_fragment].motion_y = motion_y[k+4];
916 } else {
917 s->all_fragments[current_fragment].motion_x = motion_x[0];
918 s->all_fragments[current_fragment].motion_y = motion_y[0];
919 }
920 }
921 }
922 }
923 }
924
925 return 0;
926 }
927
928 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
929 {
930 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
931 int num_blocks = s->coded_fragment_list_index;
932
933 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
934 i = blocks_decoded = num_blocks_at_qpi = 0;
935
936 bit = get_bits1(gb);
937
938 do {
939 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
940 if (run_length == 34)
941 run_length += get_bits(gb, 12);
942 blocks_decoded += run_length;
943
944 if (!bit)
945 num_blocks_at_qpi += run_length;
946
947 for (j = 0; j < run_length; i++) {
948 if (i >= s->coded_fragment_list_index)
949 return -1;
950
951 if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
952 s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
953 j++;
954 }
955 }
956
957 if (run_length == 4129)
958 bit = get_bits1(gb);
959 else
960 bit ^= 1;
961 } while (blocks_decoded < num_blocks);
962
963 num_blocks -= num_blocks_at_qpi;
964 }
965
966 return 0;
967 }
968
969 /*
970 * This function is called by unpack_dct_coeffs() to extract the VLCs from
971 * the bitstream. The VLCs encode tokens which are used to unpack DCT
972 * data. This function unpacks all the VLCs for either the Y plane or both
973 * C planes, and is called for DC coefficients or different AC coefficient
974 * levels (since different coefficient types require different VLC tables.
975 *
976 * This function returns a residual eob run. E.g, if a particular token gave
977 * instructions to EOB the next 5 fragments and there were only 2 fragments
978 * left in the current fragment range, 3 would be returned so that it could
979 * be passed into the next call to this same function.
980 */
981 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
982 VLC *table, int coeff_index,
983 int y_plane,
984 int eob_run)
985 {
986 int i;
987 int token;
988 int zero_run = 0;
989 DCTELEM coeff = 0;
990 Vp3Fragment *fragment;
991 int bits_to_get;
992 int next_fragment;
993 int previous_fragment;
994 int fragment_num;
995 int *list_head;
996
997 /* local references to structure members to avoid repeated deferences */
998 uint8_t *perm= s->scantable.permutated;
999 int *coded_fragment_list = s->coded_fragment_list;
1000 Vp3Fragment *all_fragments = s->all_fragments;
1001 uint8_t *coeff_counts = s->coeff_counts;
1002 VLC_TYPE (*vlc_table)[2] = table->table;
1003 int *fast_fragment_list = s->fast_fragment_list;
1004
1005 if (y_plane) {
1006 next_fragment = s->fragment_list_y_head;
1007 list_head = &s->fragment_list_y_head;
1008 } else {
1009 next_fragment = s->fragment_list_c_head;
1010 list_head = &s->fragment_list_c_head;
1011 }
1012
1013 i = next_fragment;
1014 previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */
1015 while (i != -1) {
1016 fragment_num = coded_fragment_list[i];
1017
1018 if (coeff_counts[fragment_num] > coeff_index) {
1019 previous_fragment = i;
1020 i = fast_fragment_list[i];
1021 continue;
1022 }
1023 fragment = &all_fragments[fragment_num];
1024
1025 if (!eob_run) {
1026 /* decode a VLC into a token */
1027 token = get_vlc2(gb, vlc_table, 5, 3);
1028 /* use the token to get a zero run, a coefficient, and an eob run */
1029 if (token <= 6) {
1030 eob_run = eob_run_base[token];
1031 if (eob_run_get_bits[token])
1032 eob_run += get_bits(gb, eob_run_get_bits[token]);
1033 coeff = zero_run = 0;
1034 } else {
1035 bits_to_get = coeff_get_bits[token];
1036 if (bits_to_get)
1037 bits_to_get = get_bits(gb, bits_to_get);
1038 coeff = coeff_tables[token][bits_to_get];
1039
1040 zero_run = zero_run_base[token];
1041 if (zero_run_get_bits[token])
1042 zero_run += get_bits(gb, zero_run_get_bits[token]);
1043 }
1044 }
1045
1046 if (!eob_run) {
1047 coeff_counts[fragment_num] += zero_run;
1048 if (coeff_counts[fragment_num] < 64){
1049 fragment->next_coeff->coeff= coeff;
1050 fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1051 fragment->next_coeff->next= s->next_coeff;
1052 s->next_coeff->next=NULL;
1053 fragment->next_coeff= s->next_coeff++;
1054 }
1055 /* previous fragment is now this fragment */
1056 previous_fragment = i;
1057 } else {
1058 coeff_counts[fragment_num] |= 128;
1059 eob_run--;
1060 /* remove this fragment from the list */
1061 if (previous_fragment != -1)
1062 fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1063 else
1064 *list_head = fast_fragment_list[i];
1065 /* previous fragment remains unchanged */
1066 }
1067
1068 i = fast_fragment_list[i];
1069 }
1070
1071 return eob_run;
1072 }
1073
1074 static void reverse_dc_prediction(Vp3DecodeContext *s,
1075 int first_fragment,
1076 int fragment_width,
1077 int fragment_height);
1078 /*
1079 * This function unpacks all of the DCT coefficient data from the
1080 * bitstream.
1081 */
1082 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1083 {
1084 int i;
1085 int dc_y_table;
1086 int dc_c_table;
1087 int ac_y_table;
1088 int ac_c_table;
1089 int residual_eob_run = 0;
1090 VLC *y_tables[64];
1091 VLC *c_tables[64];
1092
1093 /* fetch the DC table indexes */
1094 dc_y_table = get_bits(gb, 4);
1095 dc_c_table = get_bits(gb, 4);
1096
1097 /* unpack the Y plane DC coefficients */
1098 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1099 1, residual_eob_run);
1100
1101 /* reverse prediction of the Y-plane DC coefficients */
1102 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1103
1104 /* unpack the C plane DC coefficients */
1105 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1106 0, residual_eob_run);
1107
1108 /* reverse prediction of the C-plane DC coefficients */
1109 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1110 {
1111 reverse_dc_prediction(s, s->fragment_start[1],
1112 s->fragment_width / 2, s->fragment_height / 2);
1113 reverse_dc_prediction(s, s->fragment_start[2],
1114 s->fragment_width / 2, s->fragment_height / 2);
1115 }
1116
1117 /* fetch the AC table indexes */
1118 ac_y_table = get_bits(gb, 4);
1119 ac_c_table = get_bits(gb, 4);
1120
1121 /* build tables of AC VLC tables */
1122 for (i = 1; i <= 5; i++) {
1123 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1124 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1125 }
1126 for (i = 6; i <= 14; i++) {
1127 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1128 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1129 }
1130 for (i = 15; i <= 27; i++) {
1131 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1132 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1133 }
1134 for (i = 28; i <= 63; i++) {
1135 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1136 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1137 }
1138
1139 /* decode all AC coefficents */
1140 for (i = 1; i <= 63; i++) {
1141 if (s->fragment_list_y_head != -1)
1142 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1143 1, residual_eob_run);
1144
1145 if (s->fragment_list_c_head != -1)
1146 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1147 0, residual_eob_run);
1148 }
1149
1150 return 0;
1151 }
1152
1153 /*
1154 * This function reverses the DC prediction for each coded fragment in
1155 * the frame. Much of this function is adapted directly from the original
1156 * VP3 source code.
1157 */
1158 #define COMPATIBLE_FRAME(x) \
1159 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1160 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1161
1162 static void reverse_dc_prediction(Vp3DecodeContext *s,
1163 int first_fragment,
1164 int fragment_width,
1165 int fragment_height)
1166 {
1167
1168 #define PUL 8
1169 #define PU 4
1170 #define PUR 2
1171 #define PL 1
1172
1173 int x, y;
1174 int i = first_fragment;
1175
1176 int predicted_dc;
1177
1178 /* DC values for the left, up-left, up, and up-right fragments */
1179 int vl, vul, vu, vur;
1180
1181 /* indexes for the left, up-left, up, and up-right fragments */
1182 int l, ul, u, ur;
1183
1184 /*
1185 * The 6 fields mean:
1186 * 0: up-left multiplier
1187 * 1: up multiplier
1188 * 2: up-right multiplier
1189 * 3: left multiplier
1190 */
1191 static const int predictor_transform[16][4] = {
1192 { 0, 0, 0, 0},
1193 { 0, 0, 0,128}, // PL
1194 { 0, 0,128, 0}, // PUR
1195 { 0, 0, 53, 75}, // PUR|PL
1196 { 0,128, 0, 0}, // PU
1197 { 0, 64, 0, 64}, // PU|PL
1198 { 0,128, 0, 0}, // PU|PUR
1199 { 0, 0, 53, 75}, // PU|PUR|PL
1200 {128, 0, 0, 0}, // PUL
1201 { 0, 0, 0,128}, // PUL|PL
1202 { 64, 0, 64, 0}, // PUL|PUR
1203 { 0, 0, 53, 75}, // PUL|PUR|PL
1204 { 0,128, 0, 0}, // PUL|PU
1205 {-104,116, 0,116}, // PUL|PU|PL
1206 { 24, 80, 24, 0}, // PUL|PU|PUR
1207 {-104,116, 0,116} // PUL|PU|PUR|PL
1208 };
1209
1210 /* This table shows which types of blocks can use other blocks for
1211 * prediction. For example, INTRA is the only mode in this table to
1212 * have a frame number of 0. That means INTRA blocks can only predict
1213 * from other INTRA blocks. There are 2 golden frame coding types;
1214 * blocks encoding in these modes can only predict from other blocks
1215 * that were encoded with these 1 of these 2 modes. */
1216 static const unsigned char compatible_frame[9] = {
1217 1, /* MODE_INTER_NO_MV */
1218 0, /* MODE_INTRA */
1219 1, /* MODE_INTER_PLUS_MV */
1220 1, /* MODE_INTER_LAST_MV */
1221 1, /* MODE_INTER_PRIOR_MV */
1222 2, /* MODE_USING_GOLDEN */
1223 2, /* MODE_GOLDEN_MV */
1224 1, /* MODE_INTER_FOUR_MV */
1225 3 /* MODE_COPY */
1226 };
1227 int current_frame_type;
1228
1229 /* there is a last DC predictor for each of the 3 frame types */
1230 short last_dc[3];
1231
1232 int transform = 0;
1233
1234 vul = vu = vur = vl = 0;
1235 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1236
1237 /* for each fragment row... */
1238 for (y = 0; y < fragment_height; y++) {
1239
1240 /* for each fragment in a row... */
1241 for (x = 0; x < fragment_width; x++, i++) {
1242
1243 /* reverse prediction if this block was coded */
1244 if (s->all_fragments[i].coding_method != MODE_COPY) {
1245
1246 current_frame_type =
1247 compatible_frame[s->all_fragments[i].coding_method];
1248
1249 transform= 0;
1250 if(x){
1251 l= i-1;
1252 vl = DC_COEFF(l);
1253 if(COMPATIBLE_FRAME(l))
1254 transform |= PL;
1255 }
1256 if(y){
1257 u= i-fragment_width;
1258 vu = DC_COEFF(u);
1259 if(COMPATIBLE_FRAME(u))
1260 transform |= PU;
1261 if(x){
1262 ul= i-fragment_width-1;
1263 vul = DC_COEFF(ul);
1264 if(COMPATIBLE_FRAME(ul))
1265 transform |= PUL;
1266 }
1267 if(x + 1 < fragment_width){
1268 ur= i-fragment_width+1;
1269 vur = DC_COEFF(ur);
1270 if(COMPATIBLE_FRAME(ur))
1271 transform |= PUR;
1272 }
1273 }
1274
1275 if (transform == 0) {
1276
1277 /* if there were no fragments to predict from, use last
1278 * DC saved */
1279 predicted_dc = last_dc[current_frame_type];
1280 } else {
1281
1282 /* apply the appropriate predictor transform */
1283 predicted_dc =
1284 (predictor_transform[transform][0] * vul) +
1285 (predictor_transform[transform][1] * vu) +
1286 (predictor_transform[transform][2] * vur) +
1287 (predictor_transform[transform][3] * vl);
1288
1289 predicted_dc /= 128;
1290
1291 /* check for outranging on the [ul u l] and
1292 * [ul u ur l] predictors */
1293 if ((transform == 15) || (transform == 13)) {
1294 if (FFABS(predicted_dc - vu) > 128)
1295 predicted_dc = vu;
1296 else if (FFABS(predicted_dc - vl) > 128)
1297 predicted_dc = vl;
1298 else if (FFABS(predicted_dc - vul) > 128)
1299 predicted_dc = vul;
1300 }
1301 }
1302
1303 /* at long last, apply the predictor */
1304 if(s->coeffs[i].index){
1305 *s->next_coeff= s->coeffs[i];
1306 s->coeffs[i].index=0;
1307 s->coeffs[i].coeff=0;
1308 s->coeffs[i].next= s->next_coeff++;
1309 }
1310 s->coeffs[i].coeff += predicted_dc;
1311 /* save the DC */
1312 last_dc[current_frame_type] = DC_COEFF(i);
1313 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1314 s->coeff_counts[i]= 129;
1315 // s->all_fragments[i].next_coeff= s->next_coeff;
1316 s->coeffs[i].next= s->next_coeff;
1317 (s->next_coeff++)->next=NULL;
1318 }
1319 }
1320 }
1321 }
1322 }
1323
1324 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1325 {
1326 int x, y;
1327 int *bounding_values= s->bounding_values_array+127;
1328
1329 int width = s->fragment_width >> !!plane;
1330 int height = s->fragment_height >> !!plane;
1331 int fragment = s->fragment_start [plane] + ystart * width;
1332 int stride = s->current_frame.linesize[plane];
1333 uint8_t *plane_data = s->current_frame.data [plane];
1334 if (!s->flipped_image) stride = -stride;
1335 plane_data += s->data_offset[plane] + 8*ystart*stride;
1336
1337 for (y = ystart; y < yend; y++) {
1338
1339 for (x = 0; x < width; x++) {
1340 /* This code basically just deblocks on the edges of coded blocks.
1341 * However, it has to be much more complicated because of the
1342 * braindamaged deblock ordering used in VP3/Theora. Order matters
1343 * because some pixels get filtered twice. */
1344 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1345 {
1346 /* do not perform left edge filter for left columns frags */
1347 if (x > 0) {
1348 s->dsp.vp3_h_loop_filter(
1349 plane_data + 8*x,
1350 stride, bounding_values);
1351 }
1352
1353 /* do not perform top edge filter for top row fragments */
1354 if (y > 0) {
1355 s->dsp.vp3_v_loop_filter(
1356 plane_data + 8*x,
1357 stride, bounding_values);
1358 }
1359
1360 /* do not perform right edge filter for right column
1361 * fragments or if right fragment neighbor is also coded
1362 * in this frame (it will be filtered in next iteration) */
1363 if ((x < width - 1) &&
1364 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1365 s->dsp.vp3_h_loop_filter(
1366 plane_data + 8*x + 8,
1367 stride, bounding_values);
1368 }
1369
1370 /* do not perform bottom edge filter for bottom row
1371 * fragments or if bottom fragment neighbor is also coded
1372 * in this frame (it will be filtered in the next row) */
1373 if ((y < height - 1) &&
1374 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1375 s->dsp.vp3_v_loop_filter(
1376 plane_data + 8*x + 8*stride,
1377 stride, bounding_values);
1378 }
1379 }
1380
1381 fragment++;
1382 }
1383 plane_data += 8*stride;
1384 }
1385 }
1386
1387 /**
1388 * called when all pixels up to row y are complete
1389 */
1390 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1391 {
1392 int h, cy;
1393 int offset[4];
1394
1395 if(s->avctx->draw_horiz_band==NULL)
1396 return;
1397
1398 h= y - s->last_slice_end;
1399 y -= h;
1400
1401 if (!s->flipped_image) {
1402 if (y == 0)
1403 h -= s->height - s->avctx->height; // account for non-mod16
1404 y = s->height - y - h;
1405 }
1406
1407 cy = y >> 1;
1408 offset[0] = s->current_frame.linesize[0]*y;
1409 offset[1] = s->current_frame.linesize[1]*cy;
1410 offset[2] = s->current_frame.linesize[2]*cy;
1411 offset[3] = 0;
1412
1413 emms_c();
1414 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1415 s->last_slice_end= y + h;
1416 }
1417
1418 /*
1419 * Perform the final rendering for a particular slice of data.
1420 * The slice number ranges from 0..(macroblock_height - 1).
1421 */
1422 static void render_slice(Vp3DecodeContext *s, int slice)
1423 {
1424 int x;
1425 int16_t *dequantizer;
1426 DECLARE_ALIGNED_16(DCTELEM, block)[64];
1427 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1428 int motion_halfpel_index;
1429 uint8_t *motion_source;
1430 int plane;
1431
1432 if (slice >= s->macroblock_height)
1433 return;
1434
1435 for (plane = 0; plane < 3; plane++) {
1436 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1437 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1438 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1439 int stride = s->current_frame.linesize[plane];
1440 int plane_width = s->width >> !!plane;
1441 int plane_height = s->height >> !!plane;
1442 int y = slice * FRAGMENT_PIXELS << !plane ;
1443 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1444 int i = s->fragment_start[plane] + (y>>3)*(s->fragment_width>>!!plane);
1445
1446 if (!s->flipped_image) stride = -stride;
1447
1448
1449 if(FFABS(stride) > 2048)
1450 return; //various tables are fixed size
1451
1452 /* for each fragment row in the slice (both of them)... */
1453 for (; y < slice_height; y += 8) {
1454
1455 /* for each fragment in a row... */
1456 for (x = 0; x < plane_width; x += 8, i++) {
1457 int first_pixel = y*stride + x;
1458
1459 if ((i < 0) || (i >= s->fragment_count)) {
1460 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1461 return;
1462 }
1463
1464 /* transform if this block was coded */
1465 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1466 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1467
1468 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1469 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1470 motion_source= golden_plane;
1471 else
1472 motion_source= last_plane;
1473
1474 motion_source += first_pixel;
1475 motion_halfpel_index = 0;
1476
1477 /* sort out the motion vector if this fragment is coded
1478 * using a motion vector method */
1479 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1480 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1481 int src_x, src_y;
1482 motion_x = s->all_fragments[i].motion_x;
1483 motion_y = s->all_fragments[i].motion_y;
1484 if(plane){
1485 motion_x= (motion_x>>1) | (motion_x&1);
1486 motion_y= (motion_y>>1) | (motion_y&1);
1487 }
1488
1489 src_x= (motion_x>>1) + x;
1490 src_y= (motion_y>>1) + y;
1491 if ((motion_x == 127) || (motion_y == 127))
1492 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1493
1494 motion_halfpel_index = motion_x & 0x01;
1495 motion_source += (motion_x >> 1);
1496
1497 motion_halfpel_index |= (motion_y & 0x01) << 1;
1498 motion_source += ((motion_y >> 1) * stride);
1499
1500 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1501 uint8_t *temp= s->edge_emu_buffer;
1502 if(stride<0) temp -= 9*stride;
1503 else temp += 9*stride;
1504
1505 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1506 motion_source= temp;
1507 }
1508 }
1509
1510
1511 /* first, take care of copying a block from either the
1512 * previous or the golden frame */
1513 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1514 /* Note, it is possible to implement all MC cases with
1515 put_no_rnd_pixels_l2 which would look more like the
1516 VP3 source but this would be slower as
1517 put_no_rnd_pixels_tab is better optimzed */
1518 if(motion_halfpel_index != 3){
1519 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1520 output_plane + first_pixel,
1521 motion_source, stride, 8);
1522 }else{
1523 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1524 s->dsp.put_no_rnd_pixels_l2[1](
1525 output_plane + first_pixel,
1526 motion_source - d,
1527 motion_source + stride + 1 + d,
1528 stride, 8);
1529 }
1530 dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1531 }else{
1532 dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1533 }
1534
1535 /* dequantize the DCT coefficients */
1536 if(s->avctx->idct_algo==FF_IDCT_VP3){
1537 Coeff *coeff= s->coeffs + i;
1538 s->dsp.clear_block(block);
1539 while(coeff->next){
1540 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1541 coeff= coeff->next;
1542 }
1543 }else{
1544 Coeff *coeff= s->coeffs + i;
1545 s->dsp.clear_block(block);
1546 while(coeff->next){
1547 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1548 coeff= coeff->next;
1549 }
1550 }
1551
1552 /* invert DCT and place (or add) in final output */
1553
1554 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1555 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1556 block[0] += 128<<3;
1557 s->dsp.idct_put(
1558 output_plane + first_pixel,
1559 stride,
1560 block);
1561 } else {
1562 s->dsp.idct_add(
1563 output_plane + first_pixel,
1564 stride,
1565 block);
1566 }
1567 } else {
1568
1569 /* copy directly from the previous frame */
1570 s->dsp.put_pixels_tab[1][0](
1571 output_plane + first_pixel,
1572 last_plane + first_pixel,
1573 stride, 8);
1574
1575 }
1576 }
1577 // Filter the previous block row. We can't filter the current row yet
1578 // since it needs pixels from the next row
1579 if (y > 0)
1580 apply_loop_filter(s, plane, (y>>3)-1, (y>>3));
1581 }
1582 }
1583
1584 /* this looks like a good place for slice dispatch... */
1585 /* algorithm:
1586 * if (slice == s->macroblock_height - 1)
1587 * dispatch (both last slice & 2nd-to-last slice);
1588 * else if (slice > 0)
1589 * dispatch (slice - 1);
1590 */
1591
1592 // now that we've filtered the last rows, they're safe to display
1593 if (slice)
1594 vp3_draw_horiz_band(s, 16*slice);
1595 }
1596
1597 /*
1598 * This is the ffmpeg/libavcodec API init function.
1599 */
1600 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1601 {
1602 Vp3DecodeContext *s = avctx->priv_data;
1603 int i, inter, plane;
1604 int c_width;
1605 int c_height;
1606 int y_superblock_count;
1607 int c_superblock_count;
1608
1609 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1610 s->version = 0;
1611 else
1612 s->version = 1;
1613
1614 s->avctx = avctx;
1615 s->width = FFALIGN(avctx->width, 16);
1616 s->height = FFALIGN(avctx->height, 16);
1617 avctx->pix_fmt = PIX_FMT_YUV420P;
1618 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1619 if(avctx->idct_algo==FF_IDCT_AUTO)
1620 avctx->idct_algo=FF_IDCT_VP3;
1621 dsputil_init(&s->dsp, avctx);
1622
1623 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1624
1625 /* initialize to an impossible value which will force a recalculation
1626 * in the first frame decode */
1627 for (i = 0; i < 3; i++)
1628 s->qps[i] = -1;
1629
1630 s->y_superblock_width = (s->width + 31) / 32;
1631 s->y_superblock_height = (s->height + 31) / 32;
1632 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1633
1634 /* work out the dimensions for the C planes */
1635 c_width = s->width / 2;
1636 c_height = s->height / 2;
1637 s->c_superblock_width = (c_width + 31) / 32;
1638 s->c_superblock_height = (c_height + 31) / 32;
1639 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1640
1641 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1642 s->u_superblock_start = y_superblock_count;
1643 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1644 s->superblock_coding = av_malloc(s->superblock_count);
1645
1646 s->macroblock_width = (s->width + 15) / 16;
1647 s->macroblock_height = (s->height + 15) / 16;
1648 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1649
1650 s->fragment_width = s->width / FRAGMENT_PIXELS;
1651 s->fragment_height = s->height / FRAGMENT_PIXELS;
1652
1653 /* fragment count covers all 8x8 blocks for all 3 planes */
1654 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1655 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1656 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1657
1658 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1659 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1660 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1661 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1662 s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1663 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1664 !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1665 vp3_decode_end(avctx);
1666 return -1;
1667 }
1668
1669 if (!s->theora_tables)
1670 {
1671 for (i = 0; i < 64; i++) {
1672 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1673 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1674 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1675 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1676 s->base_matrix[2][i] = vp31_inter_dequant[i];
1677 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1678 }
1679
1680 for(inter=0; inter<2; inter++){
1681 for(plane=0; plane<3; plane++){
1682 s->qr_count[inter][plane]= 1;
1683 s->qr_size [inter][plane][0]= 63;
1684 s->qr_base [inter][plane][0]=
1685 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1686 }
1687 }
1688
1689 /* init VLC tables */
1690 for (i = 0; i < 16; i++) {
1691
1692 /* DC histograms */
1693 init_vlc(&s->dc_vlc[i], 5, 32,
1694 &dc_bias[i][0][1], 4, 2,
1695 &dc_bias[i][0][0], 4, 2, 0);
1696
1697 /* group 1 AC histograms */
1698 init_vlc(&s->ac_vlc_1[i], 5, 32,
1699 &ac_bias_0[i][0][1], 4, 2,
1700 &ac_bias_0[i][0][0], 4, 2, 0);
1701
1702 /* group 2 AC histograms */
1703 init_vlc(&s->ac_vlc_2[i], 5, 32,
1704 &ac_bias_1[i][0][1], 4, 2,
1705 &ac_bias_1[i][0][0], 4, 2, 0);
1706
1707 /* group 3 AC histograms */
1708 init_vlc(&s->ac_vlc_3[i], 5, 32,
1709 &ac_bias_2[i][0][1], 4, 2,
1710 &ac_bias_2[i][0][0], 4, 2, 0);
1711
1712 /* group 4 AC histograms */
1713 init_vlc(&s->ac_vlc_4[i], 5, 32,
1714 &ac_bias_3[i][0][1], 4, 2,
1715 &ac_bias_3[i][0][0], 4, 2, 0);
1716 }
1717 } else {
1718 for (i = 0; i < 16; i++) {
1719
1720 /* DC histograms */
1721 if (init_vlc(&s->dc_vlc[i], 5, 32,
1722 &s->huffman_table[i][0][1], 4, 2,
1723 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1724 goto vlc_fail;
1725
1726 /* group 1 AC histograms */
1727 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1728 &s->huffman_table[i+16][0][1], 4, 2,
1729 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1730 goto vlc_fail;
1731
1732 /* group 2 AC histograms */
1733 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1734 &s->huffman_table[i+16*2][0][1], 4, 2,
1735 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1736 goto vlc_fail;
1737
1738 /* group 3 AC histograms */
1739 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1740 &s->huffman_table[i+16*3][0][1], 4, 2,
1741 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1742 goto vlc_fail;
1743
1744 /* group 4 AC histograms */
1745 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1746 &s->huffman_table[i+16*4][0][1], 4, 2,
1747 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1748 goto vlc_fail;
1749 }
1750 }
1751
1752 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1753 &superblock_run_length_vlc_table[0][1], 4, 2,
1754 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1755
1756 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1757 &fragment_run_length_vlc_table[0][1], 4, 2,
1758 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1759
1760 init_vlc(&s->mode_code_vlc, 3, 8,
1761 &mode_code_vlc_table[0][1], 2, 1,
1762 &mode_code_vlc_table[0][0], 2, 1, 0);
1763
1764 init_vlc(&s->motion_vector_vlc, 6, 63,
1765 &motion_vector_vlc_table[0][1], 2, 1,
1766 &motion_vector_vlc_table[0][0], 2, 1, 0);
1767
1768 /* work out the block mapping tables */
1769 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1770 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1771 if (!s->superblock_fragments || !s->macroblock_coding) {
1772 vp3_decode_end(avctx);
1773 return -1;
1774 }
1775 init_block_mapping(s);
1776
1777 for (i = 0; i < 3; i++) {
1778 s->current_frame.data[i] = NULL;
1779 s->last_frame.data[i] = NULL;
1780 s->golden_frame.data[i] = NULL;
1781 }
1782
1783 return 0;
1784
1785 vlc_fail:
1786 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1787 return -1;
1788 }
1789
1790 /*
1791 * This is the ffmpeg/libavcodec API frame decode function.
1792 */
1793 static int vp3_decode_frame(AVCodecContext *avctx,
1794 void *data, int *data_size,
1795 AVPacket *avpkt)
1796 {
1797 const uint8_t *buf = avpkt->data;
1798 int buf_size = avpkt->size;
1799 Vp3DecodeContext *s = avctx->priv_data;
1800 GetBitContext gb;
1801 static int counter = 0;
1802 int i;
1803
1804 init_get_bits(&gb, buf, buf_size * 8);
1805
1806 if (s->theora && get_bits1(&gb))
1807 {
1808 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1809 return -1;
1810 }
1811
1812 s->keyframe = !get_bits1(&gb);
1813 if (!s->theora)
1814 skip_bits(&gb, 1);
1815 for (i = 0; i < 3; i++)
1816 s->last_qps[i] = s->qps[i];
1817
1818 s->nqps=0;
1819 do{
1820 s->qps[s->nqps++]= get_bits(&gb, 6);
1821 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1822 for (i = s->nqps; i < 3; i++)
1823 s->qps[i] = -1;
1824
1825 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1826 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1827 s->keyframe?"key":"", counter, s->qps[0]);
1828 counter++;
1829
1830 if (s->qps[0] != s->last_qps[0])
1831 init_loop_filter(s);
1832
1833 for (i = 0; i < s->nqps; i++)
1834 // reinit all dequantizers if the first one changed, because
1835 // the DC of the first quantizer must be used for all matrices
1836 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1837 init_dequantizer(s, i);
1838
1839 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1840 return buf_size;
1841
1842 if (s->keyframe) {
1843 if (!s->theora)
1844 {
1845 skip_bits(&gb, 4); /* width code */
1846 skip_bits(&gb, 4); /* height code */
1847 if (s->version)
1848 {
1849 s->version = get_bits(&gb, 5);
1850 if (counter == 1)
1851 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1852 }
1853 }
1854 if (s->version || s->theora)
1855 {
1856 if (get_bits1(&gb))
1857 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1858 skip_bits(&gb, 2); /* reserved? */
1859 }
1860
1861 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1862 if (s->golden_frame.data[0])
1863 avctx->release_buffer(avctx, &s->golden_frame);
1864 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1865 } else {
1866 if (s->golden_frame.data[0])
1867 avctx->release_buffer(avctx, &s->golden_frame);
1868 if (s->last_frame.data[0])
1869 avctx->release_buffer(avctx, &s->last_frame);
1870 }
1871
1872 s->golden_frame.reference = 3;
1873 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1874 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1875 return -1;
1876 }
1877
1878 /* golden frame is also the current frame */
1879 s->current_frame= s->golden_frame;
1880 } else {
1881 /* allocate a new current frame */
1882 s->current_frame.reference = 3;
1883 if (!s->golden_frame.data[0]) {
1884 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1885 return -1;
1886 }
1887 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1888 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1889 return -1;
1890 }
1891 }
1892
1893 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1894 s->current_frame.qstride= 0;
1895
1896 init_frame(s, &gb);
1897
1898 if (unpack_superblocks(s, &gb)){
1899 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1900 return -1;
1901 }
1902 if (unpack_modes(s, &gb)){
1903 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1904 return -1;
1905 }
1906 if (unpack_vectors(s, &gb)){
1907 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1908 return -1;
1909 }
1910 if (unpack_block_qpis(s, &gb)){
1911 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1912 return -1;
1913 }
1914 if (unpack_dct_coeffs(s, &gb)){
1915 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1916 return -1;
1917 }
1918
1919 for (i = 0; i < 3; i++) {
1920 if (s->flipped_image)
1921 s->data_offset[i] = 0;
1922 else
1923 s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1924 }
1925
1926 s->last_slice_end = 0;
1927 for (i = 0; i < s->macroblock_height; i++)
1928 render_slice(s, i);
1929
1930 // filter the last row
1931 for (i = 0; i < 3; i++) {
1932 int row = (s->height >> (3+!!i)) - 1;
1933 apply_loop_filter(s, i, row, row+1);
1934 }
1935 vp3_draw_horiz_band(s, s->height);
1936
1937 *data_size=sizeof(AVFrame);
1938 *(AVFrame*)data= s->current_frame;
1939
1940 /* release the last frame, if it is allocated and if it is not the
1941 * golden frame */
1942 if ((s->last_frame.data[0]) &&
1943 (s->last_frame.data[0] != s->golden_frame.data[0]))
1944 avctx->release_buffer(avctx, &s->last_frame);
1945
1946 /* shuffle frames (last = current) */
1947 s->last_frame= s->current_frame;
1948 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1949
1950 return buf_size;
1951 }
1952
1953 /*
1954 * This is the ffmpeg/libavcodec API module cleanup function.
1955 */
1956 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1957 {
1958 Vp3DecodeContext *s = avctx->priv_data;
1959 int i;
1960
1961 av_free(s->superblock_coding);
1962 av_free(s->all_fragments);
1963 av_free(s->coeff_counts);
1964 av_free(s->coeffs);
1965 av_free(s->coded_fragment_list);
1966 av_free(s->fast_fragment_list);
1967 av_free(s->superblock_fragments);
1968 av_free(s->macroblock_coding);
1969
1970 for (i = 0; i < 16; i++) {
1971 free_vlc(&s->dc_vlc[i]);
1972 free_vlc(&s->ac_vlc_1[i]);
1973 free_vlc(&s->ac_vlc_2[i]);
1974 free_vlc(&s->ac_vlc_3[i]);
1975 free_vlc(&s->ac_vlc_4[i]);
1976 }
1977
1978 free_vlc(&s->superblock_run_length_vlc);
1979 free_vlc(&s->fragment_run_length_vlc);
1980 free_vlc(&s->mode_code_vlc);
1981 free_vlc(&s->motion_vector_vlc);
1982
1983 /* release all frames */
1984 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1985 avctx->release_buffer(avctx, &s->golden_frame);
1986 if (s->last_frame.data[0])
1987 avctx->release_buffer(avctx, &s->last_frame);
1988 /* no need to release the current_frame since it will always be pointing
1989 * to the same frame as either the golden or last frame */
1990
1991 return 0;
1992 }
1993
1994 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1995 {
1996 Vp3DecodeContext *s = avctx->priv_data;
1997
1998 if (get_bits1(gb)) {
1999 int token;
2000 if (s->entries >= 32) { /* overflow */
2001 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2002 return -1;
2003 }
2004 token = get_bits(gb, 5);
2005 //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);
2006 s->huffman_table[s->hti][token][0] = s->hbits;
2007 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2008 s->entries++;
2009 }
2010 else {
2011 if (s->huff_code_size >= 32) {/* overflow */
2012 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2013 return -1;
2014 }
2015 s->huff_code_size++;
2016 s->hbits <<= 1;
2017 if (read_huffman_tree(avctx, gb))
2018 return -1;
2019 s->hbits |= 1;
2020 if (read_huffman_tree(avctx, gb))
2021 return -1;
2022 s->hbits >>= 1;
2023 s->huff_code_size--;
2024 }
2025 return 0;
2026 }
2027
2028 #if CONFIG_THEORA_DECODER
2029 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2030 {
2031 Vp3DecodeContext *s = avctx->priv_data;
2032 int visible_width, visible_height, colorspace;
2033
2034 s->theora = get_bits_long(gb, 24);
2035 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2036
2037 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2038 /* but previous versions have the image flipped relative to vp3 */
2039 if (s->theora < 0x030200)
2040 {
2041 s->flipped_image = 1;
2042 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2043 }
2044
2045 visible_width = s->width = get_bits(gb, 16) << 4;
2046 visible_height = s->height = get_bits(gb, 16) << 4;
2047
2048 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2049 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2050 s->width= s->height= 0;
2051 return -1;
2052 }
2053
2054 if (s->theora >= 0x030200) {
2055 visible_width = get_bits_long(gb, 24);
2056 visible_height = get_bits_long(gb, 24);
2057
2058 skip_bits(gb, 8); /* offset x */
2059 skip_bits(gb, 8); /* offset y */
2060 }
2061
2062 skip_bits(gb, 32); /* fps numerator */
2063 skip_bits(gb, 32); /* fps denumerator */
2064 skip_bits(gb, 24); /* aspect numerator */
2065 skip_bits(gb, 24); /* aspect denumerator */
2066
2067 if (s->theora < 0x030200)
2068 skip_bits(gb, 5); /* keyframe frequency force */
2069 colorspace = get_bits(gb, 8);
2070 skip_bits(gb, 24); /* bitrate */
2071
2072 skip_bits(gb, 6); /* quality hint */
2073
2074 if (s->theora >= 0x030200)
2075 {
2076 skip_bits(gb, 5); /* keyframe frequency force */
2077 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2078 skip_bits(gb, 3); /* reserved */
2079 }
2080
2081 // align_get_bits(gb);
2082
2083 if ( visible_width <= s->width && visible_width > s->width-16
2084 && visible_height <= s->height && visible_height > s->height-16)
2085 avcodec_set_dimensions(avctx, visible_width, visible_height);
2086 else
2087 avcodec_set_dimensions(avctx, s->width, s->height);
2088
2089 if (colorspace == 1) {
2090 avctx->color_primaries = AVCOL_PRI_BT470M;
2091 } else if (colorspace == 2) {
2092 avctx->color_primaries = AVCOL_PRI_BT470BG;
2093 }
2094 if (colorspace == 1 || colorspace == 2) {
2095 avctx->colorspace = AVCOL_SPC_BT470BG;
2096 avctx->color_trc = AVCOL_TRC_BT709;
2097 }
2098
2099 return 0;
2100 }
2101
2102 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2103 {
2104 Vp3DecodeContext *s = avctx->priv_data;
2105 int i, n, matrices, inter, plane;
2106
2107 if (s->theora >= 0x030200) {
2108 n = get_bits(gb, 3);
2109 /* loop filter limit values table */
2110 for (i = 0; i < 64; i++) {
2111 s->filter_limit_values[i] = get_bits(gb, n);
2112 if (s->filter_limit_values[i] > 127) {
2113 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2114 s->filter_limit_values[i] = 127;
2115 }
2116 }
2117 }
2118
2119 if (s->theora >= 0x030200)
2120 n = get_bits(gb, 4) + 1;
2121 else
2122 n = 16;
2123 /* quality threshold table */
2124 for (i = 0; i < 64; i++)
2125 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2126
2127 if (s->theora >= 0x030200)
2128 n = get_bits(gb, 4) + 1;
2129 else
2130 n = 16;
2131 /* dc scale factor table */
2132 for (i = 0; i < 64; i++)
2133 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2134
2135 if (s->theora >= 0x030200)
2136 matrices = get_bits(gb, 9) + 1;
2137 else
2138 matrices = 3;
2139
2140 if(matrices > 384){
2141 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2142 return -1;
2143 }
2144
2145 for(n=0; n<matrices; n++){
2146 for (i = 0; i < 64; i++)
2147 s->base_matrix[n][i]= get_bits(gb, 8);
2148 }
2149
2150 for (inter = 0; inter <= 1; inter++) {
2151 for (plane = 0; plane <= 2; plane++) {
2152 int newqr= 1;
2153 if (inter || plane > 0)
2154 newqr = get_bits1(gb);
2155 if (!newqr) {
2156 int qtj, plj;
2157 if(inter && get_bits1(gb)){
2158 qtj = 0;
2159 plj = plane;
2160 }else{
2161 qtj= (3*inter + plane - 1) / 3;
2162 plj= (plane + 2) % 3;
2163 }
2164 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2165 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2166 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2167 } else {
2168 int qri= 0;
2169 int qi = 0;
2170
2171 for(;;){
2172 i= get_bits(gb, av_log2(matrices-1)+1);
2173 if(i>= matrices){
2174 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2175 return -1;
2176 }
2177 s->qr_base[inter][plane][qri]= i;
2178 if(qi >= 63)
2179 break;
2180 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2181 s->qr_size[inter][plane][qri++]= i;
2182 qi += i;
2183 }
2184
2185 if (qi > 63) {
2186 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2187 return -1;
2188 }
2189 s->qr_count[inter][plane]= qri;
2190 }
2191 }
2192 }
2193
2194 /* Huffman tables */
2195 for (s->hti = 0; s->hti < 80; s->hti++) {
2196 s->entries = 0;
2197 s->huff_code_size = 1;
2198 if (!get_bits1(gb)) {
2199 s->hbits = 0;
2200 if(read_huffman_tree(avctx, gb))
2201 return -1;
2202 s->hbits = 1;
2203 if(read_huffman_tree(avctx, gb))
2204 return -1;
2205 }
2206 }
2207
2208 s->theora_tables = 1;
2209
2210 return 0;
2211 }
2212
2213 static av_cold int theora_decode_init(AVCodecContext *avctx)
2214 {
2215 Vp3DecodeContext *s = avctx->priv_data;
2216 GetBitContext gb;
2217 int ptype;
2218 uint8_t *header_start[3];
2219 int header_len[3];
2220 int i;
2221
2222 s->theora = 1;
2223
2224 if (!avctx->extradata_size)
2225 {
2226 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2227 return -1;
2228 }
2229
2230 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2231 42, header_start, header_len) < 0) {
2232 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2233 return -1;
2234 }
2235
2236 for(i=0;i<3;i++) {
2237 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2238
2239 ptype = get_bits(&gb, 8);
2240
2241 if (!(ptype & 0x80))
2242 {
2243 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2244 // return -1;
2245 }
2246
2247 // FIXME: Check for this as well.
2248 skip_bits_long(&gb, 6*8); /* "theora" */
2249
2250 switch(ptype)
2251 {
2252 case 0x80:
2253 theora_decode_header(avctx, &gb);
2254 break;
2255 case 0x81:
2256 // FIXME: is this needed? it breaks sometimes
2257 // theora_decode_comments(avctx, gb);
2258 break;
2259 case 0x82:
2260 if (theora_decode_tables(avctx, &gb))
2261 return -1;
2262 break;
2263 default:
2264 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2265 break;
2266 }
2267 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2268 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2269 if (s->theora < 0x030200)
2270 break;
2271 }
2272
2273 return vp3_decode_init(avctx);
2274 }
2275
2276 AVCodec theora_decoder = {
2277 "theora",
2278 CODEC_TYPE_VIDEO,
2279 CODEC_ID_THEORA,
2280 sizeof(Vp3DecodeContext),
2281 theora_decode_init,
2282 NULL,
2283 vp3_decode_end,
2284 vp3_decode_frame,
2285 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2286 NULL,
2287 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2288 };
2289 #endif
2290
2291 AVCodec vp3_decoder = {
2292 "vp3",
2293 CODEC_TYPE_VIDEO,
2294 CODEC_ID_VP3,
2295 sizeof(Vp3DecodeContext),
2296 vp3_decode_init,
2297 NULL,
2298 vp3_decode_end,
2299 vp3_decode_frame,
2300 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2301 NULL,
2302 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2303 };