Do MC and IDCT in coding (hilbert) order
[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 // This is the maximum length of a single long bit run that can be encoded
67 // for superblock coding or block qps. Theora special-cases this to read a
68 // bit instead of flipping the current bit to allow for runs longer than 4129.
69 #define MAXIMUM_LONG_BIT_RUN 4129
70
71 #define MODE_INTER_NO_MV 0
72 #define MODE_INTRA 1
73 #define MODE_INTER_PLUS_MV 2
74 #define MODE_INTER_LAST_MV 3
75 #define MODE_INTER_PRIOR_LAST 4
76 #define MODE_USING_GOLDEN 5
77 #define MODE_GOLDEN_MV 6
78 #define MODE_INTER_FOURMV 7
79 #define CODING_MODE_COUNT 8
80
81 /* special internal mode */
82 #define MODE_COPY 8
83
84 /* There are 6 preset schemes, plus a free-form scheme */
85 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
86 {
87 /* scheme 1: Last motion vector dominates */
88 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
89 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
90 MODE_INTRA, MODE_USING_GOLDEN,
91 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
92
93 /* scheme 2 */
94 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
95 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
96 MODE_INTRA, MODE_USING_GOLDEN,
97 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
98
99 /* scheme 3 */
100 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
101 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
102 MODE_INTRA, MODE_USING_GOLDEN,
103 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
104
105 /* scheme 4 */
106 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
107 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
108 MODE_INTRA, MODE_USING_GOLDEN,
109 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
110
111 /* scheme 5: No motion vector dominates */
112 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
113 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
114 MODE_INTRA, MODE_USING_GOLDEN,
115 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
116
117 /* scheme 6 */
118 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
119 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
120 MODE_INTER_PLUS_MV, MODE_INTRA,
121 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
122
123 };
124
125 static const uint8_t hilbert_offset[16][2] = {
126 {0,0}, {1,0}, {1,1}, {0,1},
127 {0,2}, {0,3}, {1,3}, {1,2},
128 {2,2}, {2,3}, {3,3}, {3,2},
129 {3,1}, {2,1}, {2,0}, {3,0}
130 };
131
132 #define MIN_DEQUANT_VAL 2
133
134 typedef struct Vp3DecodeContext {
135 AVCodecContext *avctx;
136 int theora, theora_tables;
137 int version;
138 int width, height;
139 AVFrame golden_frame;
140 AVFrame last_frame;
141 AVFrame current_frame;
142 int keyframe;
143 DSPContext dsp;
144 int flipped_image;
145 int last_slice_end;
146
147 int qps[3];
148 int nqps;
149 int last_qps[3];
150
151 int superblock_count;
152 int y_superblock_width;
153 int y_superblock_height;
154 int y_superblock_count;
155 int c_superblock_width;
156 int c_superblock_height;
157 int c_superblock_count;
158 int u_superblock_start;
159 int v_superblock_start;
160 unsigned char *superblock_coding;
161
162 int macroblock_count;
163 int macroblock_width;
164 int macroblock_height;
165
166 int fragment_count;
167 int fragment_width;
168 int fragment_height;
169
170 Vp3Fragment *all_fragments;
171 uint8_t *coeff_counts;
172 Coeff *coeffs;
173 Coeff *next_coeff;
174 int fragment_start[3];
175 int data_offset[3];
176
177 ScanTable scantable;
178
179 /* tables */
180 uint16_t coded_dc_scale_factor[64];
181 uint32_t coded_ac_scale_factor[64];
182 uint8_t base_matrix[384][64];
183 uint8_t qr_count[2][3];
184 uint8_t qr_size [2][3][64];
185 uint16_t qr_base[2][3][64];
186
187 /* this is a list of indexes into the all_fragments array indicating
188 * which of the fragments are coded */
189 int *coded_fragment_list;
190 int coded_fragment_list_index;
191
192 /* track which fragments have already been decoded; called 'fast'
193 * because this data structure avoids having to iterate through every
194 * fragment in coded_fragment_list; once a fragment has been fully
195 * decoded, it is removed from this list */
196 int *fast_fragment_list;
197 int fragment_list_y_head;
198 int fragment_list_c_head;
199
200 VLC dc_vlc[16];
201 VLC ac_vlc_1[16];
202 VLC ac_vlc_2[16];
203 VLC ac_vlc_3[16];
204 VLC ac_vlc_4[16];
205
206 VLC superblock_run_length_vlc;
207 VLC fragment_run_length_vlc;
208 VLC mode_code_vlc;
209 VLC motion_vector_vlc;
210
211 /* these arrays need to be on 16-byte boundaries since SSE2 operations
212 * index into them */
213 DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
214
215 /* This table contains superblock_count * 16 entries. Each set of 16
216 * numbers corresponds to the fragment indexes 0..15 of the superblock.
217 * An entry will be -1 to indicate that no entry corresponds to that
218 * index. */
219 int *superblock_fragments;
220
221 /* This is an array that indicates how a particular macroblock
222 * is coded. */
223 unsigned char *macroblock_coding;
224
225 int first_coded_y_fragment;
226 int first_coded_c_fragment;
227 int last_coded_y_fragment;
228 int last_coded_c_fragment;
229
230 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
231 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
232
233 /* Huffman decode */
234 int hti;
235 unsigned int hbits;
236 int entries;
237 int huff_code_size;
238 uint16_t huffman_table[80][32][2];
239
240 uint8_t filter_limit_values[64];
241 DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
242 } Vp3DecodeContext;
243
244 /************************************************************************
245 * VP3 specific functions
246 ************************************************************************/
247
248 /*
249 * This function sets up all of the various blocks mappings:
250 * superblocks <-> fragments, macroblocks <-> fragments,
251 * superblocks <-> macroblocks
252 *
253 * Returns 0 is successful; returns 1 if *anything* went wrong.
254 */
255 static int init_block_mapping(Vp3DecodeContext *s)
256 {
257 int i, j;
258 signed int hilbert_walk_mb[4];
259
260 int current_fragment = 0;
261 int current_width = 0;
262 int current_height = 0;
263 int right_edge = 0;
264 int bottom_edge = 0;
265 int superblock_row_inc = 0;
266 int mapping_index = 0;
267
268 static const signed char travel_width[16] = {
269 1, 1, 0, -1,
270 0, 0, 1, 0,
271 1, 0, 1, 0,
272 0, -1, 0, 1
273 };
274
275 static const signed char travel_height[16] = {
276 0, 0, 1, 0,
277 1, 1, 0, -1,
278 0, 1, 0, -1,
279 -1, 0, -1, 0
280 };
281
282 hilbert_walk_mb[0] = 1;
283 hilbert_walk_mb[1] = s->macroblock_width;
284 hilbert_walk_mb[2] = 1;
285 hilbert_walk_mb[3] = -s->macroblock_width;
286
287 /* iterate through each superblock (all planes) and map the fragments */
288 for (i = 0; i < s->superblock_count; i++) {
289 /* time to re-assign the limits? */
290 if (i == 0) {
291
292 /* start of Y superblocks */
293 right_edge = s->fragment_width;
294 bottom_edge = s->fragment_height;
295 current_width = -1;
296 current_height = 0;
297 superblock_row_inc = 3 * s->fragment_width -
298 (s->y_superblock_width * 4 - s->fragment_width);
299
300 /* the first operation for this variable is to advance by 1 */
301 current_fragment = -1;
302
303 } else if (i == s->u_superblock_start) {
304
305 /* start of U superblocks */
306 right_edge = s->fragment_width / 2;
307 bottom_edge = s->fragment_height / 2;
308 current_width = -1;
309 current_height = 0;
310 superblock_row_inc = 3 * (s->fragment_width / 2) -
311 (s->c_superblock_width * 4 - s->fragment_width / 2);
312
313 /* the first operation for this variable is to advance by 1 */
314 current_fragment = s->fragment_start[1] - 1;
315
316 } else if (i == s->v_superblock_start) {
317
318 /* start of V superblocks */
319 right_edge = s->fragment_width / 2;
320 bottom_edge = s->fragment_height / 2;
321 current_width = -1;
322 current_height = 0;
323 superblock_row_inc = 3 * (s->fragment_width / 2) -
324 (s->c_superblock_width * 4 - s->fragment_width / 2);
325
326 /* the first operation for this variable is to advance by 1 */
327 current_fragment = s->fragment_start[2] - 1;
328
329 }
330
331 if (current_width >= right_edge - 1) {
332 /* reset width and move to next superblock row */
333 current_width = -1;
334 current_height += 4;
335
336 /* fragment is now at the start of a new superblock row */
337 current_fragment += superblock_row_inc;
338 }
339
340 /* iterate through all 16 fragments in a superblock */
341 for (j = 0; j < 16; j++) {
342 current_fragment += travel_width[j] + right_edge * travel_height[j];
343 current_width += travel_width[j];
344 current_height += travel_height[j];
345
346 /* check if the fragment is in bounds */
347 if ((current_width < right_edge) &&
348 (current_height < bottom_edge)) {
349 s->superblock_fragments[mapping_index] = current_fragment;
350 } else {
351 s->superblock_fragments[mapping_index] = -1;
352 }
353
354 mapping_index++;
355 }
356 }
357
358 return 0; /* successful path out */
359 }
360
361 /*
362 * This function wipes out all of the fragment data.
363 */
364 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
365 {
366 int i;
367
368 /* zero out all of the fragment information */
369 s->coded_fragment_list_index = 0;
370 for (i = 0; i < s->fragment_count; i++) {
371 s->coeff_counts[i] = 0;
372 s->all_fragments[i].motion_x = 127;
373 s->all_fragments[i].motion_y = 127;
374 s->all_fragments[i].next_coeff= NULL;
375 s->all_fragments[i].qpi = 0;
376 s->coeffs[i].index=
377 s->coeffs[i].coeff=0;
378 s->coeffs[i].next= NULL;
379 }
380 }
381
382 /*
383 * This function sets up the dequantization tables used for a particular
384 * frame.
385 */
386 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
387 {
388 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
389 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
390 int i, plane, inter, qri, bmi, bmj, qistart;
391
392 for(inter=0; inter<2; inter++){
393 for(plane=0; plane<3; plane++){
394 int sum=0;
395 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
396 sum+= s->qr_size[inter][plane][qri];
397 if(s->qps[qpi] <= sum)
398 break;
399 }
400 qistart= sum - s->qr_size[inter][plane][qri];
401 bmi= s->qr_base[inter][plane][qri ];
402 bmj= s->qr_base[inter][plane][qri+1];
403 for(i=0; i<64; i++){
404 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
405 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
406 + s->qr_size[inter][plane][qri])
407 / (2*s->qr_size[inter][plane][qri]);
408
409 int qmin= 8<<(inter + !i);
410 int qscale= i ? ac_scale_factor : dc_scale_factor;
411
412 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
413 }
414 // all DC coefficients use the same quant so as not to interfere with DC prediction
415 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
416 }
417 }
418
419 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
420 }
421
422 /*
423 * This function initializes the loop filter boundary limits if the frame's
424 * quality index is different from the previous frame's.
425 *
426 * The filter_limit_values may not be larger than 127.
427 */
428 static void init_loop_filter(Vp3DecodeContext *s)
429 {
430 int *bounding_values= s->bounding_values_array+127;
431 int filter_limit;
432 int x;
433 int value;
434
435 filter_limit = s->filter_limit_values[s->qps[0]];
436
437 /* set up the bounding values */
438 memset(s->bounding_values_array, 0, 256 * sizeof(int));
439 for (x = 0; x < filter_limit; x++) {
440 bounding_values[-x] = -x;
441 bounding_values[x] = x;
442 }
443 for (x = value = filter_limit; x < 128 && value; x++, value--) {
444 bounding_values[ x] = value;
445 bounding_values[-x] = -value;
446 }
447 if (value)
448 bounding_values[128] = value;
449 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
450 }
451
452 /*
453 * This function unpacks all of the superblock/macroblock/fragment coding
454 * information from the bitstream.
455 */
456 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
457 {
458 int bit = 0;
459 int current_superblock = 0;
460 int current_run = 0;
461 int num_partial_superblocks = 0;
462 int first_c_fragment_seen;
463
464 int i, j;
465 int current_fragment;
466 int plane;
467
468 if (s->keyframe) {
469 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
470
471 } else {
472
473 /* unpack the list of partially-coded superblocks */
474 bit = get_bits1(gb);
475 while (current_superblock < s->superblock_count) {
476 current_run = get_vlc2(gb,
477 s->superblock_run_length_vlc.table, 6, 2) + 1;
478 if (current_run == 34)
479 current_run += get_bits(gb, 12);
480
481 if (current_superblock + current_run > s->superblock_count) {
482 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
483 return -1;
484 }
485
486 memset(s->superblock_coding + current_superblock, bit, current_run);
487
488 current_superblock += current_run;
489 if (bit)
490 num_partial_superblocks += current_run;
491
492 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
493 bit = get_bits1(gb);
494 else
495 bit ^= 1;
496 }
497
498 /* unpack the list of fully coded superblocks if any of the blocks were
499 * not marked as partially coded in the previous step */
500 if (num_partial_superblocks < s->superblock_count) {
501 int superblocks_decoded = 0;
502
503 current_superblock = 0;
504 bit = get_bits1(gb);
505 while (superblocks_decoded < s->superblock_count - num_partial_superblocks) {
506 current_run = get_vlc2(gb,
507 s->superblock_run_length_vlc.table, 6, 2) + 1;
508 if (current_run == 34)
509 current_run += get_bits(gb, 12);
510
511 for (j = 0; j < current_run; current_superblock++) {
512 if (current_superblock >= s->superblock_count) {
513 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
514 return -1;
515 }
516
517 /* skip any superblocks already marked as partially coded */
518 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
519 s->superblock_coding[current_superblock] = 2*bit;
520 j++;
521 }
522 }
523 superblocks_decoded += current_run;
524
525 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
526 bit = get_bits1(gb);
527 else
528 bit ^= 1;
529 }
530 }
531
532 /* if there were partial blocks, initialize bitstream for
533 * unpacking fragment codings */
534 if (num_partial_superblocks) {
535
536 current_run = 0;
537 bit = get_bits1(gb);
538 /* toggle the bit because as soon as the first run length is
539 * fetched the bit will be toggled again */
540 bit ^= 1;
541 }
542 }
543
544 /* figure out which fragments are coded; iterate through each
545 * superblock (all planes) */
546 s->coded_fragment_list_index = 0;
547 s->next_coeff= s->coeffs + s->fragment_count;
548 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
549 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
550 first_c_fragment_seen = 0;
551 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
552
553 for (plane = 0; plane < 3; plane++) {
554 int sb_start = (int[]){ 0, s->u_superblock_start, s->v_superblock_start }[plane];
555 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
556
557 for (i = sb_start; i < sb_end; i++) {
558
559 /* iterate through all 16 fragments in a superblock */
560 for (j = 0; j < 16; j++) {
561
562 /* if the fragment is in bounds, check its coding status */
563 current_fragment = s->superblock_fragments[i * 16 + j];
564 if (current_fragment >= s->fragment_count) {
565 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
566 current_fragment, s->fragment_count);
567 return 1;
568 }
569 if (current_fragment != -1) {
570 int coded = s->superblock_coding[i];
571
572 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
573
574 /* fragment may or may not be coded; this is the case
575 * that cares about the fragment coding runs */
576 if (current_run-- == 0) {
577 bit ^= 1;
578 current_run = get_vlc2(gb,
579 s->fragment_run_length_vlc.table, 5, 2);
580 }
581 coded = bit;
582 }
583
584 if (coded) {
585 /* default mode; actual mode will be decoded in
586 * the next phase */
587 s->all_fragments[current_fragment].coding_method =
588 MODE_INTER_NO_MV;
589 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
590 s->coded_fragment_list[s->coded_fragment_list_index] =
591 current_fragment;
592 if (plane && !first_c_fragment_seen) {
593 s->first_coded_c_fragment = s->coded_fragment_list_index;
594 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
595 first_c_fragment_seen = 1;
596 }
597 s->coded_fragment_list_index++;
598 } else {
599 /* not coded; copy this fragment from the prior frame */
600 s->all_fragments[current_fragment].coding_method =
601 MODE_COPY;
602 }
603 }
604 }
605 }
606 }
607
608 if (!first_c_fragment_seen)
609 /* only Y fragments coded in this frame */
610 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
611 else
612 /* end the list of coded C fragments */
613 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
614
615 for (i = 0; i < s->fragment_count - 1; i++) {
616 s->fast_fragment_list[i] = i + 1;
617 }
618 s->fast_fragment_list[s->fragment_count - 1] = -1;
619
620 if (s->last_coded_y_fragment == -1)
621 s->fragment_list_y_head = -1;
622 else {
623 s->fragment_list_y_head = s->first_coded_y_fragment;
624 s->fast_fragment_list[s->last_coded_y_fragment] = -1;
625 }
626
627 if (s->last_coded_c_fragment == -1)
628 s->fragment_list_c_head = -1;
629 else {
630 s->fragment_list_c_head = s->first_coded_c_fragment;
631 s->fast_fragment_list[s->last_coded_c_fragment] = -1;
632 }
633
634 return 0;
635 }
636
637 /*
638 * This function unpacks all the coding mode data for individual macroblocks
639 * from the bitstream.
640 */
641 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
642 {
643 int i, j, k, sb_x, sb_y;
644 int scheme;
645 int current_macroblock;
646 int current_fragment;
647 int coding_mode;
648 int custom_mode_alphabet[CODING_MODE_COUNT];
649 const int *alphabet;
650
651 if (s->keyframe) {
652 for (i = 0; i < s->fragment_count; i++)
653 s->all_fragments[i].coding_method = MODE_INTRA;
654
655 } else {
656
657 /* fetch the mode coding scheme for this frame */
658 scheme = get_bits(gb, 3);
659
660 /* is it a custom coding scheme? */
661 if (scheme == 0) {
662 for (i = 0; i < 8; i++)
663 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
664 for (i = 0; i < 8; i++)
665 custom_mode_alphabet[get_bits(gb, 3)] = i;
666 alphabet = custom_mode_alphabet;
667 } else
668 alphabet = ModeAlphabet[scheme-1];
669
670 /* iterate through all of the macroblocks that contain 1 or more
671 * coded fragments */
672 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
673 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
674
675 for (j = 0; j < 4; j++) {
676 int mb_x = 2*sb_x + (j>>1);
677 int mb_y = 2*sb_y + (((j>>1)+j)&1);
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
701 coding_mode = alphabet
702 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
703
704 s->macroblock_coding[current_macroblock] = coding_mode;
705 for (k = 0; k < 4; k++) {
706 current_fragment =
707 BLOCK_Y*s->fragment_width + BLOCK_X;
708 if (s->all_fragments[current_fragment].coding_method !=
709 MODE_COPY)
710 s->all_fragments[current_fragment].coding_method =
711 coding_mode;
712 }
713 for (k = 0; k < 2; k++) {
714 current_fragment = s->fragment_start[k+1] +
715 mb_y*(s->fragment_width>>1) + mb_x;
716 if (s->all_fragments[current_fragment].coding_method !=
717 MODE_COPY)
718 s->all_fragments[current_fragment].coding_method =
719 coding_mode;
720 }
721 }
722 }
723 }
724 }
725
726 return 0;
727 }
728
729 /*
730 * This function unpacks all the motion vectors for the individual
731 * macroblocks from the bitstream.
732 */
733 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
734 {
735 int j, k, sb_x, sb_y;
736 int coding_mode;
737 int motion_x[6];
738 int motion_y[6];
739 int last_motion_x = 0;
740 int last_motion_y = 0;
741 int prior_last_motion_x = 0;
742 int prior_last_motion_y = 0;
743 int current_macroblock;
744 int current_fragment;
745
746 if (s->keyframe)
747 return 0;
748
749 memset(motion_x, 0, 6 * sizeof(int));
750 memset(motion_y, 0, 6 * sizeof(int));
751
752 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
753 coding_mode = get_bits1(gb);
754
755 /* iterate through all of the macroblocks that contain 1 or more
756 * coded fragments */
757 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
758 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
759
760 for (j = 0; j < 4; j++) {
761 int mb_x = 2*sb_x + (j>>1);
762 int mb_y = 2*sb_y + (((j>>1)+j)&1);
763 current_macroblock = mb_y * s->macroblock_width + mb_x;
764
765 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
766 (s->macroblock_coding[current_macroblock] == MODE_COPY))
767 continue;
768
769 switch (s->macroblock_coding[current_macroblock]) {
770
771 case MODE_INTER_PLUS_MV:
772 case MODE_GOLDEN_MV:
773 /* all 6 fragments use the same motion vector */
774 if (coding_mode == 0) {
775 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
776 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
777 } else {
778 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
779 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
780 }
781
782 /* vector maintenance, only on MODE_INTER_PLUS_MV */
783 if (s->macroblock_coding[current_macroblock] ==
784 MODE_INTER_PLUS_MV) {
785 prior_last_motion_x = last_motion_x;
786 prior_last_motion_y = last_motion_y;
787 last_motion_x = motion_x[0];
788 last_motion_y = motion_y[0];
789 }
790 break;
791
792 case MODE_INTER_FOURMV:
793 /* vector maintenance */
794 prior_last_motion_x = last_motion_x;
795 prior_last_motion_y = last_motion_y;
796
797 /* fetch 4 vectors from the bitstream, one for each
798 * Y fragment, then average for the C fragment vectors */
799 motion_x[4] = motion_y[4] = 0;
800 for (k = 0; k < 4; k++) {
801 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
802 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
803 if (coding_mode == 0) {
804 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
805 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
806 } else {
807 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
808 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
809 }
810 last_motion_x = motion_x[k];
811 last_motion_y = motion_y[k];
812 } else {
813 motion_x[k] = 0;
814 motion_y[k] = 0;
815 }
816 motion_x[4] += motion_x[k];
817 motion_y[4] += motion_y[k];
818 }
819
820 motion_x[5]=
821 motion_x[4]= RSHIFT(motion_x[4], 2);
822 motion_y[5]=
823 motion_y[4]= RSHIFT(motion_y[4], 2);
824 break;
825
826 case MODE_INTER_LAST_MV:
827 /* all 6 fragments use the last motion vector */
828 motion_x[0] = last_motion_x;
829 motion_y[0] = last_motion_y;
830
831 /* no vector maintenance (last vector remains the
832 * last vector) */
833 break;
834
835 case MODE_INTER_PRIOR_LAST:
836 /* all 6 fragments use the motion vector prior to the
837 * last motion vector */
838 motion_x[0] = prior_last_motion_x;
839 motion_y[0] = prior_last_motion_y;
840
841 /* vector maintenance */
842 prior_last_motion_x = last_motion_x;
843 prior_last_motion_y = last_motion_y;
844 last_motion_x = motion_x[0];
845 last_motion_y = motion_y[0];
846 break;
847
848 default:
849 /* covers intra, inter without MV, golden without MV */
850 motion_x[0] = 0;
851 motion_y[0] = 0;
852
853 /* no vector maintenance */
854 break;
855 }
856
857 /* assign the motion vectors to the correct fragments */
858 for (k = 0; k < 4; k++) {
859 current_fragment =
860 BLOCK_Y*s->fragment_width + BLOCK_X;
861 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
862 s->all_fragments[current_fragment].motion_x = motion_x[k];
863 s->all_fragments[current_fragment].motion_y = motion_y[k];
864 } else {
865 s->all_fragments[current_fragment].motion_x = motion_x[0];
866 s->all_fragments[current_fragment].motion_y = motion_y[0];
867 }
868 }
869 for (k = 0; k < 2; k++) {
870 current_fragment = s->fragment_start[k+1] +
871 mb_y*(s->fragment_width>>1) + mb_x;
872 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
873 s->all_fragments[current_fragment].motion_x = motion_x[k+4];
874 s->all_fragments[current_fragment].motion_y = motion_y[k+4];
875 } else {
876 s->all_fragments[current_fragment].motion_x = motion_x[0];
877 s->all_fragments[current_fragment].motion_y = motion_y[0];
878 }
879 }
880 }
881 }
882 }
883
884 return 0;
885 }
886
887 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
888 {
889 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
890 int num_blocks = s->coded_fragment_list_index;
891
892 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
893 i = blocks_decoded = num_blocks_at_qpi = 0;
894
895 bit = get_bits1(gb);
896
897 do {
898 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
899 if (run_length == 34)
900 run_length += get_bits(gb, 12);
901 blocks_decoded += run_length;
902
903 if (!bit)
904 num_blocks_at_qpi += run_length;
905
906 for (j = 0; j < run_length; i++) {
907 if (i >= s->coded_fragment_list_index)
908 return -1;
909
910 if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
911 s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
912 j++;
913 }
914 }
915
916 if (run_length == MAXIMUM_LONG_BIT_RUN)
917 bit = get_bits1(gb);
918 else
919 bit ^= 1;
920 } while (blocks_decoded < num_blocks);
921
922 num_blocks -= num_blocks_at_qpi;
923 }
924
925 return 0;
926 }
927
928 /*
929 * This function is called by unpack_dct_coeffs() to extract the VLCs from
930 * the bitstream. The VLCs encode tokens which are used to unpack DCT
931 * data. This function unpacks all the VLCs for either the Y plane or both
932 * C planes, and is called for DC coefficients or different AC coefficient
933 * levels (since different coefficient types require different VLC tables.
934 *
935 * This function returns a residual eob run. E.g, if a particular token gave
936 * instructions to EOB the next 5 fragments and there were only 2 fragments
937 * left in the current fragment range, 3 would be returned so that it could
938 * be passed into the next call to this same function.
939 */
940 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
941 VLC *table, int coeff_index,
942 int y_plane,
943 int eob_run)
944 {
945 int i;
946 int token;
947 int zero_run = 0;
948 DCTELEM coeff = 0;
949 Vp3Fragment *fragment;
950 int bits_to_get;
951 int next_fragment;
952 int previous_fragment;
953 int fragment_num;
954 int *list_head;
955
956 /* local references to structure members to avoid repeated deferences */
957 uint8_t *perm= s->scantable.permutated;
958 int *coded_fragment_list = s->coded_fragment_list;
959 Vp3Fragment *all_fragments = s->all_fragments;
960 uint8_t *coeff_counts = s->coeff_counts;
961 VLC_TYPE (*vlc_table)[2] = table->table;
962 int *fast_fragment_list = s->fast_fragment_list;
963
964 if (y_plane) {
965 next_fragment = s->fragment_list_y_head;
966 list_head = &s->fragment_list_y_head;
967 } else {
968 next_fragment = s->fragment_list_c_head;
969 list_head = &s->fragment_list_c_head;
970 }
971
972 i = next_fragment;
973 previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */
974 while (i != -1) {
975 fragment_num = coded_fragment_list[i];
976
977 if (coeff_counts[fragment_num] > coeff_index) {
978 previous_fragment = i;
979 i = fast_fragment_list[i];
980 continue;
981 }
982 fragment = &all_fragments[fragment_num];
983
984 if (!eob_run) {
985 /* decode a VLC into a token */
986 token = get_vlc2(gb, vlc_table, 5, 3);
987 /* use the token to get a zero run, a coefficient, and an eob run */
988 if (token <= 6) {
989 eob_run = eob_run_base[token];
990 if (eob_run_get_bits[token])
991 eob_run += get_bits(gb, eob_run_get_bits[token]);
992 coeff = zero_run = 0;
993 } else {
994 bits_to_get = coeff_get_bits[token];
995 if (bits_to_get)
996 bits_to_get = get_bits(gb, bits_to_get);
997 coeff = coeff_tables[token][bits_to_get];
998
999 zero_run = zero_run_base[token];
1000 if (zero_run_get_bits[token])
1001 zero_run += get_bits(gb, zero_run_get_bits[token]);
1002 }
1003 }
1004
1005 if (!eob_run) {
1006 coeff_counts[fragment_num] += zero_run;
1007 if (coeff_counts[fragment_num] < 64){
1008 fragment->next_coeff->coeff= coeff;
1009 fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1010 fragment->next_coeff->next= s->next_coeff;
1011 s->next_coeff->next=NULL;
1012 fragment->next_coeff= s->next_coeff++;
1013 }
1014 /* previous fragment is now this fragment */
1015 previous_fragment = i;
1016 } else {
1017 coeff_counts[fragment_num] |= 128;
1018 eob_run--;
1019 /* remove this fragment from the list */
1020 if (previous_fragment != -1)
1021 fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1022 else
1023 *list_head = fast_fragment_list[i];
1024 /* previous fragment remains unchanged */
1025 }
1026
1027 i = fast_fragment_list[i];
1028 }
1029
1030 return eob_run;
1031 }
1032
1033 static void reverse_dc_prediction(Vp3DecodeContext *s,
1034 int first_fragment,
1035 int fragment_width,
1036 int fragment_height);
1037 /*
1038 * This function unpacks all of the DCT coefficient data from the
1039 * bitstream.
1040 */
1041 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1042 {
1043 int i;
1044 int dc_y_table;
1045 int dc_c_table;
1046 int ac_y_table;
1047 int ac_c_table;
1048 int residual_eob_run = 0;
1049 VLC *y_tables[64];
1050 VLC *c_tables[64];
1051
1052 /* fetch the DC table indexes */
1053 dc_y_table = get_bits(gb, 4);
1054 dc_c_table = get_bits(gb, 4);
1055
1056 /* unpack the Y plane DC coefficients */
1057 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1058 1, residual_eob_run);
1059
1060 /* reverse prediction of the Y-plane DC coefficients */
1061 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1062
1063 /* unpack the C plane DC coefficients */
1064 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1065 0, residual_eob_run);
1066
1067 /* reverse prediction of the C-plane DC coefficients */
1068 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1069 {
1070 reverse_dc_prediction(s, s->fragment_start[1],
1071 s->fragment_width / 2, s->fragment_height / 2);
1072 reverse_dc_prediction(s, s->fragment_start[2],
1073 s->fragment_width / 2, s->fragment_height / 2);
1074 }
1075
1076 /* fetch the AC table indexes */
1077 ac_y_table = get_bits(gb, 4);
1078 ac_c_table = get_bits(gb, 4);
1079
1080 /* build tables of AC VLC tables */
1081 for (i = 1; i <= 5; i++) {
1082 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1083 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1084 }
1085 for (i = 6; i <= 14; i++) {
1086 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1087 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1088 }
1089 for (i = 15; i <= 27; i++) {
1090 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1091 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1092 }
1093 for (i = 28; i <= 63; i++) {
1094 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1095 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1096 }
1097
1098 /* decode all AC coefficents */
1099 for (i = 1; i <= 63; i++) {
1100 if (s->fragment_list_y_head != -1)
1101 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1102 1, residual_eob_run);
1103
1104 if (s->fragment_list_c_head != -1)
1105 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1106 0, residual_eob_run);
1107 }
1108
1109 return 0;
1110 }
1111
1112 /*
1113 * This function reverses the DC prediction for each coded fragment in
1114 * the frame. Much of this function is adapted directly from the original
1115 * VP3 source code.
1116 */
1117 #define COMPATIBLE_FRAME(x) \
1118 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1119 #define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1120
1121 static void reverse_dc_prediction(Vp3DecodeContext *s,
1122 int first_fragment,
1123 int fragment_width,
1124 int fragment_height)
1125 {
1126
1127 #define PUL 8
1128 #define PU 4
1129 #define PUR 2
1130 #define PL 1
1131
1132 int x, y;
1133 int i = first_fragment;
1134
1135 int predicted_dc;
1136
1137 /* DC values for the left, up-left, up, and up-right fragments */
1138 int vl, vul, vu, vur;
1139
1140 /* indexes for the left, up-left, up, and up-right fragments */
1141 int l, ul, u, ur;
1142
1143 /*
1144 * The 6 fields mean:
1145 * 0: up-left multiplier
1146 * 1: up multiplier
1147 * 2: up-right multiplier
1148 * 3: left multiplier
1149 */
1150 static const int predictor_transform[16][4] = {
1151 { 0, 0, 0, 0},
1152 { 0, 0, 0,128}, // PL
1153 { 0, 0,128, 0}, // PUR
1154 { 0, 0, 53, 75}, // PUR|PL
1155 { 0,128, 0, 0}, // PU
1156 { 0, 64, 0, 64}, // PU|PL
1157 { 0,128, 0, 0}, // PU|PUR
1158 { 0, 0, 53, 75}, // PU|PUR|PL
1159 {128, 0, 0, 0}, // PUL
1160 { 0, 0, 0,128}, // PUL|PL
1161 { 64, 0, 64, 0}, // PUL|PUR
1162 { 0, 0, 53, 75}, // PUL|PUR|PL
1163 { 0,128, 0, 0}, // PUL|PU
1164 {-104,116, 0,116}, // PUL|PU|PL
1165 { 24, 80, 24, 0}, // PUL|PU|PUR
1166 {-104,116, 0,116} // PUL|PU|PUR|PL
1167 };
1168
1169 /* This table shows which types of blocks can use other blocks for
1170 * prediction. For example, INTRA is the only mode in this table to
1171 * have a frame number of 0. That means INTRA blocks can only predict
1172 * from other INTRA blocks. There are 2 golden frame coding types;
1173 * blocks encoding in these modes can only predict from other blocks
1174 * that were encoded with these 1 of these 2 modes. */
1175 static const unsigned char compatible_frame[9] = {
1176 1, /* MODE_INTER_NO_MV */
1177 0, /* MODE_INTRA */
1178 1, /* MODE_INTER_PLUS_MV */
1179 1, /* MODE_INTER_LAST_MV */
1180 1, /* MODE_INTER_PRIOR_MV */
1181 2, /* MODE_USING_GOLDEN */
1182 2, /* MODE_GOLDEN_MV */
1183 1, /* MODE_INTER_FOUR_MV */
1184 3 /* MODE_COPY */
1185 };
1186 int current_frame_type;
1187
1188 /* there is a last DC predictor for each of the 3 frame types */
1189 short last_dc[3];
1190
1191 int transform = 0;
1192
1193 vul = vu = vur = vl = 0;
1194 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1195
1196 /* for each fragment row... */
1197 for (y = 0; y < fragment_height; y++) {
1198
1199 /* for each fragment in a row... */
1200 for (x = 0; x < fragment_width; x++, i++) {
1201
1202 /* reverse prediction if this block was coded */
1203 if (s->all_fragments[i].coding_method != MODE_COPY) {
1204
1205 current_frame_type =
1206 compatible_frame[s->all_fragments[i].coding_method];
1207
1208 transform= 0;
1209 if(x){
1210 l= i-1;
1211 vl = DC_COEFF(l);
1212 if(COMPATIBLE_FRAME(l))
1213 transform |= PL;
1214 }
1215 if(y){
1216 u= i-fragment_width;
1217 vu = DC_COEFF(u);
1218 if(COMPATIBLE_FRAME(u))
1219 transform |= PU;
1220 if(x){
1221 ul= i-fragment_width-1;
1222 vul = DC_COEFF(ul);
1223 if(COMPATIBLE_FRAME(ul))
1224 transform |= PUL;
1225 }
1226 if(x + 1 < fragment_width){
1227 ur= i-fragment_width+1;
1228 vur = DC_COEFF(ur);
1229 if(COMPATIBLE_FRAME(ur))
1230 transform |= PUR;
1231 }
1232 }
1233
1234 if (transform == 0) {
1235
1236 /* if there were no fragments to predict from, use last
1237 * DC saved */
1238 predicted_dc = last_dc[current_frame_type];
1239 } else {
1240
1241 /* apply the appropriate predictor transform */
1242 predicted_dc =
1243 (predictor_transform[transform][0] * vul) +
1244 (predictor_transform[transform][1] * vu) +
1245 (predictor_transform[transform][2] * vur) +
1246 (predictor_transform[transform][3] * vl);
1247
1248 predicted_dc /= 128;
1249
1250 /* check for outranging on the [ul u l] and
1251 * [ul u ur l] predictors */
1252 if ((transform == 15) || (transform == 13)) {
1253 if (FFABS(predicted_dc - vu) > 128)
1254 predicted_dc = vu;
1255 else if (FFABS(predicted_dc - vl) > 128)
1256 predicted_dc = vl;
1257 else if (FFABS(predicted_dc - vul) > 128)
1258 predicted_dc = vul;
1259 }
1260 }
1261
1262 /* at long last, apply the predictor */
1263 if(s->coeffs[i].index){
1264 *s->next_coeff= s->coeffs[i];
1265 s->coeffs[i].index=0;
1266 s->coeffs[i].coeff=0;
1267 s->coeffs[i].next= s->next_coeff++;
1268 }
1269 s->coeffs[i].coeff += predicted_dc;
1270 /* save the DC */
1271 last_dc[current_frame_type] = DC_COEFF(i);
1272 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1273 s->coeff_counts[i]= 129;
1274 // s->all_fragments[i].next_coeff= s->next_coeff;
1275 s->coeffs[i].next= s->next_coeff;
1276 (s->next_coeff++)->next=NULL;
1277 }
1278 }
1279 }
1280 }
1281 }
1282
1283 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1284 {
1285 int x, y;
1286 int *bounding_values= s->bounding_values_array+127;
1287
1288 int width = s->fragment_width >> !!plane;
1289 int height = s->fragment_height >> !!plane;
1290 int fragment = s->fragment_start [plane] + ystart * width;
1291 int stride = s->current_frame.linesize[plane];
1292 uint8_t *plane_data = s->current_frame.data [plane];
1293 if (!s->flipped_image) stride = -stride;
1294 plane_data += s->data_offset[plane] + 8*ystart*stride;
1295
1296 for (y = ystart; y < yend; y++) {
1297
1298 for (x = 0; x < width; x++) {
1299 /* This code basically just deblocks on the edges of coded blocks.
1300 * However, it has to be much more complicated because of the
1301 * braindamaged deblock ordering used in VP3/Theora. Order matters
1302 * because some pixels get filtered twice. */
1303 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1304 {
1305 /* do not perform left edge filter for left columns frags */
1306 if (x > 0) {
1307 s->dsp.vp3_h_loop_filter(
1308 plane_data + 8*x,
1309 stride, bounding_values);
1310 }
1311
1312 /* do not perform top edge filter for top row fragments */
1313 if (y > 0) {
1314 s->dsp.vp3_v_loop_filter(
1315 plane_data + 8*x,
1316 stride, bounding_values);
1317 }
1318
1319 /* do not perform right edge filter for right column
1320 * fragments or if right fragment neighbor is also coded
1321 * in this frame (it will be filtered in next iteration) */
1322 if ((x < width - 1) &&
1323 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1324 s->dsp.vp3_h_loop_filter(
1325 plane_data + 8*x + 8,
1326 stride, bounding_values);
1327 }
1328
1329 /* do not perform bottom edge filter for bottom row
1330 * fragments or if bottom fragment neighbor is also coded
1331 * in this frame (it will be filtered in the next row) */
1332 if ((y < height - 1) &&
1333 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1334 s->dsp.vp3_v_loop_filter(
1335 plane_data + 8*x + 8*stride,
1336 stride, bounding_values);
1337 }
1338 }
1339
1340 fragment++;
1341 }
1342 plane_data += 8*stride;
1343 }
1344 }
1345
1346 /**
1347 * called when all pixels up to row y are complete
1348 */
1349 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1350 {
1351 int h, cy;
1352 int offset[4];
1353
1354 if(s->avctx->draw_horiz_band==NULL)
1355 return;
1356
1357 h= y - s->last_slice_end;
1358 y -= h;
1359
1360 if (!s->flipped_image) {
1361 if (y == 0)
1362 h -= s->height - s->avctx->height; // account for non-mod16
1363 y = s->height - y - h;
1364 }
1365
1366 cy = y >> 1;
1367 offset[0] = s->current_frame.linesize[0]*y;
1368 offset[1] = s->current_frame.linesize[1]*cy;
1369 offset[2] = s->current_frame.linesize[2]*cy;
1370 offset[3] = 0;
1371
1372 emms_c();
1373 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1374 s->last_slice_end= y + h;
1375 }
1376
1377 /*
1378 * Perform the final rendering for a particular slice of data.
1379 * The slice number ranges from 0..(c_superblock_height - 1).
1380 */
1381 static void render_slice(Vp3DecodeContext *s, int slice)
1382 {
1383 int x, y, i, j;
1384 int16_t *dequantizer;
1385 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1386 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1387 int motion_halfpel_index;
1388 uint8_t *motion_source;
1389 int plane, first_pixel;
1390
1391 if (slice >= s->c_superblock_height)
1392 return;
1393
1394 for (plane = 0; plane < 3; plane++) {
1395 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1396 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1397 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1398 int stride = s->current_frame.linesize[plane];
1399 int plane_width = s->width >> !!plane;
1400 int plane_height = s->height >> !!plane;
1401
1402 int sb_x, sb_y = slice << !plane;
1403 int slice_height = sb_y + (plane ? 1 : 2);
1404 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1405
1406 int fragment_width = s->fragment_width >> !!plane;
1407 int fragment_height = s->fragment_height >> !!plane;
1408 int fragment_start = s->fragment_start[plane];
1409
1410 if (!s->flipped_image) stride = -stride;
1411 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1412 continue;
1413
1414
1415 if(FFABS(stride) > 2048)
1416 return; //various tables are fixed size
1417
1418 /* for each superblock row in the slice (both of them)... */
1419 for (; sb_y < slice_height; sb_y++) {
1420
1421 /* for each superblock in a row... */
1422 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1423
1424 /* for each block in a superblock... */
1425 for (j = 0; j < 16; j++) {
1426 x = 4*sb_x + hilbert_offset[j][0];
1427 y = 4*sb_y + hilbert_offset[j][1];
1428
1429 i = fragment_start + y*fragment_width + x;
1430
1431 // bounds check
1432 if (x >= fragment_width || y >= fragment_height)
1433 continue;
1434
1435 first_pixel = 8*y*stride + 8*x;
1436
1437 /* transform if this block was coded */
1438 if (s->all_fragments[i].coding_method != MODE_COPY) {
1439
1440 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1441 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1442 motion_source= golden_plane;
1443 else
1444 motion_source= last_plane;
1445
1446 motion_source += first_pixel;
1447 motion_halfpel_index = 0;
1448
1449 /* sort out the motion vector if this fragment is coded
1450 * using a motion vector method */
1451 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1452 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1453 int src_x, src_y;
1454 motion_x = s->all_fragments[i].motion_x;
1455 motion_y = s->all_fragments[i].motion_y;
1456 if(plane){
1457 motion_x= (motion_x>>1) | (motion_x&1);
1458 motion_y= (motion_y>>1) | (motion_y&1);
1459 }
1460
1461 src_x= (motion_x>>1) + 8*x;
1462 src_y= (motion_y>>1) + 8*y;
1463 if ((motion_x == 127) || (motion_y == 127))
1464 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1465
1466 motion_halfpel_index = motion_x & 0x01;
1467 motion_source += (motion_x >> 1);
1468
1469 motion_halfpel_index |= (motion_y & 0x01) << 1;
1470 motion_source += ((motion_y >> 1) * stride);
1471
1472 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1473 uint8_t *temp= s->edge_emu_buffer;
1474 if(stride<0) temp -= 9*stride;
1475 else temp += 9*stride;
1476
1477 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1478 motion_source= temp;
1479 }
1480 }
1481
1482
1483 /* first, take care of copying a block from either the
1484 * previous or the golden frame */
1485 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1486 /* Note, it is possible to implement all MC cases with
1487 put_no_rnd_pixels_l2 which would look more like the
1488 VP3 source but this would be slower as
1489 put_no_rnd_pixels_tab is better optimzed */
1490 if(motion_halfpel_index != 3){
1491 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1492 output_plane + first_pixel,
1493 motion_source, stride, 8);
1494 }else{
1495 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1496 s->dsp.put_no_rnd_pixels_l2[1](
1497 output_plane + first_pixel,
1498 motion_source - d,
1499 motion_source + stride + 1 + d,
1500 stride, 8);
1501 }
1502 dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1503 }else{
1504 dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1505 }
1506
1507 /* dequantize the DCT coefficients */
1508 if(s->avctx->idct_algo==FF_IDCT_VP3){
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];
1513 coeff= coeff->next;
1514 }
1515 }else{
1516 Coeff *coeff= s->coeffs + i;
1517 s->dsp.clear_block(block);
1518 while(coeff->next){
1519 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1520 coeff= coeff->next;
1521 }
1522 }
1523
1524 /* invert DCT and place (or add) in final output */
1525
1526 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1527 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1528 block[0] += 128<<3;
1529 s->dsp.idct_put(
1530 output_plane + first_pixel,
1531 stride,
1532 block);
1533 } else {
1534 s->dsp.idct_add(
1535 output_plane + first_pixel,
1536 stride,
1537 block);
1538 }
1539 } else {
1540
1541 /* copy directly from the previous frame */
1542 s->dsp.put_pixels_tab[1][0](
1543 output_plane + first_pixel,
1544 last_plane + first_pixel,
1545 stride, 8);
1546
1547 }
1548 }
1549 }
1550
1551 // Filter up to the last row in the superblock row
1552 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1553 }
1554 }
1555
1556 /* this looks like a good place for slice dispatch... */
1557 /* algorithm:
1558 * if (slice == s->macroblock_height - 1)
1559 * dispatch (both last slice & 2nd-to-last slice);
1560 * else if (slice > 0)
1561 * dispatch (slice - 1);
1562 */
1563
1564 vp3_draw_horiz_band(s, 64*slice + 64-16);
1565 }
1566
1567 /*
1568 * This is the ffmpeg/libavcodec API init function.
1569 */
1570 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1571 {
1572 Vp3DecodeContext *s = avctx->priv_data;
1573 int i, inter, plane;
1574 int c_width;
1575 int c_height;
1576
1577 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1578 s->version = 0;
1579 else
1580 s->version = 1;
1581
1582 s->avctx = avctx;
1583 s->width = FFALIGN(avctx->width, 16);
1584 s->height = FFALIGN(avctx->height, 16);
1585 avctx->pix_fmt = PIX_FMT_YUV420P;
1586 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1587 if(avctx->idct_algo==FF_IDCT_AUTO)
1588 avctx->idct_algo=FF_IDCT_VP3;
1589 dsputil_init(&s->dsp, avctx);
1590
1591 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1592
1593 /* initialize to an impossible value which will force a recalculation
1594 * in the first frame decode */
1595 for (i = 0; i < 3; i++)
1596 s->qps[i] = -1;
1597
1598 s->y_superblock_width = (s->width + 31) / 32;
1599 s->y_superblock_height = (s->height + 31) / 32;
1600 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1601
1602 /* work out the dimensions for the C planes */
1603 c_width = s->width / 2;
1604 c_height = s->height / 2;
1605 s->c_superblock_width = (c_width + 31) / 32;
1606 s->c_superblock_height = (c_height + 31) / 32;
1607 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1608
1609 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1610 s->u_superblock_start = s->y_superblock_count;
1611 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1612 s->superblock_coding = av_malloc(s->superblock_count);
1613
1614 s->macroblock_width = (s->width + 15) / 16;
1615 s->macroblock_height = (s->height + 15) / 16;
1616 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1617
1618 s->fragment_width = s->width / FRAGMENT_PIXELS;
1619 s->fragment_height = s->height / FRAGMENT_PIXELS;
1620
1621 /* fragment count covers all 8x8 blocks for all 3 planes */
1622 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1623 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1624 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1625
1626 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1627 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1628 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1629 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1630 s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1631 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1632 !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1633 vp3_decode_end(avctx);
1634 return -1;
1635 }
1636
1637 if (!s->theora_tables)
1638 {
1639 for (i = 0; i < 64; i++) {
1640 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1641 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1642 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1643 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1644 s->base_matrix[2][i] = vp31_inter_dequant[i];
1645 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1646 }
1647
1648 for(inter=0; inter<2; inter++){
1649 for(plane=0; plane<3; plane++){
1650 s->qr_count[inter][plane]= 1;
1651 s->qr_size [inter][plane][0]= 63;
1652 s->qr_base [inter][plane][0]=
1653 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1654 }
1655 }
1656
1657 /* init VLC tables */
1658 for (i = 0; i < 16; i++) {
1659
1660 /* DC histograms */
1661 init_vlc(&s->dc_vlc[i], 5, 32,
1662 &dc_bias[i][0][1], 4, 2,
1663 &dc_bias[i][0][0], 4, 2, 0);
1664
1665 /* group 1 AC histograms */
1666 init_vlc(&s->ac_vlc_1[i], 5, 32,
1667 &ac_bias_0[i][0][1], 4, 2,
1668 &ac_bias_0[i][0][0], 4, 2, 0);
1669
1670 /* group 2 AC histograms */
1671 init_vlc(&s->ac_vlc_2[i], 5, 32,
1672 &ac_bias_1[i][0][1], 4, 2,
1673 &ac_bias_1[i][0][0], 4, 2, 0);
1674
1675 /* group 3 AC histograms */
1676 init_vlc(&s->ac_vlc_3[i], 5, 32,
1677 &ac_bias_2[i][0][1], 4, 2,
1678 &ac_bias_2[i][0][0], 4, 2, 0);
1679
1680 /* group 4 AC histograms */
1681 init_vlc(&s->ac_vlc_4[i], 5, 32,
1682 &ac_bias_3[i][0][1], 4, 2,
1683 &ac_bias_3[i][0][0], 4, 2, 0);
1684 }
1685 } else {
1686 for (i = 0; i < 16; i++) {
1687
1688 /* DC histograms */
1689 if (init_vlc(&s->dc_vlc[i], 5, 32,
1690 &s->huffman_table[i][0][1], 4, 2,
1691 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1692 goto vlc_fail;
1693
1694 /* group 1 AC histograms */
1695 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1696 &s->huffman_table[i+16][0][1], 4, 2,
1697 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1698 goto vlc_fail;
1699
1700 /* group 2 AC histograms */
1701 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1702 &s->huffman_table[i+16*2][0][1], 4, 2,
1703 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1704 goto vlc_fail;
1705
1706 /* group 3 AC histograms */
1707 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1708 &s->huffman_table[i+16*3][0][1], 4, 2,
1709 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1710 goto vlc_fail;
1711
1712 /* group 4 AC histograms */
1713 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1714 &s->huffman_table[i+16*4][0][1], 4, 2,
1715 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1716 goto vlc_fail;
1717 }
1718 }
1719
1720 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1721 &superblock_run_length_vlc_table[0][1], 4, 2,
1722 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1723
1724 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1725 &fragment_run_length_vlc_table[0][1], 4, 2,
1726 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1727
1728 init_vlc(&s->mode_code_vlc, 3, 8,
1729 &mode_code_vlc_table[0][1], 2, 1,
1730 &mode_code_vlc_table[0][0], 2, 1, 0);
1731
1732 init_vlc(&s->motion_vector_vlc, 6, 63,
1733 &motion_vector_vlc_table[0][1], 2, 1,
1734 &motion_vector_vlc_table[0][0], 2, 1, 0);
1735
1736 /* work out the block mapping tables */
1737 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1738 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1739 if (!s->superblock_fragments || !s->macroblock_coding) {
1740 vp3_decode_end(avctx);
1741 return -1;
1742 }
1743 init_block_mapping(s);
1744
1745 for (i = 0; i < 3; i++) {
1746 s->current_frame.data[i] = NULL;
1747 s->last_frame.data[i] = NULL;
1748 s->golden_frame.data[i] = NULL;
1749 }
1750
1751 return 0;
1752
1753 vlc_fail:
1754 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1755 return -1;
1756 }
1757
1758 /*
1759 * This is the ffmpeg/libavcodec API frame decode function.
1760 */
1761 static int vp3_decode_frame(AVCodecContext *avctx,
1762 void *data, int *data_size,
1763 AVPacket *avpkt)
1764 {
1765 const uint8_t *buf = avpkt->data;
1766 int buf_size = avpkt->size;
1767 Vp3DecodeContext *s = avctx->priv_data;
1768 GetBitContext gb;
1769 static int counter = 0;
1770 int i;
1771
1772 init_get_bits(&gb, buf, buf_size * 8);
1773
1774 if (s->theora && get_bits1(&gb))
1775 {
1776 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1777 return -1;
1778 }
1779
1780 s->keyframe = !get_bits1(&gb);
1781 if (!s->theora)
1782 skip_bits(&gb, 1);
1783 for (i = 0; i < 3; i++)
1784 s->last_qps[i] = s->qps[i];
1785
1786 s->nqps=0;
1787 do{
1788 s->qps[s->nqps++]= get_bits(&gb, 6);
1789 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1790 for (i = s->nqps; i < 3; i++)
1791 s->qps[i] = -1;
1792
1793 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1794 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1795 s->keyframe?"key":"", counter, s->qps[0]);
1796 counter++;
1797
1798 if (s->qps[0] != s->last_qps[0])
1799 init_loop_filter(s);
1800
1801 for (i = 0; i < s->nqps; i++)
1802 // reinit all dequantizers if the first one changed, because
1803 // the DC of the first quantizer must be used for all matrices
1804 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1805 init_dequantizer(s, i);
1806
1807 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1808 return buf_size;
1809
1810 if (s->keyframe) {
1811 if (!s->theora)
1812 {
1813 skip_bits(&gb, 4); /* width code */
1814 skip_bits(&gb, 4); /* height code */
1815 if (s->version)
1816 {
1817 s->version = get_bits(&gb, 5);
1818 if (counter == 1)
1819 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1820 }
1821 }
1822 if (s->version || s->theora)
1823 {
1824 if (get_bits1(&gb))
1825 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1826 skip_bits(&gb, 2); /* reserved? */
1827 }
1828
1829 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1830 if (s->golden_frame.data[0])
1831 avctx->release_buffer(avctx, &s->golden_frame);
1832 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1833 } else {
1834 if (s->golden_frame.data[0])
1835 avctx->release_buffer(avctx, &s->golden_frame);
1836 if (s->last_frame.data[0])
1837 avctx->release_buffer(avctx, &s->last_frame);
1838 }
1839
1840 s->golden_frame.reference = 3;
1841 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1842 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1843 return -1;
1844 }
1845
1846 /* golden frame is also the current frame */
1847 s->current_frame= s->golden_frame;
1848 } else {
1849 /* allocate a new current frame */
1850 s->current_frame.reference = 3;
1851 if (!s->golden_frame.data[0]) {
1852 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1853 return -1;
1854 }
1855 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1856 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1857 return -1;
1858 }
1859 }
1860
1861 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1862 s->current_frame.qstride= 0;
1863
1864 init_frame(s, &gb);
1865
1866 if (unpack_superblocks(s, &gb)){
1867 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1868 return -1;
1869 }
1870 if (unpack_modes(s, &gb)){
1871 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1872 return -1;
1873 }
1874 if (unpack_vectors(s, &gb)){
1875 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1876 return -1;
1877 }
1878 if (unpack_block_qpis(s, &gb)){
1879 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1880 return -1;
1881 }
1882 if (unpack_dct_coeffs(s, &gb)){
1883 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1884 return -1;
1885 }
1886
1887 for (i = 0; i < 3; i++) {
1888 if (s->flipped_image)
1889 s->data_offset[i] = 0;
1890 else
1891 s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1892 }
1893
1894 s->last_slice_end = 0;
1895 for (i = 0; i < s->c_superblock_height; i++)
1896 render_slice(s, i);
1897
1898 // filter the last row
1899 for (i = 0; i < 3; i++) {
1900 int row = (s->height >> (3+!!i)) - 1;
1901 apply_loop_filter(s, i, row, row+1);
1902 }
1903 vp3_draw_horiz_band(s, s->height);
1904
1905 *data_size=sizeof(AVFrame);
1906 *(AVFrame*)data= s->current_frame;
1907
1908 /* release the last frame, if it is allocated and if it is not the
1909 * golden frame */
1910 if ((s->last_frame.data[0]) &&
1911 (s->last_frame.data[0] != s->golden_frame.data[0]))
1912 avctx->release_buffer(avctx, &s->last_frame);
1913
1914 /* shuffle frames (last = current) */
1915 s->last_frame= s->current_frame;
1916 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1917
1918 return buf_size;
1919 }
1920
1921 /*
1922 * This is the ffmpeg/libavcodec API module cleanup function.
1923 */
1924 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1925 {
1926 Vp3DecodeContext *s = avctx->priv_data;
1927 int i;
1928
1929 av_free(s->superblock_coding);
1930 av_free(s->all_fragments);
1931 av_free(s->coeff_counts);
1932 av_free(s->coeffs);
1933 av_free(s->coded_fragment_list);
1934 av_free(s->fast_fragment_list);
1935 av_free(s->superblock_fragments);
1936 av_free(s->macroblock_coding);
1937
1938 for (i = 0; i < 16; i++) {
1939 free_vlc(&s->dc_vlc[i]);
1940 free_vlc(&s->ac_vlc_1[i]);
1941 free_vlc(&s->ac_vlc_2[i]);
1942 free_vlc(&s->ac_vlc_3[i]);
1943 free_vlc(&s->ac_vlc_4[i]);
1944 }
1945
1946 free_vlc(&s->superblock_run_length_vlc);
1947 free_vlc(&s->fragment_run_length_vlc);
1948 free_vlc(&s->mode_code_vlc);
1949 free_vlc(&s->motion_vector_vlc);
1950
1951 /* release all frames */
1952 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1953 avctx->release_buffer(avctx, &s->golden_frame);
1954 if (s->last_frame.data[0])
1955 avctx->release_buffer(avctx, &s->last_frame);
1956 /* no need to release the current_frame since it will always be pointing
1957 * to the same frame as either the golden or last frame */
1958
1959 return 0;
1960 }
1961
1962 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1963 {
1964 Vp3DecodeContext *s = avctx->priv_data;
1965
1966 if (get_bits1(gb)) {
1967 int token;
1968 if (s->entries >= 32) { /* overflow */
1969 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1970 return -1;
1971 }
1972 token = get_bits(gb, 5);
1973 //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);
1974 s->huffman_table[s->hti][token][0] = s->hbits;
1975 s->huffman_table[s->hti][token][1] = s->huff_code_size;
1976 s->entries++;
1977 }
1978 else {
1979 if (s->huff_code_size >= 32) {/* overflow */
1980 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1981 return -1;
1982 }
1983 s->huff_code_size++;
1984 s->hbits <<= 1;
1985 if (read_huffman_tree(avctx, gb))
1986 return -1;
1987 s->hbits |= 1;
1988 if (read_huffman_tree(avctx, gb))
1989 return -1;
1990 s->hbits >>= 1;
1991 s->huff_code_size--;
1992 }
1993 return 0;
1994 }
1995
1996 #if CONFIG_THEORA_DECODER
1997 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1998 {
1999 Vp3DecodeContext *s = avctx->priv_data;
2000 int visible_width, visible_height, colorspace;
2001
2002 s->theora = get_bits_long(gb, 24);
2003 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2004
2005 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2006 /* but previous versions have the image flipped relative to vp3 */
2007 if (s->theora < 0x030200)
2008 {
2009 s->flipped_image = 1;
2010 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2011 }
2012
2013 visible_width = s->width = get_bits(gb, 16) << 4;
2014 visible_height = s->height = get_bits(gb, 16) << 4;
2015
2016 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2017 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2018 s->width= s->height= 0;
2019 return -1;
2020 }
2021
2022 if (s->theora >= 0x030200) {
2023 visible_width = get_bits_long(gb, 24);
2024 visible_height = get_bits_long(gb, 24);
2025
2026 skip_bits(gb, 8); /* offset x */
2027 skip_bits(gb, 8); /* offset y */
2028 }
2029
2030 skip_bits(gb, 32); /* fps numerator */
2031 skip_bits(gb, 32); /* fps denumerator */
2032 skip_bits(gb, 24); /* aspect numerator */
2033 skip_bits(gb, 24); /* aspect denumerator */
2034
2035 if (s->theora < 0x030200)
2036 skip_bits(gb, 5); /* keyframe frequency force */
2037 colorspace = get_bits(gb, 8);
2038 skip_bits(gb, 24); /* bitrate */
2039
2040 skip_bits(gb, 6); /* quality hint */
2041
2042 if (s->theora >= 0x030200)
2043 {
2044 skip_bits(gb, 5); /* keyframe frequency force */
2045 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2046 skip_bits(gb, 3); /* reserved */
2047 }
2048
2049 // align_get_bits(gb);
2050
2051 if ( visible_width <= s->width && visible_width > s->width-16
2052 && visible_height <= s->height && visible_height > s->height-16)
2053 avcodec_set_dimensions(avctx, visible_width, visible_height);
2054 else
2055 avcodec_set_dimensions(avctx, s->width, s->height);
2056
2057 if (colorspace == 1) {
2058 avctx->color_primaries = AVCOL_PRI_BT470M;
2059 } else if (colorspace == 2) {
2060 avctx->color_primaries = AVCOL_PRI_BT470BG;
2061 }
2062 if (colorspace == 1 || colorspace == 2) {
2063 avctx->colorspace = AVCOL_SPC_BT470BG;
2064 avctx->color_trc = AVCOL_TRC_BT709;
2065 }
2066
2067 return 0;
2068 }
2069
2070 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2071 {
2072 Vp3DecodeContext *s = avctx->priv_data;
2073 int i, n, matrices, inter, plane;
2074
2075 if (s->theora >= 0x030200) {
2076 n = get_bits(gb, 3);
2077 /* loop filter limit values table */
2078 for (i = 0; i < 64; i++) {
2079 s->filter_limit_values[i] = get_bits(gb, n);
2080 if (s->filter_limit_values[i] > 127) {
2081 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2082 s->filter_limit_values[i] = 127;
2083 }
2084 }
2085 }
2086
2087 if (s->theora >= 0x030200)
2088 n = get_bits(gb, 4) + 1;
2089 else
2090 n = 16;
2091 /* quality threshold table */
2092 for (i = 0; i < 64; i++)
2093 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2094
2095 if (s->theora >= 0x030200)
2096 n = get_bits(gb, 4) + 1;
2097 else
2098 n = 16;
2099 /* dc scale factor table */
2100 for (i = 0; i < 64; i++)
2101 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2102
2103 if (s->theora >= 0x030200)
2104 matrices = get_bits(gb, 9) + 1;
2105 else
2106 matrices = 3;
2107
2108 if(matrices > 384){
2109 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2110 return -1;
2111 }
2112
2113 for(n=0; n<matrices; n++){
2114 for (i = 0; i < 64; i++)
2115 s->base_matrix[n][i]= get_bits(gb, 8);
2116 }
2117
2118 for (inter = 0; inter <= 1; inter++) {
2119 for (plane = 0; plane <= 2; plane++) {
2120 int newqr= 1;
2121 if (inter || plane > 0)
2122 newqr = get_bits1(gb);
2123 if (!newqr) {
2124 int qtj, plj;
2125 if(inter && get_bits1(gb)){
2126 qtj = 0;
2127 plj = plane;
2128 }else{
2129 qtj= (3*inter + plane - 1) / 3;
2130 plj= (plane + 2) % 3;
2131 }
2132 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2133 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2134 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2135 } else {
2136 int qri= 0;
2137 int qi = 0;
2138
2139 for(;;){
2140 i= get_bits(gb, av_log2(matrices-1)+1);
2141 if(i>= matrices){
2142 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2143 return -1;
2144 }
2145 s->qr_base[inter][plane][qri]= i;
2146 if(qi >= 63)
2147 break;
2148 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2149 s->qr_size[inter][plane][qri++]= i;
2150 qi += i;
2151 }
2152
2153 if (qi > 63) {
2154 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2155 return -1;
2156 }
2157 s->qr_count[inter][plane]= qri;
2158 }
2159 }
2160 }
2161
2162 /* Huffman tables */
2163 for (s->hti = 0; s->hti < 80; s->hti++) {
2164 s->entries = 0;
2165 s->huff_code_size = 1;
2166 if (!get_bits1(gb)) {
2167 s->hbits = 0;
2168 if(read_huffman_tree(avctx, gb))
2169 return -1;
2170 s->hbits = 1;
2171 if(read_huffman_tree(avctx, gb))
2172 return -1;
2173 }
2174 }
2175
2176 s->theora_tables = 1;
2177
2178 return 0;
2179 }
2180
2181 static av_cold int theora_decode_init(AVCodecContext *avctx)
2182 {
2183 Vp3DecodeContext *s = avctx->priv_data;
2184 GetBitContext gb;
2185 int ptype;
2186 uint8_t *header_start[3];
2187 int header_len[3];
2188 int i;
2189
2190 s->theora = 1;
2191
2192 if (!avctx->extradata_size)
2193 {
2194 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2195 return -1;
2196 }
2197
2198 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2199 42, header_start, header_len) < 0) {
2200 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2201 return -1;
2202 }
2203
2204 for(i=0;i<3;i++) {
2205 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2206
2207 ptype = get_bits(&gb, 8);
2208
2209 if (!(ptype & 0x80))
2210 {
2211 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2212 // return -1;
2213 }
2214
2215 // FIXME: Check for this as well.
2216 skip_bits_long(&gb, 6*8); /* "theora" */
2217
2218 switch(ptype)
2219 {
2220 case 0x80:
2221 theora_decode_header(avctx, &gb);
2222 break;
2223 case 0x81:
2224 // FIXME: is this needed? it breaks sometimes
2225 // theora_decode_comments(avctx, gb);
2226 break;
2227 case 0x82:
2228 if (theora_decode_tables(avctx, &gb))
2229 return -1;
2230 break;
2231 default:
2232 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2233 break;
2234 }
2235 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2236 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2237 if (s->theora < 0x030200)
2238 break;
2239 }
2240
2241 return vp3_decode_init(avctx);
2242 }
2243
2244 AVCodec theora_decoder = {
2245 "theora",
2246 CODEC_TYPE_VIDEO,
2247 CODEC_ID_THEORA,
2248 sizeof(Vp3DecodeContext),
2249 theora_decode_init,
2250 NULL,
2251 vp3_decode_end,
2252 vp3_decode_frame,
2253 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2254 NULL,
2255 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2256 };
2257 #endif
2258
2259 AVCodec vp3_decoder = {
2260 "vp3",
2261 CODEC_TYPE_VIDEO,
2262 CODEC_ID_VP3,
2263 sizeof(Vp3DecodeContext),
2264 vp3_decode_init,
2265 NULL,
2266 vp3_decode_end,
2267 vp3_decode_frame,
2268 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2269 NULL,
2270 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2271 };