Delay translating DCT tokens into coefficients until immediately before IDCT
[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 //FIXME split things out into their own arrays
48 typedef struct Vp3Fragment {
49 int16_t dc;
50 uint8_t coding_method;
51 int8_t motion_x;
52 int8_t motion_y;
53 uint8_t qpi;
54 } Vp3Fragment;
55
56 #define SB_NOT_CODED 0
57 #define SB_PARTIALLY_CODED 1
58 #define SB_FULLY_CODED 2
59
60 // This is the maximum length of a single long bit run that can be encoded
61 // for superblock coding or block qps. Theora special-cases this to read a
62 // bit instead of flipping the current bit to allow for runs longer than 4129.
63 #define MAXIMUM_LONG_BIT_RUN 4129
64
65 #define MODE_INTER_NO_MV 0
66 #define MODE_INTRA 1
67 #define MODE_INTER_PLUS_MV 2
68 #define MODE_INTER_LAST_MV 3
69 #define MODE_INTER_PRIOR_LAST 4
70 #define MODE_USING_GOLDEN 5
71 #define MODE_GOLDEN_MV 6
72 #define MODE_INTER_FOURMV 7
73 #define CODING_MODE_COUNT 8
74
75 /* special internal mode */
76 #define MODE_COPY 8
77
78 /* There are 6 preset schemes, plus a free-form scheme */
79 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
80 {
81 /* scheme 1: Last motion vector dominates */
82 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
83 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
84 MODE_INTRA, MODE_USING_GOLDEN,
85 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
86
87 /* scheme 2 */
88 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
89 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
90 MODE_INTRA, MODE_USING_GOLDEN,
91 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
92
93 /* scheme 3 */
94 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
95 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
96 MODE_INTRA, MODE_USING_GOLDEN,
97 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
98
99 /* scheme 4 */
100 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
101 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
102 MODE_INTRA, MODE_USING_GOLDEN,
103 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
104
105 /* scheme 5: No motion vector dominates */
106 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
107 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
108 MODE_INTRA, MODE_USING_GOLDEN,
109 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
110
111 /* scheme 6 */
112 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
113 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
114 MODE_INTER_PLUS_MV, MODE_INTRA,
115 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
116
117 };
118
119 static const uint8_t hilbert_offset[16][2] = {
120 {0,0}, {1,0}, {1,1}, {0,1},
121 {0,2}, {0,3}, {1,3}, {1,2},
122 {2,2}, {2,3}, {3,3}, {3,2},
123 {3,1}, {2,1}, {2,0}, {3,0}
124 };
125
126 #define MIN_DEQUANT_VAL 2
127
128 typedef struct Vp3DecodeContext {
129 AVCodecContext *avctx;
130 int theora, theora_tables;
131 int version;
132 int width, height;
133 AVFrame golden_frame;
134 AVFrame last_frame;
135 AVFrame current_frame;
136 int keyframe;
137 DSPContext dsp;
138 int flipped_image;
139 int last_slice_end;
140
141 int qps[3];
142 int nqps;
143 int last_qps[3];
144
145 int superblock_count;
146 int y_superblock_width;
147 int y_superblock_height;
148 int y_superblock_count;
149 int c_superblock_width;
150 int c_superblock_height;
151 int c_superblock_count;
152 int u_superblock_start;
153 int v_superblock_start;
154 unsigned char *superblock_coding;
155
156 int macroblock_count;
157 int macroblock_width;
158 int macroblock_height;
159
160 int fragment_count;
161 int fragment_width;
162 int fragment_height;
163
164 Vp3Fragment *all_fragments;
165 int fragment_start[3];
166 int data_offset[3];
167
168 ScanTable scantable;
169
170 /* tables */
171 uint16_t coded_dc_scale_factor[64];
172 uint32_t coded_ac_scale_factor[64];
173 uint8_t base_matrix[384][64];
174 uint8_t qr_count[2][3];
175 uint8_t qr_size [2][3][64];
176 uint16_t qr_base[2][3][64];
177
178 /**
179 * This is a list of all tokens in bitstream order. Reordering takes place
180 * by pulling from each level during IDCT. As a consequence, IDCT must be
181 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
182 * otherwise. The 32 different tokens with up to 12 bits of extradata are
183 * collapsed into 3 types, packed as follows:
184 * (from the low to high bits)
185 *
186 * 2 bits: type (0,1,2)
187 * 0: EOB run, 14 bits for run length (12 needed)
188 * 1: zero run, 7 bits for run length
189 * 7 bits for the next coefficient (3 needed)
190 * 2: coefficient, 14 bits (11 needed)
191 *
192 * Coefficients are signed, so are packed in the highest bits for automatic
193 * sign extension.
194 */
195 int16_t *dct_tokens[3][64];
196 int16_t *dct_tokens_base;
197 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
198 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
199 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
200
201 /**
202 * number of blocks that contain DCT coefficients at the given level or higher
203 */
204 int num_coded_frags[3][64];
205 int total_num_coded_frags;
206
207 /* this is a list of indexes into the all_fragments array indicating
208 * which of the fragments are coded */
209 int *coded_fragment_list[3];
210
211 VLC dc_vlc[16];
212 VLC ac_vlc_1[16];
213 VLC ac_vlc_2[16];
214 VLC ac_vlc_3[16];
215 VLC ac_vlc_4[16];
216
217 VLC superblock_run_length_vlc;
218 VLC fragment_run_length_vlc;
219 VLC mode_code_vlc;
220 VLC motion_vector_vlc;
221
222 /* these arrays need to be on 16-byte boundaries since SSE2 operations
223 * index into them */
224 DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
225
226 /* This table contains superblock_count * 16 entries. Each set of 16
227 * numbers corresponds to the fragment indexes 0..15 of the superblock.
228 * An entry will be -1 to indicate that no entry corresponds to that
229 * index. */
230 int *superblock_fragments;
231
232 /* This is an array that indicates how a particular macroblock
233 * is coded. */
234 unsigned char *macroblock_coding;
235
236 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
237 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
238
239 /* Huffman decode */
240 int hti;
241 unsigned int hbits;
242 int entries;
243 int huff_code_size;
244 uint16_t huffman_table[80][32][2];
245
246 uint8_t filter_limit_values[64];
247 DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
248 } Vp3DecodeContext;
249
250 /************************************************************************
251 * VP3 specific functions
252 ************************************************************************/
253
254 /*
255 * This function sets up all of the various blocks mappings:
256 * superblocks <-> fragments, macroblocks <-> fragments,
257 * superblocks <-> macroblocks
258 *
259 * Returns 0 is successful; returns 1 if *anything* went wrong.
260 */
261 static int init_block_mapping(Vp3DecodeContext *s)
262 {
263 int i, j;
264 signed int hilbert_walk_mb[4];
265
266 int current_fragment = 0;
267 int current_width = 0;
268 int current_height = 0;
269 int right_edge = 0;
270 int bottom_edge = 0;
271 int superblock_row_inc = 0;
272 int mapping_index = 0;
273
274 static const signed char travel_width[16] = {
275 1, 1, 0, -1,
276 0, 0, 1, 0,
277 1, 0, 1, 0,
278 0, -1, 0, 1
279 };
280
281 static const signed char travel_height[16] = {
282 0, 0, 1, 0,
283 1, 1, 0, -1,
284 0, 1, 0, -1,
285 -1, 0, -1, 0
286 };
287
288 hilbert_walk_mb[0] = 1;
289 hilbert_walk_mb[1] = s->macroblock_width;
290 hilbert_walk_mb[2] = 1;
291 hilbert_walk_mb[3] = -s->macroblock_width;
292
293 /* iterate through each superblock (all planes) and map the fragments */
294 for (i = 0; i < s->superblock_count; i++) {
295 /* time to re-assign the limits? */
296 if (i == 0) {
297
298 /* start of Y superblocks */
299 right_edge = s->fragment_width;
300 bottom_edge = s->fragment_height;
301 current_width = -1;
302 current_height = 0;
303 superblock_row_inc = 3 * s->fragment_width -
304 (s->y_superblock_width * 4 - s->fragment_width);
305
306 /* the first operation for this variable is to advance by 1 */
307 current_fragment = -1;
308
309 } else if (i == s->u_superblock_start) {
310
311 /* start of U superblocks */
312 right_edge = s->fragment_width / 2;
313 bottom_edge = s->fragment_height / 2;
314 current_width = -1;
315 current_height = 0;
316 superblock_row_inc = 3 * (s->fragment_width / 2) -
317 (s->c_superblock_width * 4 - s->fragment_width / 2);
318
319 /* the first operation for this variable is to advance by 1 */
320 current_fragment = s->fragment_start[1] - 1;
321
322 } else if (i == s->v_superblock_start) {
323
324 /* start of V superblocks */
325 right_edge = s->fragment_width / 2;
326 bottom_edge = s->fragment_height / 2;
327 current_width = -1;
328 current_height = 0;
329 superblock_row_inc = 3 * (s->fragment_width / 2) -
330 (s->c_superblock_width * 4 - s->fragment_width / 2);
331
332 /* the first operation for this variable is to advance by 1 */
333 current_fragment = s->fragment_start[2] - 1;
334
335 }
336
337 if (current_width >= right_edge - 1) {
338 /* reset width and move to next superblock row */
339 current_width = -1;
340 current_height += 4;
341
342 /* fragment is now at the start of a new superblock row */
343 current_fragment += superblock_row_inc;
344 }
345
346 /* iterate through all 16 fragments in a superblock */
347 for (j = 0; j < 16; j++) {
348 current_fragment += travel_width[j] + right_edge * travel_height[j];
349 current_width += travel_width[j];
350 current_height += travel_height[j];
351
352 /* check if the fragment is in bounds */
353 if ((current_width < right_edge) &&
354 (current_height < bottom_edge)) {
355 s->superblock_fragments[mapping_index] = current_fragment;
356 } else {
357 s->superblock_fragments[mapping_index] = -1;
358 }
359
360 mapping_index++;
361 }
362 }
363
364 return 0; /* successful path out */
365 }
366
367 /*
368 * This function wipes out all of the fragment data.
369 */
370 static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
371 {
372 int i;
373
374 /* zero out all of the fragment information */
375 for (i = 0; i < s->fragment_count; i++) {
376 s->all_fragments[i].motion_x = 127;
377 s->all_fragments[i].motion_y = 127;
378 s->all_fragments[i].dc = 0;
379 s->all_fragments[i].qpi = 0;
380 }
381 }
382
383 /*
384 * This function sets up the dequantization tables used for a particular
385 * frame.
386 */
387 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
388 {
389 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
390 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
391 int i, plane, inter, qri, bmi, bmj, qistart;
392
393 for(inter=0; inter<2; inter++){
394 for(plane=0; plane<3; plane++){
395 int sum=0;
396 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
397 sum+= s->qr_size[inter][plane][qri];
398 if(s->qps[qpi] <= sum)
399 break;
400 }
401 qistart= sum - s->qr_size[inter][plane][qri];
402 bmi= s->qr_base[inter][plane][qri ];
403 bmj= s->qr_base[inter][plane][qri+1];
404 for(i=0; i<64; i++){
405 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
406 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
407 + s->qr_size[inter][plane][qri])
408 / (2*s->qr_size[inter][plane][qri]);
409
410 int qmin= 8<<(inter + !i);
411 int qscale= i ? ac_scale_factor : dc_scale_factor;
412
413 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
414 }
415 // all DC coefficients use the same quant so as not to interfere with DC prediction
416 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
417 }
418 }
419
420 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
421 }
422
423 /*
424 * This function initializes the loop filter boundary limits if the frame's
425 * quality index is different from the previous frame's.
426 *
427 * The filter_limit_values may not be larger than 127.
428 */
429 static void init_loop_filter(Vp3DecodeContext *s)
430 {
431 int *bounding_values= s->bounding_values_array+127;
432 int filter_limit;
433 int x;
434 int value;
435
436 filter_limit = s->filter_limit_values[s->qps[0]];
437
438 /* set up the bounding values */
439 memset(s->bounding_values_array, 0, 256 * sizeof(int));
440 for (x = 0; x < filter_limit; x++) {
441 bounding_values[-x] = -x;
442 bounding_values[x] = x;
443 }
444 for (x = value = filter_limit; x < 128 && value; x++, value--) {
445 bounding_values[ x] = value;
446 bounding_values[-x] = -value;
447 }
448 if (value)
449 bounding_values[128] = value;
450 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
451 }
452
453 /*
454 * This function unpacks all of the superblock/macroblock/fragment coding
455 * information from the bitstream.
456 */
457 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
458 {
459 int bit = 0;
460 int current_superblock = 0;
461 int current_run = 0;
462 int num_partial_superblocks = 0;
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->total_num_coded_frags = 0;
547 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
548
549 for (plane = 0; plane < 3; plane++) {
550 int sb_start = (int[]){ 0, s->u_superblock_start, s->v_superblock_start }[plane];
551 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
552 int num_coded_frags = 0;
553
554 for (i = sb_start; i < sb_end; i++) {
555
556 /* iterate through all 16 fragments in a superblock */
557 for (j = 0; j < 16; j++) {
558
559 /* if the fragment is in bounds, check its coding status */
560 current_fragment = s->superblock_fragments[i * 16 + j];
561 if (current_fragment >= s->fragment_count) {
562 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
563 current_fragment, s->fragment_count);
564 return 1;
565 }
566 if (current_fragment != -1) {
567 int coded = s->superblock_coding[i];
568
569 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
570
571 /* fragment may or may not be coded; this is the case
572 * that cares about the fragment coding runs */
573 if (current_run-- == 0) {
574 bit ^= 1;
575 current_run = get_vlc2(gb,
576 s->fragment_run_length_vlc.table, 5, 2);
577 }
578 coded = bit;
579 }
580
581 if (coded) {
582 /* default mode; actual mode will be decoded in
583 * the next phase */
584 s->all_fragments[current_fragment].coding_method =
585 MODE_INTER_NO_MV;
586 s->coded_fragment_list[plane][num_coded_frags++] =
587 current_fragment;
588 } else {
589 /* not coded; copy this fragment from the prior frame */
590 s->all_fragments[current_fragment].coding_method =
591 MODE_COPY;
592 }
593 }
594 }
595 }
596 s->total_num_coded_frags += num_coded_frags;
597 for (i = 0; i < 64; i++)
598 s->num_coded_frags[plane][i] = num_coded_frags;
599 if (plane < 2)
600 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
601 }
602 return 0;
603 }
604
605 /*
606 * This function unpacks all the coding mode data for individual macroblocks
607 * from the bitstream.
608 */
609 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
610 {
611 int i, j, k, sb_x, sb_y;
612 int scheme;
613 int current_macroblock;
614 int current_fragment;
615 int coding_mode;
616 int custom_mode_alphabet[CODING_MODE_COUNT];
617 const int *alphabet;
618
619 if (s->keyframe) {
620 for (i = 0; i < s->fragment_count; i++)
621 s->all_fragments[i].coding_method = MODE_INTRA;
622
623 } else {
624
625 /* fetch the mode coding scheme for this frame */
626 scheme = get_bits(gb, 3);
627
628 /* is it a custom coding scheme? */
629 if (scheme == 0) {
630 for (i = 0; i < 8; i++)
631 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
632 for (i = 0; i < 8; i++)
633 custom_mode_alphabet[get_bits(gb, 3)] = i;
634 alphabet = custom_mode_alphabet;
635 } else
636 alphabet = ModeAlphabet[scheme-1];
637
638 /* iterate through all of the macroblocks that contain 1 or more
639 * coded fragments */
640 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
641 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
642
643 for (j = 0; j < 4; j++) {
644 int mb_x = 2*sb_x + (j>>1);
645 int mb_y = 2*sb_y + (((j>>1)+j)&1);
646 current_macroblock = mb_y * s->macroblock_width + mb_x;
647
648 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
649 continue;
650
651 #define BLOCK_X (2*mb_x + (k&1))
652 #define BLOCK_Y (2*mb_y + (k>>1))
653 /* coding modes are only stored if the macroblock has at least one
654 * luma block coded, otherwise it must be INTER_NO_MV */
655 for (k = 0; k < 4; k++) {
656 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
657 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
658 break;
659 }
660 if (k == 4) {
661 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
662 continue;
663 }
664
665 /* mode 7 means get 3 bits for each coding mode */
666 if (scheme == 7)
667 coding_mode = get_bits(gb, 3);
668 else
669 coding_mode = alphabet
670 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
671
672 s->macroblock_coding[current_macroblock] = coding_mode;
673 for (k = 0; k < 4; k++) {
674 current_fragment =
675 BLOCK_Y*s->fragment_width + BLOCK_X;
676 if (s->all_fragments[current_fragment].coding_method !=
677 MODE_COPY)
678 s->all_fragments[current_fragment].coding_method =
679 coding_mode;
680 }
681 for (k = 0; k < 2; k++) {
682 current_fragment = s->fragment_start[k+1] +
683 mb_y*(s->fragment_width>>1) + mb_x;
684 if (s->all_fragments[current_fragment].coding_method !=
685 MODE_COPY)
686 s->all_fragments[current_fragment].coding_method =
687 coding_mode;
688 }
689 }
690 }
691 }
692 }
693
694 return 0;
695 }
696
697 /*
698 * This function unpacks all the motion vectors for the individual
699 * macroblocks from the bitstream.
700 */
701 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
702 {
703 int j, k, sb_x, sb_y;
704 int coding_mode;
705 int motion_x[6];
706 int motion_y[6];
707 int last_motion_x = 0;
708 int last_motion_y = 0;
709 int prior_last_motion_x = 0;
710 int prior_last_motion_y = 0;
711 int current_macroblock;
712 int current_fragment;
713
714 if (s->keyframe)
715 return 0;
716
717 memset(motion_x, 0, 6 * sizeof(int));
718 memset(motion_y, 0, 6 * sizeof(int));
719
720 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
721 coding_mode = get_bits1(gb);
722
723 /* iterate through all of the macroblocks that contain 1 or more
724 * coded fragments */
725 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
726 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
727
728 for (j = 0; j < 4; j++) {
729 int mb_x = 2*sb_x + (j>>1);
730 int mb_y = 2*sb_y + (((j>>1)+j)&1);
731 current_macroblock = mb_y * s->macroblock_width + mb_x;
732
733 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
734 (s->macroblock_coding[current_macroblock] == MODE_COPY))
735 continue;
736
737 switch (s->macroblock_coding[current_macroblock]) {
738
739 case MODE_INTER_PLUS_MV:
740 case MODE_GOLDEN_MV:
741 /* all 6 fragments use the same motion vector */
742 if (coding_mode == 0) {
743 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
744 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
745 } else {
746 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
747 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
748 }
749
750 /* vector maintenance, only on MODE_INTER_PLUS_MV */
751 if (s->macroblock_coding[current_macroblock] ==
752 MODE_INTER_PLUS_MV) {
753 prior_last_motion_x = last_motion_x;
754 prior_last_motion_y = last_motion_y;
755 last_motion_x = motion_x[0];
756 last_motion_y = motion_y[0];
757 }
758 break;
759
760 case MODE_INTER_FOURMV:
761 /* vector maintenance */
762 prior_last_motion_x = last_motion_x;
763 prior_last_motion_y = last_motion_y;
764
765 /* fetch 4 vectors from the bitstream, one for each
766 * Y fragment, then average for the C fragment vectors */
767 motion_x[4] = motion_y[4] = 0;
768 for (k = 0; k < 4; k++) {
769 current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
770 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
771 if (coding_mode == 0) {
772 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
773 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
774 } else {
775 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
776 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
777 }
778 last_motion_x = motion_x[k];
779 last_motion_y = motion_y[k];
780 } else {
781 motion_x[k] = 0;
782 motion_y[k] = 0;
783 }
784 motion_x[4] += motion_x[k];
785 motion_y[4] += motion_y[k];
786 }
787
788 motion_x[5]=
789 motion_x[4]= RSHIFT(motion_x[4], 2);
790 motion_y[5]=
791 motion_y[4]= RSHIFT(motion_y[4], 2);
792 break;
793
794 case MODE_INTER_LAST_MV:
795 /* all 6 fragments use the last motion vector */
796 motion_x[0] = last_motion_x;
797 motion_y[0] = last_motion_y;
798
799 /* no vector maintenance (last vector remains the
800 * last vector) */
801 break;
802
803 case MODE_INTER_PRIOR_LAST:
804 /* all 6 fragments use the motion vector prior to the
805 * last motion vector */
806 motion_x[0] = prior_last_motion_x;
807 motion_y[0] = prior_last_motion_y;
808
809 /* vector maintenance */
810 prior_last_motion_x = last_motion_x;
811 prior_last_motion_y = last_motion_y;
812 last_motion_x = motion_x[0];
813 last_motion_y = motion_y[0];
814 break;
815
816 default:
817 /* covers intra, inter without MV, golden without MV */
818 motion_x[0] = 0;
819 motion_y[0] = 0;
820
821 /* no vector maintenance */
822 break;
823 }
824
825 /* assign the motion vectors to the correct fragments */
826 for (k = 0; k < 4; k++) {
827 current_fragment =
828 BLOCK_Y*s->fragment_width + BLOCK_X;
829 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
830 s->all_fragments[current_fragment].motion_x = motion_x[k];
831 s->all_fragments[current_fragment].motion_y = motion_y[k];
832 } else {
833 s->all_fragments[current_fragment].motion_x = motion_x[0];
834 s->all_fragments[current_fragment].motion_y = motion_y[0];
835 }
836 }
837 for (k = 0; k < 2; k++) {
838 current_fragment = s->fragment_start[k+1] +
839 mb_y*(s->fragment_width>>1) + mb_x;
840 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
841 s->all_fragments[current_fragment].motion_x = motion_x[k+4];
842 s->all_fragments[current_fragment].motion_y = motion_y[k+4];
843 } else {
844 s->all_fragments[current_fragment].motion_x = motion_x[0];
845 s->all_fragments[current_fragment].motion_y = motion_y[0];
846 }
847 }
848 }
849 }
850 }
851
852 return 0;
853 }
854
855 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
856 {
857 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
858 int num_blocks = s->total_num_coded_frags;
859
860 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
861 i = blocks_decoded = num_blocks_at_qpi = 0;
862
863 bit = get_bits1(gb);
864
865 do {
866 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
867 if (run_length == 34)
868 run_length += get_bits(gb, 12);
869 blocks_decoded += run_length;
870
871 if (!bit)
872 num_blocks_at_qpi += run_length;
873
874 for (j = 0; j < run_length; i++) {
875 if (i >= s->total_num_coded_frags)
876 return -1;
877
878 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
879 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
880 j++;
881 }
882 }
883
884 if (run_length == MAXIMUM_LONG_BIT_RUN)
885 bit = get_bits1(gb);
886 else
887 bit ^= 1;
888 } while (blocks_decoded < num_blocks);
889
890 num_blocks -= num_blocks_at_qpi;
891 }
892
893 return 0;
894 }
895
896 /*
897 * This function is called by unpack_dct_coeffs() to extract the VLCs from
898 * the bitstream. The VLCs encode tokens which are used to unpack DCT
899 * data. This function unpacks all the VLCs for either the Y plane or both
900 * C planes, and is called for DC coefficients or different AC coefficient
901 * levels (since different coefficient types require different VLC tables.
902 *
903 * This function returns a residual eob run. E.g, if a particular token gave
904 * instructions to EOB the next 5 fragments and there were only 2 fragments
905 * left in the current fragment range, 3 would be returned so that it could
906 * be passed into the next call to this same function.
907 */
908 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
909 VLC *table, int coeff_index,
910 int plane,
911 int eob_run)
912 {
913 int i, j = 0;
914 int token;
915 int zero_run = 0;
916 DCTELEM coeff = 0;
917 int bits_to_get;
918 int blocks_ended;
919 int coeff_i = 0;
920 int num_coeffs = s->num_coded_frags[plane][coeff_index];
921 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
922
923 /* local references to structure members to avoid repeated deferences */
924 int *coded_fragment_list = s->coded_fragment_list[plane];
925 Vp3Fragment *all_fragments = s->all_fragments;
926 VLC_TYPE (*vlc_table)[2] = table->table;
927
928 if (num_coeffs < 0)
929 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
930
931 if (eob_run > num_coeffs) {
932 coeff_i = blocks_ended = num_coeffs;
933 eob_run -= num_coeffs;
934 } else {
935 coeff_i = blocks_ended = eob_run;
936 eob_run = 0;
937 }
938
939 // insert fake EOB token to cover the split between planes or zzi
940 if (blocks_ended)
941 dct_tokens[j++] = blocks_ended << 2;
942
943 while (coeff_i < num_coeffs) {
944 /* decode a VLC into a token */
945 token = get_vlc2(gb, vlc_table, 5, 3);
946 /* use the token to get a zero run, a coefficient, and an eob run */
947 if (token <= 6) {
948 eob_run = eob_run_base[token];
949 if (eob_run_get_bits[token])
950 eob_run += get_bits(gb, eob_run_get_bits[token]);
951
952 // record only the number of blocks ended in this plane,
953 // any spill will be recorded in the next plane.
954 if (eob_run > num_coeffs - coeff_i) {
955 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
956 blocks_ended += num_coeffs - coeff_i;
957 eob_run -= num_coeffs - coeff_i;
958 coeff_i = num_coeffs;
959 } else {
960 dct_tokens[j++] = TOKEN_EOB(eob_run);
961 blocks_ended += eob_run;
962 coeff_i += eob_run;
963 eob_run = 0;
964 }
965 } else {
966 bits_to_get = coeff_get_bits[token];
967 if (bits_to_get)
968 bits_to_get = get_bits(gb, bits_to_get);
969 coeff = coeff_tables[token][bits_to_get];
970
971 zero_run = zero_run_base[token];
972 if (zero_run_get_bits[token])
973 zero_run += get_bits(gb, zero_run_get_bits[token]);
974
975 if (zero_run) {
976 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
977 } else {
978 // Save DC into the fragment structure. DC prediction is
979 // done in raster order, so the actual DC can't be in with
980 // other tokens. We still need the token in dct_tokens[]
981 // however, or else the structure collapses on itself.
982 if (!coeff_index)
983 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
984
985 dct_tokens[j++] = TOKEN_COEFF(coeff);
986 }
987
988 if (coeff_index + zero_run > 64) {
989 av_log(s->avctx, AV_LOG_ERROR, "Invalid zero run of %d with"
990 " %d coeffs left\n", zero_run, 64-coeff_index);
991 zero_run = 64 - coeff_index;
992 }
993
994 // zero runs code multiple coefficients,
995 // so don't try to decode coeffs for those higher levels
996 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
997 s->num_coded_frags[plane][i]--;
998 coeff_i++;
999 }
1000 }
1001
1002 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1003 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1004
1005 // decrement the number of blocks that have higher coeffecients for each
1006 // EOB run at this level
1007 if (blocks_ended)
1008 for (i = coeff_index+1; i < 64; i++)
1009 s->num_coded_frags[plane][i] -= blocks_ended;
1010
1011 // setup the next buffer
1012 if (plane < 2)
1013 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
1014 else if (coeff_index < 63)
1015 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
1016
1017 return eob_run;
1018 }
1019
1020 static void reverse_dc_prediction(Vp3DecodeContext *s,
1021 int first_fragment,
1022 int fragment_width,
1023 int fragment_height);
1024 /*
1025 * This function unpacks all of the DCT coefficient data from the
1026 * bitstream.
1027 */
1028 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1029 {
1030 int i;
1031 int dc_y_table;
1032 int dc_c_table;
1033 int ac_y_table;
1034 int ac_c_table;
1035 int residual_eob_run = 0;
1036 VLC *y_tables[64];
1037 VLC *c_tables[64];
1038
1039 s->dct_tokens[0][0] = s->dct_tokens_base;
1040
1041 /* fetch the DC table indexes */
1042 dc_y_table = get_bits(gb, 4);
1043 dc_c_table = get_bits(gb, 4);
1044
1045 /* unpack the Y plane DC coefficients */
1046 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1047 0, residual_eob_run);
1048
1049 /* reverse prediction of the Y-plane DC coefficients */
1050 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1051
1052 /* unpack the C plane DC coefficients */
1053 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1054 1, residual_eob_run);
1055 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1056 2, residual_eob_run);
1057
1058 /* reverse prediction of the C-plane DC coefficients */
1059 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1060 {
1061 reverse_dc_prediction(s, s->fragment_start[1],
1062 s->fragment_width / 2, s->fragment_height / 2);
1063 reverse_dc_prediction(s, s->fragment_start[2],
1064 s->fragment_width / 2, s->fragment_height / 2);
1065 }
1066
1067 /* fetch the AC table indexes */
1068 ac_y_table = get_bits(gb, 4);
1069 ac_c_table = get_bits(gb, 4);
1070
1071 /* build tables of AC VLC tables */
1072 for (i = 1; i <= 5; i++) {
1073 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1074 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1075 }
1076 for (i = 6; i <= 14; i++) {
1077 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1078 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1079 }
1080 for (i = 15; i <= 27; i++) {
1081 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1082 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1083 }
1084 for (i = 28; i <= 63; i++) {
1085 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1086 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1087 }
1088
1089 /* decode all AC coefficents */
1090 for (i = 1; i <= 63; i++) {
1091 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1092 0, residual_eob_run);
1093
1094 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1095 1, residual_eob_run);
1096 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1097 2, residual_eob_run);
1098 }
1099
1100 return 0;
1101 }
1102
1103 /*
1104 * This function reverses the DC prediction for each coded fragment in
1105 * the frame. Much of this function is adapted directly from the original
1106 * VP3 source code.
1107 */
1108 #define COMPATIBLE_FRAME(x) \
1109 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1110 #define DC_COEFF(u) s->all_fragments[u].dc
1111
1112 static void reverse_dc_prediction(Vp3DecodeContext *s,
1113 int first_fragment,
1114 int fragment_width,
1115 int fragment_height)
1116 {
1117
1118 #define PUL 8
1119 #define PU 4
1120 #define PUR 2
1121 #define PL 1
1122
1123 int x, y;
1124 int i = first_fragment;
1125
1126 int predicted_dc;
1127
1128 /* DC values for the left, up-left, up, and up-right fragments */
1129 int vl, vul, vu, vur;
1130
1131 /* indexes for the left, up-left, up, and up-right fragments */
1132 int l, ul, u, ur;
1133
1134 /*
1135 * The 6 fields mean:
1136 * 0: up-left multiplier
1137 * 1: up multiplier
1138 * 2: up-right multiplier
1139 * 3: left multiplier
1140 */
1141 static const int predictor_transform[16][4] = {
1142 { 0, 0, 0, 0},
1143 { 0, 0, 0,128}, // PL
1144 { 0, 0,128, 0}, // PUR
1145 { 0, 0, 53, 75}, // PUR|PL
1146 { 0,128, 0, 0}, // PU
1147 { 0, 64, 0, 64}, // PU|PL
1148 { 0,128, 0, 0}, // PU|PUR
1149 { 0, 0, 53, 75}, // PU|PUR|PL
1150 {128, 0, 0, 0}, // PUL
1151 { 0, 0, 0,128}, // PUL|PL
1152 { 64, 0, 64, 0}, // PUL|PUR
1153 { 0, 0, 53, 75}, // PUL|PUR|PL
1154 { 0,128, 0, 0}, // PUL|PU
1155 {-104,116, 0,116}, // PUL|PU|PL
1156 { 24, 80, 24, 0}, // PUL|PU|PUR
1157 {-104,116, 0,116} // PUL|PU|PUR|PL
1158 };
1159
1160 /* This table shows which types of blocks can use other blocks for
1161 * prediction. For example, INTRA is the only mode in this table to
1162 * have a frame number of 0. That means INTRA blocks can only predict
1163 * from other INTRA blocks. There are 2 golden frame coding types;
1164 * blocks encoding in these modes can only predict from other blocks
1165 * that were encoded with these 1 of these 2 modes. */
1166 static const unsigned char compatible_frame[9] = {
1167 1, /* MODE_INTER_NO_MV */
1168 0, /* MODE_INTRA */
1169 1, /* MODE_INTER_PLUS_MV */
1170 1, /* MODE_INTER_LAST_MV */
1171 1, /* MODE_INTER_PRIOR_MV */
1172 2, /* MODE_USING_GOLDEN */
1173 2, /* MODE_GOLDEN_MV */
1174 1, /* MODE_INTER_FOUR_MV */
1175 3 /* MODE_COPY */
1176 };
1177 int current_frame_type;
1178
1179 /* there is a last DC predictor for each of the 3 frame types */
1180 short last_dc[3];
1181
1182 int transform = 0;
1183
1184 vul = vu = vur = vl = 0;
1185 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1186
1187 /* for each fragment row... */
1188 for (y = 0; y < fragment_height; y++) {
1189
1190 /* for each fragment in a row... */
1191 for (x = 0; x < fragment_width; x++, i++) {
1192
1193 /* reverse prediction if this block was coded */
1194 if (s->all_fragments[i].coding_method != MODE_COPY) {
1195
1196 current_frame_type =
1197 compatible_frame[s->all_fragments[i].coding_method];
1198
1199 transform= 0;
1200 if(x){
1201 l= i-1;
1202 vl = DC_COEFF(l);
1203 if(COMPATIBLE_FRAME(l))
1204 transform |= PL;
1205 }
1206 if(y){
1207 u= i-fragment_width;
1208 vu = DC_COEFF(u);
1209 if(COMPATIBLE_FRAME(u))
1210 transform |= PU;
1211 if(x){
1212 ul= i-fragment_width-1;
1213 vul = DC_COEFF(ul);
1214 if(COMPATIBLE_FRAME(ul))
1215 transform |= PUL;
1216 }
1217 if(x + 1 < fragment_width){
1218 ur= i-fragment_width+1;
1219 vur = DC_COEFF(ur);
1220 if(COMPATIBLE_FRAME(ur))
1221 transform |= PUR;
1222 }
1223 }
1224
1225 if (transform == 0) {
1226
1227 /* if there were no fragments to predict from, use last
1228 * DC saved */
1229 predicted_dc = last_dc[current_frame_type];
1230 } else {
1231
1232 /* apply the appropriate predictor transform */
1233 predicted_dc =
1234 (predictor_transform[transform][0] * vul) +
1235 (predictor_transform[transform][1] * vu) +
1236 (predictor_transform[transform][2] * vur) +
1237 (predictor_transform[transform][3] * vl);
1238
1239 predicted_dc /= 128;
1240
1241 /* check for outranging on the [ul u l] and
1242 * [ul u ur l] predictors */
1243 if ((transform == 15) || (transform == 13)) {
1244 if (FFABS(predicted_dc - vu) > 128)
1245 predicted_dc = vu;
1246 else if (FFABS(predicted_dc - vl) > 128)
1247 predicted_dc = vl;
1248 else if (FFABS(predicted_dc - vul) > 128)
1249 predicted_dc = vul;
1250 }
1251 }
1252
1253 /* at long last, apply the predictor */
1254 DC_COEFF(i) += predicted_dc;
1255 /* save the DC */
1256 last_dc[current_frame_type] = DC_COEFF(i);
1257 }
1258 }
1259 }
1260 }
1261
1262 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1263 {
1264 int x, y;
1265 int *bounding_values= s->bounding_values_array+127;
1266
1267 int width = s->fragment_width >> !!plane;
1268 int height = s->fragment_height >> !!plane;
1269 int fragment = s->fragment_start [plane] + ystart * width;
1270 int stride = s->current_frame.linesize[plane];
1271 uint8_t *plane_data = s->current_frame.data [plane];
1272 if (!s->flipped_image) stride = -stride;
1273 plane_data += s->data_offset[plane] + 8*ystart*stride;
1274
1275 for (y = ystart; y < yend; y++) {
1276
1277 for (x = 0; x < width; x++) {
1278 /* This code basically just deblocks on the edges of coded blocks.
1279 * However, it has to be much more complicated because of the
1280 * braindamaged deblock ordering used in VP3/Theora. Order matters
1281 * because some pixels get filtered twice. */
1282 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1283 {
1284 /* do not perform left edge filter for left columns frags */
1285 if (x > 0) {
1286 s->dsp.vp3_h_loop_filter(
1287 plane_data + 8*x,
1288 stride, bounding_values);
1289 }
1290
1291 /* do not perform top edge filter for top row fragments */
1292 if (y > 0) {
1293 s->dsp.vp3_v_loop_filter(
1294 plane_data + 8*x,
1295 stride, bounding_values);
1296 }
1297
1298 /* do not perform right edge filter for right column
1299 * fragments or if right fragment neighbor is also coded
1300 * in this frame (it will be filtered in next iteration) */
1301 if ((x < width - 1) &&
1302 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1303 s->dsp.vp3_h_loop_filter(
1304 plane_data + 8*x + 8,
1305 stride, bounding_values);
1306 }
1307
1308 /* do not perform bottom edge filter for bottom row
1309 * fragments or if bottom fragment neighbor is also coded
1310 * in this frame (it will be filtered in the next row) */
1311 if ((y < height - 1) &&
1312 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1313 s->dsp.vp3_v_loop_filter(
1314 plane_data + 8*x + 8*stride,
1315 stride, bounding_values);
1316 }
1317 }
1318
1319 fragment++;
1320 }
1321 plane_data += 8*stride;
1322 }
1323 }
1324
1325 /**
1326 * Pulls DCT tokens from the 64 levels to decode and dequant the coefficients
1327 * for the next block in coding order
1328 */
1329 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1330 int plane, int inter, DCTELEM block[64])
1331 {
1332 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1333 uint8_t *perm = s->scantable.permutated;
1334 int i = 0;
1335
1336 do {
1337 int token = *s->dct_tokens[plane][i];
1338 switch (token & 3) {
1339 case 0: // EOB
1340 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1341 s->dct_tokens[plane][i]++;
1342 else
1343 *s->dct_tokens[plane][i] = token & ~3;
1344 goto end;
1345 case 1: // zero run
1346 s->dct_tokens[plane][i]++;
1347 i += (token >> 2) & 0x7f;
1348 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1349 i++;
1350 break;
1351 case 2: // coeff
1352 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1353 s->dct_tokens[plane][i++]++;
1354 break;
1355 default:
1356 av_log(s->avctx, AV_LOG_ERROR, "internal: invalid token type\n");
1357 return i;
1358 }
1359 } while (i < 64);
1360 end:
1361 // the actual DC+prediction is in the fragment structure
1362 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1363 return i;
1364 }
1365
1366 /**
1367 * called when all pixels up to row y are complete
1368 */
1369 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1370 {
1371 int h, cy;
1372 int offset[4];
1373
1374 if(s->avctx->draw_horiz_band==NULL)
1375 return;
1376
1377 h= y - s->last_slice_end;
1378 y -= h;
1379
1380 if (!s->flipped_image) {
1381 if (y == 0)
1382 h -= s->height - s->avctx->height; // account for non-mod16
1383 y = s->height - y - h;
1384 }
1385
1386 cy = y >> 1;
1387 offset[0] = s->current_frame.linesize[0]*y;
1388 offset[1] = s->current_frame.linesize[1]*cy;
1389 offset[2] = s->current_frame.linesize[2]*cy;
1390 offset[3] = 0;
1391
1392 emms_c();
1393 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1394 s->last_slice_end= y + h;
1395 }
1396
1397 /*
1398 * Perform the final rendering for a particular slice of data.
1399 * The slice number ranges from 0..(c_superblock_height - 1).
1400 */
1401 static void render_slice(Vp3DecodeContext *s, int slice)
1402 {
1403 int x, y, i, j;
1404 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1405 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1406 int motion_halfpel_index;
1407 uint8_t *motion_source;
1408 int plane, first_pixel;
1409
1410 if (slice >= s->c_superblock_height)
1411 return;
1412
1413 for (plane = 0; plane < 3; plane++) {
1414 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1415 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1416 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1417 int stride = s->current_frame.linesize[plane];
1418 int plane_width = s->width >> !!plane;
1419 int plane_height = s->height >> !!plane;
1420
1421 int sb_x, sb_y = slice << !plane;
1422 int slice_height = sb_y + (plane ? 1 : 2);
1423 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1424
1425 int fragment_width = s->fragment_width >> !!plane;
1426 int fragment_height = s->fragment_height >> !!plane;
1427 int fragment_start = s->fragment_start[plane];
1428
1429 if (!s->flipped_image) stride = -stride;
1430 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1431 continue;
1432
1433
1434 if(FFABS(stride) > 2048)
1435 return; //various tables are fixed size
1436
1437 /* for each superblock row in the slice (both of them)... */
1438 for (; sb_y < slice_height; sb_y++) {
1439
1440 /* for each superblock in a row... */
1441 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1442
1443 /* for each block in a superblock... */
1444 for (j = 0; j < 16; j++) {
1445 x = 4*sb_x + hilbert_offset[j][0];
1446 y = 4*sb_y + hilbert_offset[j][1];
1447
1448 i = fragment_start + y*fragment_width + x;
1449
1450 // bounds check
1451 if (x >= fragment_width || y >= fragment_height)
1452 continue;
1453
1454 first_pixel = 8*y*stride + 8*x;
1455
1456 /* transform if this block was coded */
1457 if (s->all_fragments[i].coding_method != MODE_COPY) {
1458 int intra = s->all_fragments[i].coding_method == MODE_INTRA;
1459
1460 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1461 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1462 motion_source= golden_plane;
1463 else
1464 motion_source= last_plane;
1465
1466 motion_source += first_pixel;
1467 motion_halfpel_index = 0;
1468
1469 /* sort out the motion vector if this fragment is coded
1470 * using a motion vector method */
1471 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1472 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1473 int src_x, src_y;
1474 motion_x = s->all_fragments[i].motion_x;
1475 motion_y = s->all_fragments[i].motion_y;
1476 if(plane){
1477 motion_x= (motion_x>>1) | (motion_x&1);
1478 motion_y= (motion_y>>1) | (motion_y&1);
1479 }
1480
1481 src_x= (motion_x>>1) + 8*x;
1482 src_y= (motion_y>>1) + 8*y;
1483 if ((motion_x == 127) || (motion_y == 127))
1484 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1485
1486 motion_halfpel_index = motion_x & 0x01;
1487 motion_source += (motion_x >> 1);
1488
1489 motion_halfpel_index |= (motion_y & 0x01) << 1;
1490 motion_source += ((motion_y >> 1) * stride);
1491
1492 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1493 uint8_t *temp= s->edge_emu_buffer;
1494 if(stride<0) temp -= 9*stride;
1495 else temp += 9*stride;
1496
1497 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1498 motion_source= temp;
1499 }
1500 }
1501
1502
1503 /* first, take care of copying a block from either the
1504 * previous or the golden frame */
1505 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1506 /* Note, it is possible to implement all MC cases with
1507 put_no_rnd_pixels_l2 which would look more like the
1508 VP3 source but this would be slower as
1509 put_no_rnd_pixels_tab is better optimzed */
1510 if(motion_halfpel_index != 3){
1511 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1512 output_plane + first_pixel,
1513 motion_source, stride, 8);
1514 }else{
1515 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1516 s->dsp.put_no_rnd_pixels_l2[1](
1517 output_plane + first_pixel,
1518 motion_source - d,
1519 motion_source + stride + 1 + d,
1520 stride, 8);
1521 }
1522 }
1523
1524 s->dsp.clear_block(block);
1525 vp3_dequant(s, s->all_fragments + i, plane, !intra, block);
1526
1527 /* invert DCT and place (or add) in final output */
1528
1529 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1530 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1531 block[0] += 128<<3;
1532 s->dsp.idct_put(
1533 output_plane + first_pixel,
1534 stride,
1535 block);
1536 } else {
1537 s->dsp.idct_add(
1538 output_plane + first_pixel,
1539 stride,
1540 block);
1541 }
1542 } else {
1543
1544 /* copy directly from the previous frame */
1545 s->dsp.put_pixels_tab[1][0](
1546 output_plane + first_pixel,
1547 last_plane + first_pixel,
1548 stride, 8);
1549
1550 }
1551 }
1552 }
1553
1554 // Filter up to the last row in the superblock row
1555 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1556 }
1557 }
1558
1559 /* this looks like a good place for slice dispatch... */
1560 /* algorithm:
1561 * if (slice == s->macroblock_height - 1)
1562 * dispatch (both last slice & 2nd-to-last slice);
1563 * else if (slice > 0)
1564 * dispatch (slice - 1);
1565 */
1566
1567 vp3_draw_horiz_band(s, 64*slice + 64-16);
1568 }
1569
1570 /*
1571 * This is the ffmpeg/libavcodec API init function.
1572 */
1573 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1574 {
1575 Vp3DecodeContext *s = avctx->priv_data;
1576 int i, inter, plane;
1577 int c_width;
1578 int c_height;
1579
1580 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1581 s->version = 0;
1582 else
1583 s->version = 1;
1584
1585 s->avctx = avctx;
1586 s->width = FFALIGN(avctx->width, 16);
1587 s->height = FFALIGN(avctx->height, 16);
1588 avctx->pix_fmt = PIX_FMT_YUV420P;
1589 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1590 if(avctx->idct_algo==FF_IDCT_AUTO)
1591 avctx->idct_algo=FF_IDCT_VP3;
1592 dsputil_init(&s->dsp, avctx);
1593
1594 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1595
1596 /* initialize to an impossible value which will force a recalculation
1597 * in the first frame decode */
1598 for (i = 0; i < 3; i++)
1599 s->qps[i] = -1;
1600
1601 s->y_superblock_width = (s->width + 31) / 32;
1602 s->y_superblock_height = (s->height + 31) / 32;
1603 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1604
1605 /* work out the dimensions for the C planes */
1606 c_width = s->width / 2;
1607 c_height = s->height / 2;
1608 s->c_superblock_width = (c_width + 31) / 32;
1609 s->c_superblock_height = (c_height + 31) / 32;
1610 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1611
1612 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1613 s->u_superblock_start = s->y_superblock_count;
1614 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1615 s->superblock_coding = av_malloc(s->superblock_count);
1616
1617 s->macroblock_width = (s->width + 15) / 16;
1618 s->macroblock_height = (s->height + 15) / 16;
1619 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1620
1621 s->fragment_width = s->width / FRAGMENT_PIXELS;
1622 s->fragment_height = s->height / FRAGMENT_PIXELS;
1623
1624 /* fragment count covers all 8x8 blocks for all 3 planes */
1625 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1626 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1627 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1628
1629 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1630 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1631 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1632 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1633 !s->coded_fragment_list[0]) {
1634 vp3_decode_end(avctx);
1635 return -1;
1636 }
1637
1638 if (!s->theora_tables)
1639 {
1640 for (i = 0; i < 64; i++) {
1641 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1642 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1643 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1644 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1645 s->base_matrix[2][i] = vp31_inter_dequant[i];
1646 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1647 }
1648
1649 for(inter=0; inter<2; inter++){
1650 for(plane=0; plane<3; plane++){
1651 s->qr_count[inter][plane]= 1;
1652 s->qr_size [inter][plane][0]= 63;
1653 s->qr_base [inter][plane][0]=
1654 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1655 }
1656 }
1657
1658 /* init VLC tables */
1659 for (i = 0; i < 16; i++) {
1660
1661 /* DC histograms */
1662 init_vlc(&s->dc_vlc[i], 5, 32,
1663 &dc_bias[i][0][1], 4, 2,
1664 &dc_bias[i][0][0], 4, 2, 0);
1665
1666 /* group 1 AC histograms */
1667 init_vlc(&s->ac_vlc_1[i], 5, 32,
1668 &ac_bias_0[i][0][1], 4, 2,
1669 &ac_bias_0[i][0][0], 4, 2, 0);
1670
1671 /* group 2 AC histograms */
1672 init_vlc(&s->ac_vlc_2[i], 5, 32,
1673 &ac_bias_1[i][0][1], 4, 2,
1674 &ac_bias_1[i][0][0], 4, 2, 0);
1675
1676 /* group 3 AC histograms */
1677 init_vlc(&s->ac_vlc_3[i], 5, 32,
1678 &ac_bias_2[i][0][1], 4, 2,
1679 &ac_bias_2[i][0][0], 4, 2, 0);
1680
1681 /* group 4 AC histograms */
1682 init_vlc(&s->ac_vlc_4[i], 5, 32,
1683 &ac_bias_3[i][0][1], 4, 2,
1684 &ac_bias_3[i][0][0], 4, 2, 0);
1685 }
1686 } else {
1687 for (i = 0; i < 16; i++) {
1688
1689 /* DC histograms */
1690 if (init_vlc(&s->dc_vlc[i], 5, 32,
1691 &s->huffman_table[i][0][1], 4, 2,
1692 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1693 goto vlc_fail;
1694
1695 /* group 1 AC histograms */
1696 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1697 &s->huffman_table[i+16][0][1], 4, 2,
1698 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1699 goto vlc_fail;
1700
1701 /* group 2 AC histograms */
1702 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1703 &s->huffman_table[i+16*2][0][1], 4, 2,
1704 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1705 goto vlc_fail;
1706
1707 /* group 3 AC histograms */
1708 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1709 &s->huffman_table[i+16*3][0][1], 4, 2,
1710 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1711 goto vlc_fail;
1712
1713 /* group 4 AC histograms */
1714 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1715 &s->huffman_table[i+16*4][0][1], 4, 2,
1716 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1717 goto vlc_fail;
1718 }
1719 }
1720
1721 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1722 &superblock_run_length_vlc_table[0][1], 4, 2,
1723 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1724
1725 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1726 &fragment_run_length_vlc_table[0][1], 4, 2,
1727 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1728
1729 init_vlc(&s->mode_code_vlc, 3, 8,
1730 &mode_code_vlc_table[0][1], 2, 1,
1731 &mode_code_vlc_table[0][0], 2, 1, 0);
1732
1733 init_vlc(&s->motion_vector_vlc, 6, 63,
1734 &motion_vector_vlc_table[0][1], 2, 1,
1735 &motion_vector_vlc_table[0][0], 2, 1, 0);
1736
1737 /* work out the block mapping tables */
1738 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1739 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1740 if (!s->superblock_fragments || !s->macroblock_coding) {
1741 vp3_decode_end(avctx);
1742 return -1;
1743 }
1744 init_block_mapping(s);
1745
1746 for (i = 0; i < 3; i++) {
1747 s->current_frame.data[i] = NULL;
1748 s->last_frame.data[i] = NULL;
1749 s->golden_frame.data[i] = NULL;
1750 }
1751
1752 return 0;
1753
1754 vlc_fail:
1755 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1756 return -1;
1757 }
1758
1759 /*
1760 * This is the ffmpeg/libavcodec API frame decode function.
1761 */
1762 static int vp3_decode_frame(AVCodecContext *avctx,
1763 void *data, int *data_size,
1764 AVPacket *avpkt)
1765 {
1766 const uint8_t *buf = avpkt->data;
1767 int buf_size = avpkt->size;
1768 Vp3DecodeContext *s = avctx->priv_data;
1769 GetBitContext gb;
1770 static int counter = 0;
1771 int i;
1772
1773 init_get_bits(&gb, buf, buf_size * 8);
1774
1775 if (s->theora && get_bits1(&gb))
1776 {
1777 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1778 return -1;
1779 }
1780
1781 s->keyframe = !get_bits1(&gb);
1782 if (!s->theora)
1783 skip_bits(&gb, 1);
1784 for (i = 0; i < 3; i++)
1785 s->last_qps[i] = s->qps[i];
1786
1787 s->nqps=0;
1788 do{
1789 s->qps[s->nqps++]= get_bits(&gb, 6);
1790 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1791 for (i = s->nqps; i < 3; i++)
1792 s->qps[i] = -1;
1793
1794 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1795 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1796 s->keyframe?"key":"", counter, s->qps[0]);
1797 counter++;
1798
1799 if (s->qps[0] != s->last_qps[0])
1800 init_loop_filter(s);
1801
1802 for (i = 0; i < s->nqps; i++)
1803 // reinit all dequantizers if the first one changed, because
1804 // the DC of the first quantizer must be used for all matrices
1805 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1806 init_dequantizer(s, i);
1807
1808 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1809 return buf_size;
1810
1811 if (s->keyframe) {
1812 if (!s->theora)
1813 {
1814 skip_bits(&gb, 4); /* width code */
1815 skip_bits(&gb, 4); /* height code */
1816 if (s->version)
1817 {
1818 s->version = get_bits(&gb, 5);
1819 if (counter == 1)
1820 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1821 }
1822 }
1823 if (s->version || s->theora)
1824 {
1825 if (get_bits1(&gb))
1826 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1827 skip_bits(&gb, 2); /* reserved? */
1828 }
1829
1830 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1831 if (s->golden_frame.data[0])
1832 avctx->release_buffer(avctx, &s->golden_frame);
1833 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1834 } else {
1835 if (s->golden_frame.data[0])
1836 avctx->release_buffer(avctx, &s->golden_frame);
1837 if (s->last_frame.data[0])
1838 avctx->release_buffer(avctx, &s->last_frame);
1839 }
1840
1841 s->golden_frame.reference = 3;
1842 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1843 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1844 return -1;
1845 }
1846
1847 /* golden frame is also the current frame */
1848 s->current_frame= s->golden_frame;
1849 } else {
1850 /* allocate a new current frame */
1851 s->current_frame.reference = 3;
1852 if (!s->golden_frame.data[0]) {
1853 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1854 return -1;
1855 }
1856 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1857 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1858 return -1;
1859 }
1860 }
1861
1862 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1863 s->current_frame.qstride= 0;
1864
1865 init_frame(s, &gb);
1866
1867 if (unpack_superblocks(s, &gb)){
1868 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1869 return -1;
1870 }
1871 if (unpack_modes(s, &gb)){
1872 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1873 return -1;
1874 }
1875 if (unpack_vectors(s, &gb)){
1876 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1877 return -1;
1878 }
1879 if (unpack_block_qpis(s, &gb)){
1880 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1881 return -1;
1882 }
1883 if (unpack_dct_coeffs(s, &gb)){
1884 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1885 return -1;
1886 }
1887
1888 for (i = 0; i < 3; i++) {
1889 if (s->flipped_image)
1890 s->data_offset[i] = 0;
1891 else
1892 s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1893 }
1894
1895 s->last_slice_end = 0;
1896 for (i = 0; i < s->c_superblock_height; i++)
1897 render_slice(s, i);
1898
1899 // filter the last row
1900 for (i = 0; i < 3; i++) {
1901 int row = (s->height >> (3+!!i)) - 1;
1902 apply_loop_filter(s, i, row, row+1);
1903 }
1904 vp3_draw_horiz_band(s, s->height);
1905
1906 *data_size=sizeof(AVFrame);
1907 *(AVFrame*)data= s->current_frame;
1908
1909 /* release the last frame, if it is allocated and if it is not the
1910 * golden frame */
1911 if ((s->last_frame.data[0]) &&
1912 (s->last_frame.data[0] != s->golden_frame.data[0]))
1913 avctx->release_buffer(avctx, &s->last_frame);
1914
1915 /* shuffle frames (last = current) */
1916 s->last_frame= s->current_frame;
1917 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1918
1919 return buf_size;
1920 }
1921
1922 /*
1923 * This is the ffmpeg/libavcodec API module cleanup function.
1924 */
1925 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1926 {
1927 Vp3DecodeContext *s = avctx->priv_data;
1928 int i;
1929
1930 av_free(s->superblock_coding);
1931 av_free(s->all_fragments);
1932 av_free(s->coded_fragment_list[0]);
1933 av_free(s->dct_tokens_base);
1934 av_free(s->superblock_fragments);
1935 av_free(s->macroblock_coding);
1936
1937 for (i = 0; i < 16; i++) {
1938 free_vlc(&s->dc_vlc[i]);
1939 free_vlc(&s->ac_vlc_1[i]);
1940 free_vlc(&s->ac_vlc_2[i]);
1941 free_vlc(&s->ac_vlc_3[i]);
1942 free_vlc(&s->ac_vlc_4[i]);
1943 }
1944
1945 free_vlc(&s->superblock_run_length_vlc);
1946 free_vlc(&s->fragment_run_length_vlc);
1947 free_vlc(&s->mode_code_vlc);
1948 free_vlc(&s->motion_vector_vlc);
1949
1950 /* release all frames */
1951 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1952 avctx->release_buffer(avctx, &s->golden_frame);
1953 if (s->last_frame.data[0])
1954 avctx->release_buffer(avctx, &s->last_frame);
1955 /* no need to release the current_frame since it will always be pointing
1956 * to the same frame as either the golden or last frame */
1957
1958 return 0;
1959 }
1960
1961 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1962 {
1963 Vp3DecodeContext *s = avctx->priv_data;
1964
1965 if (get_bits1(gb)) {
1966 int token;
1967 if (s->entries >= 32) { /* overflow */
1968 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1969 return -1;
1970 }
1971 token = get_bits(gb, 5);
1972 //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);
1973 s->huffman_table[s->hti][token][0] = s->hbits;
1974 s->huffman_table[s->hti][token][1] = s->huff_code_size;
1975 s->entries++;
1976 }
1977 else {
1978 if (s->huff_code_size >= 32) {/* overflow */
1979 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1980 return -1;
1981 }
1982 s->huff_code_size++;
1983 s->hbits <<= 1;
1984 if (read_huffman_tree(avctx, gb))
1985 return -1;
1986 s->hbits |= 1;
1987 if (read_huffman_tree(avctx, gb))
1988 return -1;
1989 s->hbits >>= 1;
1990 s->huff_code_size--;
1991 }
1992 return 0;
1993 }
1994
1995 #if CONFIG_THEORA_DECODER
1996 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1997 {
1998 Vp3DecodeContext *s = avctx->priv_data;
1999 int visible_width, visible_height, colorspace;
2000
2001 s->theora = get_bits_long(gb, 24);
2002 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2003
2004 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2005 /* but previous versions have the image flipped relative to vp3 */
2006 if (s->theora < 0x030200)
2007 {
2008 s->flipped_image = 1;
2009 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2010 }
2011
2012 visible_width = s->width = get_bits(gb, 16) << 4;
2013 visible_height = s->height = get_bits(gb, 16) << 4;
2014
2015 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2016 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2017 s->width= s->height= 0;
2018 return -1;
2019 }
2020
2021 if (s->theora >= 0x030200) {
2022 visible_width = get_bits_long(gb, 24);
2023 visible_height = get_bits_long(gb, 24);
2024
2025 skip_bits(gb, 8); /* offset x */
2026 skip_bits(gb, 8); /* offset y */
2027 }
2028
2029 skip_bits(gb, 32); /* fps numerator */
2030 skip_bits(gb, 32); /* fps denumerator */
2031 skip_bits(gb, 24); /* aspect numerator */
2032 skip_bits(gb, 24); /* aspect denumerator */
2033
2034 if (s->theora < 0x030200)
2035 skip_bits(gb, 5); /* keyframe frequency force */
2036 colorspace = get_bits(gb, 8);
2037 skip_bits(gb, 24); /* bitrate */
2038
2039 skip_bits(gb, 6); /* quality hint */
2040
2041 if (s->theora >= 0x030200)
2042 {
2043 skip_bits(gb, 5); /* keyframe frequency force */
2044 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2045 skip_bits(gb, 3); /* reserved */
2046 }
2047
2048 // align_get_bits(gb);
2049
2050 if ( visible_width <= s->width && visible_width > s->width-16
2051 && visible_height <= s->height && visible_height > s->height-16)
2052 avcodec_set_dimensions(avctx, visible_width, visible_height);
2053 else
2054 avcodec_set_dimensions(avctx, s->width, s->height);
2055
2056 if (colorspace == 1) {
2057 avctx->color_primaries = AVCOL_PRI_BT470M;
2058 } else if (colorspace == 2) {
2059 avctx->color_primaries = AVCOL_PRI_BT470BG;
2060 }
2061 if (colorspace == 1 || colorspace == 2) {
2062 avctx->colorspace = AVCOL_SPC_BT470BG;
2063 avctx->color_trc = AVCOL_TRC_BT709;
2064 }
2065
2066 return 0;
2067 }
2068
2069 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2070 {
2071 Vp3DecodeContext *s = avctx->priv_data;
2072 int i, n, matrices, inter, plane;
2073
2074 if (s->theora >= 0x030200) {
2075 n = get_bits(gb, 3);
2076 /* loop filter limit values table */
2077 for (i = 0; i < 64; i++) {
2078 s->filter_limit_values[i] = get_bits(gb, n);
2079 if (s->filter_limit_values[i] > 127) {
2080 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2081 s->filter_limit_values[i] = 127;
2082 }
2083 }
2084 }
2085
2086 if (s->theora >= 0x030200)
2087 n = get_bits(gb, 4) + 1;
2088 else
2089 n = 16;
2090 /* quality threshold table */
2091 for (i = 0; i < 64; i++)
2092 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2093
2094 if (s->theora >= 0x030200)
2095 n = get_bits(gb, 4) + 1;
2096 else
2097 n = 16;
2098 /* dc scale factor table */
2099 for (i = 0; i < 64; i++)
2100 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2101
2102 if (s->theora >= 0x030200)
2103 matrices = get_bits(gb, 9) + 1;
2104 else
2105 matrices = 3;
2106
2107 if(matrices > 384){
2108 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2109 return -1;
2110 }
2111
2112 for(n=0; n<matrices; n++){
2113 for (i = 0; i < 64; i++)
2114 s->base_matrix[n][i]= get_bits(gb, 8);
2115 }
2116
2117 for (inter = 0; inter <= 1; inter++) {
2118 for (plane = 0; plane <= 2; plane++) {
2119 int newqr= 1;
2120 if (inter || plane > 0)
2121 newqr = get_bits1(gb);
2122 if (!newqr) {
2123 int qtj, plj;
2124 if(inter && get_bits1(gb)){
2125 qtj = 0;
2126 plj = plane;
2127 }else{
2128 qtj= (3*inter + plane - 1) / 3;
2129 plj= (plane + 2) % 3;
2130 }
2131 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2132 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2133 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2134 } else {
2135 int qri= 0;
2136 int qi = 0;
2137
2138 for(;;){
2139 i= get_bits(gb, av_log2(matrices-1)+1);
2140 if(i>= matrices){
2141 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2142 return -1;
2143 }
2144 s->qr_base[inter][plane][qri]= i;
2145 if(qi >= 63)
2146 break;
2147 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2148 s->qr_size[inter][plane][qri++]= i;
2149 qi += i;
2150 }
2151
2152 if (qi > 63) {
2153 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2154 return -1;
2155 }
2156 s->qr_count[inter][plane]= qri;
2157 }
2158 }
2159 }
2160
2161 /* Huffman tables */
2162 for (s->hti = 0; s->hti < 80; s->hti++) {
2163 s->entries = 0;
2164 s->huff_code_size = 1;
2165 if (!get_bits1(gb)) {
2166 s->hbits = 0;
2167 if(read_huffman_tree(avctx, gb))
2168 return -1;
2169 s->hbits = 1;
2170 if(read_huffman_tree(avctx, gb))
2171 return -1;
2172 }
2173 }
2174
2175 s->theora_tables = 1;
2176
2177 return 0;
2178 }
2179
2180 static av_cold int theora_decode_init(AVCodecContext *avctx)
2181 {
2182 Vp3DecodeContext *s = avctx->priv_data;
2183 GetBitContext gb;
2184 int ptype;
2185 uint8_t *header_start[3];
2186 int header_len[3];
2187 int i;
2188
2189 s->theora = 1;
2190
2191 if (!avctx->extradata_size)
2192 {
2193 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2194 return -1;
2195 }
2196
2197 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2198 42, header_start, header_len) < 0) {
2199 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2200 return -1;
2201 }
2202
2203 for(i=0;i<3;i++) {
2204 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2205
2206 ptype = get_bits(&gb, 8);
2207
2208 if (!(ptype & 0x80))
2209 {
2210 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2211 // return -1;
2212 }
2213
2214 // FIXME: Check for this as well.
2215 skip_bits_long(&gb, 6*8); /* "theora" */
2216
2217 switch(ptype)
2218 {
2219 case 0x80:
2220 theora_decode_header(avctx, &gb);
2221 break;
2222 case 0x81:
2223 // FIXME: is this needed? it breaks sometimes
2224 // theora_decode_comments(avctx, gb);
2225 break;
2226 case 0x82:
2227 if (theora_decode_tables(avctx, &gb))
2228 return -1;
2229 break;
2230 default:
2231 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2232 break;
2233 }
2234 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2235 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2236 if (s->theora < 0x030200)
2237 break;
2238 }
2239
2240 return vp3_decode_init(avctx);
2241 }
2242
2243 AVCodec theora_decoder = {
2244 "theora",
2245 CODEC_TYPE_VIDEO,
2246 CODEC_ID_THEORA,
2247 sizeof(Vp3DecodeContext),
2248 theora_decode_init,
2249 NULL,
2250 vp3_decode_end,
2251 vp3_decode_frame,
2252 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2253 NULL,
2254 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2255 };
2256 #endif
2257
2258 AVCodec vp3_decoder = {
2259 "vp3",
2260 CODEC_TYPE_VIDEO,
2261 CODEC_ID_VP3,
2262 sizeof(Vp3DecodeContext),
2263 vp3_decode_init,
2264 NULL,
2265 vp3_decode_end,
2266 vp3_decode_frame,
2267 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2268 NULL,
2269 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2270 };