lavc: use designated initialisers for all codecs.
[libav.git] / libavcodec / vp3.c
1 /*
2 * Copyright (C) 2003-2004 the ffmpeg project
3 *
4 * This file is part of Libav.
5 *
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 /**
22 * @file
23 * On2 VP3 Video Decoder
24 *
25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26 * For more information about the VP3 coding process, visit:
27 * http://wiki.multimedia.cx/index.php?title=On2_VP3
28 *
29 * Theora decoder by Alex Beregszaszi
30 */
31
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <string.h>
35
36 #include "libavutil/imgutils.h"
37 #include "avcodec.h"
38 #include "dsputil.h"
39 #include "get_bits.h"
40
41 #include "vp3data.h"
42 #include "xiph.h"
43 #include "thread.h"
44
45 #define FRAGMENT_PIXELS 8
46
47 static av_cold int vp3_decode_end(AVCodecContext *avctx);
48
49 //FIXME split things out into their own arrays
50 typedef struct Vp3Fragment {
51 int16_t dc;
52 uint8_t coding_method;
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 int chroma_x_shift, chroma_y_shift;
134 AVFrame golden_frame;
135 AVFrame last_frame;
136 AVFrame current_frame;
137 int keyframe;
138 DSPContext dsp;
139 int flipped_image;
140 int last_slice_end;
141 int skip_loop_filter;
142
143 int qps[3];
144 int nqps;
145 int last_qps[3];
146
147 int superblock_count;
148 int y_superblock_width;
149 int y_superblock_height;
150 int y_superblock_count;
151 int c_superblock_width;
152 int c_superblock_height;
153 int c_superblock_count;
154 int u_superblock_start;
155 int v_superblock_start;
156 unsigned char *superblock_coding;
157
158 int macroblock_count;
159 int macroblock_width;
160 int macroblock_height;
161
162 int fragment_count;
163 int fragment_width[2];
164 int fragment_height[2];
165
166 Vp3Fragment *all_fragments;
167 int fragment_start[3];
168 int data_offset[3];
169
170 int8_t (*motion_val[2])[2];
171
172 ScanTable scantable;
173
174 /* tables */
175 uint16_t coded_dc_scale_factor[64];
176 uint32_t coded_ac_scale_factor[64];
177 uint8_t base_matrix[384][64];
178 uint8_t qr_count[2][3];
179 uint8_t qr_size [2][3][64];
180 uint16_t qr_base[2][3][64];
181
182 /**
183 * This is a list of all tokens in bitstream order. Reordering takes place
184 * by pulling from each level during IDCT. As a consequence, IDCT must be
185 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
186 * otherwise. The 32 different tokens with up to 12 bits of extradata are
187 * collapsed into 3 types, packed as follows:
188 * (from the low to high bits)
189 *
190 * 2 bits: type (0,1,2)
191 * 0: EOB run, 14 bits for run length (12 needed)
192 * 1: zero run, 7 bits for run length
193 * 7 bits for the next coefficient (3 needed)
194 * 2: coefficient, 14 bits (11 needed)
195 *
196 * Coefficients are signed, so are packed in the highest bits for automatic
197 * sign extension.
198 */
199 int16_t *dct_tokens[3][64];
200 int16_t *dct_tokens_base;
201 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
202 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
203 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
204
205 /**
206 * number of blocks that contain DCT coefficients at the given level or higher
207 */
208 int num_coded_frags[3][64];
209 int total_num_coded_frags;
210
211 /* this is a list of indexes into the all_fragments array indicating
212 * which of the fragments are coded */
213 int *coded_fragment_list[3];
214
215 VLC dc_vlc[16];
216 VLC ac_vlc_1[16];
217 VLC ac_vlc_2[16];
218 VLC ac_vlc_3[16];
219 VLC ac_vlc_4[16];
220
221 VLC superblock_run_length_vlc;
222 VLC fragment_run_length_vlc;
223 VLC mode_code_vlc;
224 VLC motion_vector_vlc;
225
226 /* these arrays need to be on 16-byte boundaries since SSE2 operations
227 * index into them */
228 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
229
230 /* This table contains superblock_count * 16 entries. Each set of 16
231 * numbers corresponds to the fragment indexes 0..15 of the superblock.
232 * An entry will be -1 to indicate that no entry corresponds to that
233 * index. */
234 int *superblock_fragments;
235
236 /* This is an array that indicates how a particular macroblock
237 * is coded. */
238 unsigned char *macroblock_coding;
239
240 uint8_t *edge_emu_buffer;
241
242 /* Huffman decode */
243 int hti;
244 unsigned int hbits;
245 int entries;
246 int huff_code_size;
247 uint32_t huffman_table[80][32][2];
248
249 uint8_t filter_limit_values[64];
250 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
251 } Vp3DecodeContext;
252
253 /************************************************************************
254 * VP3 specific functions
255 ************************************************************************/
256
257 /*
258 * This function sets up all of the various blocks mappings:
259 * superblocks <-> fragments, macroblocks <-> fragments,
260 * superblocks <-> macroblocks
261 *
262 * @return 0 is successful; returns 1 if *anything* went wrong.
263 */
264 static int init_block_mapping(Vp3DecodeContext *s)
265 {
266 int sb_x, sb_y, plane;
267 int x, y, i, j = 0;
268
269 for (plane = 0; plane < 3; plane++) {
270 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
271 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
272 int frag_width = s->fragment_width[!!plane];
273 int frag_height = s->fragment_height[!!plane];
274
275 for (sb_y = 0; sb_y < sb_height; sb_y++)
276 for (sb_x = 0; sb_x < sb_width; sb_x++)
277 for (i = 0; i < 16; i++) {
278 x = 4*sb_x + hilbert_offset[i][0];
279 y = 4*sb_y + hilbert_offset[i][1];
280
281 if (x < frag_width && y < frag_height)
282 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
283 else
284 s->superblock_fragments[j++] = -1;
285 }
286 }
287
288 return 0; /* successful path out */
289 }
290
291 /*
292 * This function sets up the dequantization tables used for a particular
293 * frame.
294 */
295 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
296 {
297 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
298 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
299 int i, plane, inter, qri, bmi, bmj, qistart;
300
301 for(inter=0; inter<2; inter++){
302 for(plane=0; plane<3; plane++){
303 int sum=0;
304 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
305 sum+= s->qr_size[inter][plane][qri];
306 if(s->qps[qpi] <= sum)
307 break;
308 }
309 qistart= sum - s->qr_size[inter][plane][qri];
310 bmi= s->qr_base[inter][plane][qri ];
311 bmj= s->qr_base[inter][plane][qri+1];
312 for(i=0; i<64; i++){
313 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
314 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
315 + s->qr_size[inter][plane][qri])
316 / (2*s->qr_size[inter][plane][qri]);
317
318 int qmin= 8<<(inter + !i);
319 int qscale= i ? ac_scale_factor : dc_scale_factor;
320
321 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
322 }
323 // all DC coefficients use the same quant so as not to interfere with DC prediction
324 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
325 }
326 }
327 }
328
329 /*
330 * This function initializes the loop filter boundary limits if the frame's
331 * quality index is different from the previous frame's.
332 *
333 * The filter_limit_values may not be larger than 127.
334 */
335 static void init_loop_filter(Vp3DecodeContext *s)
336 {
337 int *bounding_values= s->bounding_values_array+127;
338 int filter_limit;
339 int x;
340 int value;
341
342 filter_limit = s->filter_limit_values[s->qps[0]];
343
344 /* set up the bounding values */
345 memset(s->bounding_values_array, 0, 256 * sizeof(int));
346 for (x = 0; x < filter_limit; x++) {
347 bounding_values[-x] = -x;
348 bounding_values[x] = x;
349 }
350 for (x = value = filter_limit; x < 128 && value; x++, value--) {
351 bounding_values[ x] = value;
352 bounding_values[-x] = -value;
353 }
354 if (value)
355 bounding_values[128] = value;
356 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
357 }
358
359 /*
360 * This function unpacks all of the superblock/macroblock/fragment coding
361 * information from the bitstream.
362 */
363 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
364 {
365 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
366 int bit = 0;
367 int current_superblock = 0;
368 int current_run = 0;
369 int num_partial_superblocks = 0;
370
371 int i, j;
372 int current_fragment;
373 int plane;
374
375 if (s->keyframe) {
376 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
377
378 } else {
379
380 /* unpack the list of partially-coded superblocks */
381 bit = get_bits1(gb) ^ 1;
382 current_run = 0;
383
384 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
385 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
386 bit = get_bits1(gb);
387 else
388 bit ^= 1;
389
390 current_run = get_vlc2(gb,
391 s->superblock_run_length_vlc.table, 6, 2) + 1;
392 if (current_run == 34)
393 current_run += get_bits(gb, 12);
394
395 if (current_superblock + current_run > s->superblock_count) {
396 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
397 return -1;
398 }
399
400 memset(s->superblock_coding + current_superblock, bit, current_run);
401
402 current_superblock += current_run;
403 if (bit)
404 num_partial_superblocks += current_run;
405 }
406
407 /* unpack the list of fully coded superblocks if any of the blocks were
408 * not marked as partially coded in the previous step */
409 if (num_partial_superblocks < s->superblock_count) {
410 int superblocks_decoded = 0;
411
412 current_superblock = 0;
413 bit = get_bits1(gb) ^ 1;
414 current_run = 0;
415
416 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
417 && get_bits_left(gb) > 0) {
418
419 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
420 bit = get_bits1(gb);
421 else
422 bit ^= 1;
423
424 current_run = get_vlc2(gb,
425 s->superblock_run_length_vlc.table, 6, 2) + 1;
426 if (current_run == 34)
427 current_run += get_bits(gb, 12);
428
429 for (j = 0; j < current_run; current_superblock++) {
430 if (current_superblock >= s->superblock_count) {
431 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
432 return -1;
433 }
434
435 /* skip any superblocks already marked as partially coded */
436 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
437 s->superblock_coding[current_superblock] = 2*bit;
438 j++;
439 }
440 }
441 superblocks_decoded += current_run;
442 }
443 }
444
445 /* if there were partial blocks, initialize bitstream for
446 * unpacking fragment codings */
447 if (num_partial_superblocks) {
448
449 current_run = 0;
450 bit = get_bits1(gb);
451 /* toggle the bit because as soon as the first run length is
452 * fetched the bit will be toggled again */
453 bit ^= 1;
454 }
455 }
456
457 /* figure out which fragments are coded; iterate through each
458 * superblock (all planes) */
459 s->total_num_coded_frags = 0;
460 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
461
462 for (plane = 0; plane < 3; plane++) {
463 int sb_start = superblock_starts[plane];
464 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
465 int num_coded_frags = 0;
466
467 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
468
469 /* iterate through all 16 fragments in a superblock */
470 for (j = 0; j < 16; j++) {
471
472 /* if the fragment is in bounds, check its coding status */
473 current_fragment = s->superblock_fragments[i * 16 + j];
474 if (current_fragment != -1) {
475 int coded = s->superblock_coding[i];
476
477 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
478
479 /* fragment may or may not be coded; this is the case
480 * that cares about the fragment coding runs */
481 if (current_run-- == 0) {
482 bit ^= 1;
483 current_run = get_vlc2(gb,
484 s->fragment_run_length_vlc.table, 5, 2);
485 }
486 coded = bit;
487 }
488
489 if (coded) {
490 /* default mode; actual mode will be decoded in
491 * the next phase */
492 s->all_fragments[current_fragment].coding_method =
493 MODE_INTER_NO_MV;
494 s->coded_fragment_list[plane][num_coded_frags++] =
495 current_fragment;
496 } else {
497 /* not coded; copy this fragment from the prior frame */
498 s->all_fragments[current_fragment].coding_method =
499 MODE_COPY;
500 }
501 }
502 }
503 }
504 s->total_num_coded_frags += num_coded_frags;
505 for (i = 0; i < 64; i++)
506 s->num_coded_frags[plane][i] = num_coded_frags;
507 if (plane < 2)
508 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
509 }
510 return 0;
511 }
512
513 /*
514 * This function unpacks all the coding mode data for individual macroblocks
515 * from the bitstream.
516 */
517 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
518 {
519 int i, j, k, sb_x, sb_y;
520 int scheme;
521 int current_macroblock;
522 int current_fragment;
523 int coding_mode;
524 int custom_mode_alphabet[CODING_MODE_COUNT];
525 const int *alphabet;
526 Vp3Fragment *frag;
527
528 if (s->keyframe) {
529 for (i = 0; i < s->fragment_count; i++)
530 s->all_fragments[i].coding_method = MODE_INTRA;
531
532 } else {
533
534 /* fetch the mode coding scheme for this frame */
535 scheme = get_bits(gb, 3);
536
537 /* is it a custom coding scheme? */
538 if (scheme == 0) {
539 for (i = 0; i < 8; i++)
540 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
541 for (i = 0; i < 8; i++)
542 custom_mode_alphabet[get_bits(gb, 3)] = i;
543 alphabet = custom_mode_alphabet;
544 } else
545 alphabet = ModeAlphabet[scheme-1];
546
547 /* iterate through all of the macroblocks that contain 1 or more
548 * coded fragments */
549 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
550 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
551 if (get_bits_left(gb) <= 0)
552 return -1;
553
554 for (j = 0; j < 4; j++) {
555 int mb_x = 2*sb_x + (j>>1);
556 int mb_y = 2*sb_y + (((j>>1)+j)&1);
557 current_macroblock = mb_y * s->macroblock_width + mb_x;
558
559 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
560 continue;
561
562 #define BLOCK_X (2*mb_x + (k&1))
563 #define BLOCK_Y (2*mb_y + (k>>1))
564 /* coding modes are only stored if the macroblock has at least one
565 * luma block coded, otherwise it must be INTER_NO_MV */
566 for (k = 0; k < 4; k++) {
567 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
568 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
569 break;
570 }
571 if (k == 4) {
572 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
573 continue;
574 }
575
576 /* mode 7 means get 3 bits for each coding mode */
577 if (scheme == 7)
578 coding_mode = get_bits(gb, 3);
579 else
580 coding_mode = alphabet
581 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
582
583 s->macroblock_coding[current_macroblock] = coding_mode;
584 for (k = 0; k < 4; k++) {
585 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
586 if (frag->coding_method != MODE_COPY)
587 frag->coding_method = coding_mode;
588 }
589
590 #define SET_CHROMA_MODES \
591 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
592 frag[s->fragment_start[1]].coding_method = coding_mode;\
593 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
594 frag[s->fragment_start[2]].coding_method = coding_mode;
595
596 if (s->chroma_y_shift) {
597 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
598 SET_CHROMA_MODES
599 } else if (s->chroma_x_shift) {
600 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
601 for (k = 0; k < 2; k++) {
602 SET_CHROMA_MODES
603 frag += s->fragment_width[1];
604 }
605 } else {
606 for (k = 0; k < 4; k++) {
607 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
608 SET_CHROMA_MODES
609 }
610 }
611 }
612 }
613 }
614 }
615
616 return 0;
617 }
618
619 /*
620 * This function unpacks all the motion vectors for the individual
621 * macroblocks from the bitstream.
622 */
623 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
624 {
625 int j, k, sb_x, sb_y;
626 int coding_mode;
627 int motion_x[4];
628 int motion_y[4];
629 int last_motion_x = 0;
630 int last_motion_y = 0;
631 int prior_last_motion_x = 0;
632 int prior_last_motion_y = 0;
633 int current_macroblock;
634 int current_fragment;
635 int frag;
636
637 if (s->keyframe)
638 return 0;
639
640 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
641 coding_mode = get_bits1(gb);
642
643 /* iterate through all of the macroblocks that contain 1 or more
644 * coded fragments */
645 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
646 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
647 if (get_bits_left(gb) <= 0)
648 return -1;
649
650 for (j = 0; j < 4; j++) {
651 int mb_x = 2*sb_x + (j>>1);
652 int mb_y = 2*sb_y + (((j>>1)+j)&1);
653 current_macroblock = mb_y * s->macroblock_width + mb_x;
654
655 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
656 (s->macroblock_coding[current_macroblock] == MODE_COPY))
657 continue;
658
659 switch (s->macroblock_coding[current_macroblock]) {
660
661 case MODE_INTER_PLUS_MV:
662 case MODE_GOLDEN_MV:
663 /* all 6 fragments use the same motion vector */
664 if (coding_mode == 0) {
665 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
666 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
667 } else {
668 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
669 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
670 }
671
672 /* vector maintenance, only on MODE_INTER_PLUS_MV */
673 if (s->macroblock_coding[current_macroblock] ==
674 MODE_INTER_PLUS_MV) {
675 prior_last_motion_x = last_motion_x;
676 prior_last_motion_y = last_motion_y;
677 last_motion_x = motion_x[0];
678 last_motion_y = motion_y[0];
679 }
680 break;
681
682 case MODE_INTER_FOURMV:
683 /* vector maintenance */
684 prior_last_motion_x = last_motion_x;
685 prior_last_motion_y = last_motion_y;
686
687 /* fetch 4 vectors from the bitstream, one for each
688 * Y fragment, then average for the C fragment vectors */
689 for (k = 0; k < 4; k++) {
690 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
691 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
692 if (coding_mode == 0) {
693 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
694 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
695 } else {
696 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
697 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
698 }
699 last_motion_x = motion_x[k];
700 last_motion_y = motion_y[k];
701 } else {
702 motion_x[k] = 0;
703 motion_y[k] = 0;
704 }
705 }
706 break;
707
708 case MODE_INTER_LAST_MV:
709 /* all 6 fragments use the last motion vector */
710 motion_x[0] = last_motion_x;
711 motion_y[0] = last_motion_y;
712
713 /* no vector maintenance (last vector remains the
714 * last vector) */
715 break;
716
717 case MODE_INTER_PRIOR_LAST:
718 /* all 6 fragments use the motion vector prior to the
719 * last motion vector */
720 motion_x[0] = prior_last_motion_x;
721 motion_y[0] = prior_last_motion_y;
722
723 /* vector maintenance */
724 prior_last_motion_x = last_motion_x;
725 prior_last_motion_y = last_motion_y;
726 last_motion_x = motion_x[0];
727 last_motion_y = motion_y[0];
728 break;
729
730 default:
731 /* covers intra, inter without MV, golden without MV */
732 motion_x[0] = 0;
733 motion_y[0] = 0;
734
735 /* no vector maintenance */
736 break;
737 }
738
739 /* assign the motion vectors to the correct fragments */
740 for (k = 0; k < 4; k++) {
741 current_fragment =
742 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
743 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
744 s->motion_val[0][current_fragment][0] = motion_x[k];
745 s->motion_val[0][current_fragment][1] = motion_y[k];
746 } else {
747 s->motion_val[0][current_fragment][0] = motion_x[0];
748 s->motion_val[0][current_fragment][1] = motion_y[0];
749 }
750 }
751
752 if (s->chroma_y_shift) {
753 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
754 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
755 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
756 }
757 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
758 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
759 frag = mb_y*s->fragment_width[1] + mb_x;
760 s->motion_val[1][frag][0] = motion_x[0];
761 s->motion_val[1][frag][1] = motion_y[0];
762 } else if (s->chroma_x_shift) {
763 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
764 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
765 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
766 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
767 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
768 } else {
769 motion_x[1] = motion_x[0];
770 motion_y[1] = motion_y[0];
771 }
772 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
773 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
774
775 frag = 2*mb_y*s->fragment_width[1] + mb_x;
776 for (k = 0; k < 2; k++) {
777 s->motion_val[1][frag][0] = motion_x[k];
778 s->motion_val[1][frag][1] = motion_y[k];
779 frag += s->fragment_width[1];
780 }
781 } else {
782 for (k = 0; k < 4; k++) {
783 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
784 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
785 s->motion_val[1][frag][0] = motion_x[k];
786 s->motion_val[1][frag][1] = motion_y[k];
787 } else {
788 s->motion_val[1][frag][0] = motion_x[0];
789 s->motion_val[1][frag][1] = motion_y[0];
790 }
791 }
792 }
793 }
794 }
795 }
796
797 return 0;
798 }
799
800 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
801 {
802 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
803 int num_blocks = s->total_num_coded_frags;
804
805 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
806 i = blocks_decoded = num_blocks_at_qpi = 0;
807
808 bit = get_bits1(gb) ^ 1;
809 run_length = 0;
810
811 do {
812 if (run_length == MAXIMUM_LONG_BIT_RUN)
813 bit = get_bits1(gb);
814 else
815 bit ^= 1;
816
817 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
818 if (run_length == 34)
819 run_length += get_bits(gb, 12);
820 blocks_decoded += run_length;
821
822 if (!bit)
823 num_blocks_at_qpi += run_length;
824
825 for (j = 0; j < run_length; i++) {
826 if (i >= s->total_num_coded_frags)
827 return -1;
828
829 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
830 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
831 j++;
832 }
833 }
834 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
835
836 num_blocks -= num_blocks_at_qpi;
837 }
838
839 return 0;
840 }
841
842 /*
843 * This function is called by unpack_dct_coeffs() to extract the VLCs from
844 * the bitstream. The VLCs encode tokens which are used to unpack DCT
845 * data. This function unpacks all the VLCs for either the Y plane or both
846 * C planes, and is called for DC coefficients or different AC coefficient
847 * levels (since different coefficient types require different VLC tables.
848 *
849 * This function returns a residual eob run. E.g, if a particular token gave
850 * instructions to EOB the next 5 fragments and there were only 2 fragments
851 * left in the current fragment range, 3 would be returned so that it could
852 * be passed into the next call to this same function.
853 */
854 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
855 VLC *table, int coeff_index,
856 int plane,
857 int eob_run)
858 {
859 int i, j = 0;
860 int token;
861 int zero_run = 0;
862 DCTELEM coeff = 0;
863 int bits_to_get;
864 int blocks_ended;
865 int coeff_i = 0;
866 int num_coeffs = s->num_coded_frags[plane][coeff_index];
867 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
868
869 /* local references to structure members to avoid repeated deferences */
870 int *coded_fragment_list = s->coded_fragment_list[plane];
871 Vp3Fragment *all_fragments = s->all_fragments;
872 VLC_TYPE (*vlc_table)[2] = table->table;
873
874 if (num_coeffs < 0)
875 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
876
877 if (eob_run > num_coeffs) {
878 coeff_i = blocks_ended = num_coeffs;
879 eob_run -= num_coeffs;
880 } else {
881 coeff_i = blocks_ended = eob_run;
882 eob_run = 0;
883 }
884
885 // insert fake EOB token to cover the split between planes or zzi
886 if (blocks_ended)
887 dct_tokens[j++] = blocks_ended << 2;
888
889 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
890 /* decode a VLC into a token */
891 token = get_vlc2(gb, vlc_table, 11, 3);
892 /* use the token to get a zero run, a coefficient, and an eob run */
893 if (token <= 6) {
894 eob_run = eob_run_base[token];
895 if (eob_run_get_bits[token])
896 eob_run += get_bits(gb, eob_run_get_bits[token]);
897
898 // record only the number of blocks ended in this plane,
899 // any spill will be recorded in the next plane.
900 if (eob_run > num_coeffs - coeff_i) {
901 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
902 blocks_ended += num_coeffs - coeff_i;
903 eob_run -= num_coeffs - coeff_i;
904 coeff_i = num_coeffs;
905 } else {
906 dct_tokens[j++] = TOKEN_EOB(eob_run);
907 blocks_ended += eob_run;
908 coeff_i += eob_run;
909 eob_run = 0;
910 }
911 } else {
912 bits_to_get = coeff_get_bits[token];
913 if (bits_to_get)
914 bits_to_get = get_bits(gb, bits_to_get);
915 coeff = coeff_tables[token][bits_to_get];
916
917 zero_run = zero_run_base[token];
918 if (zero_run_get_bits[token])
919 zero_run += get_bits(gb, zero_run_get_bits[token]);
920
921 if (zero_run) {
922 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
923 } else {
924 // Save DC into the fragment structure. DC prediction is
925 // done in raster order, so the actual DC can't be in with
926 // other tokens. We still need the token in dct_tokens[]
927 // however, or else the structure collapses on itself.
928 if (!coeff_index)
929 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
930
931 dct_tokens[j++] = TOKEN_COEFF(coeff);
932 }
933
934 if (coeff_index + zero_run > 64) {
935 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
936 " %d coeffs left\n", zero_run, 64-coeff_index);
937 zero_run = 64 - coeff_index;
938 }
939
940 // zero runs code multiple coefficients,
941 // so don't try to decode coeffs for those higher levels
942 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
943 s->num_coded_frags[plane][i]--;
944 coeff_i++;
945 }
946 }
947
948 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
949 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
950
951 // decrement the number of blocks that have higher coeffecients for each
952 // EOB run at this level
953 if (blocks_ended)
954 for (i = coeff_index+1; i < 64; i++)
955 s->num_coded_frags[plane][i] -= blocks_ended;
956
957 // setup the next buffer
958 if (plane < 2)
959 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
960 else if (coeff_index < 63)
961 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
962
963 return eob_run;
964 }
965
966 static void reverse_dc_prediction(Vp3DecodeContext *s,
967 int first_fragment,
968 int fragment_width,
969 int fragment_height);
970 /*
971 * This function unpacks all of the DCT coefficient data from the
972 * bitstream.
973 */
974 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
975 {
976 int i;
977 int dc_y_table;
978 int dc_c_table;
979 int ac_y_table;
980 int ac_c_table;
981 int residual_eob_run = 0;
982 VLC *y_tables[64];
983 VLC *c_tables[64];
984
985 s->dct_tokens[0][0] = s->dct_tokens_base;
986
987 /* fetch the DC table indexes */
988 dc_y_table = get_bits(gb, 4);
989 dc_c_table = get_bits(gb, 4);
990
991 /* unpack the Y plane DC coefficients */
992 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
993 0, residual_eob_run);
994
995 /* reverse prediction of the Y-plane DC coefficients */
996 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
997
998 /* unpack the C plane DC coefficients */
999 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1000 1, residual_eob_run);
1001 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1002 2, residual_eob_run);
1003
1004 /* reverse prediction of the C-plane DC coefficients */
1005 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1006 {
1007 reverse_dc_prediction(s, s->fragment_start[1],
1008 s->fragment_width[1], s->fragment_height[1]);
1009 reverse_dc_prediction(s, s->fragment_start[2],
1010 s->fragment_width[1], s->fragment_height[1]);
1011 }
1012
1013 /* fetch the AC table indexes */
1014 ac_y_table = get_bits(gb, 4);
1015 ac_c_table = get_bits(gb, 4);
1016
1017 /* build tables of AC VLC tables */
1018 for (i = 1; i <= 5; i++) {
1019 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1020 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1021 }
1022 for (i = 6; i <= 14; i++) {
1023 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1024 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1025 }
1026 for (i = 15; i <= 27; i++) {
1027 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1028 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1029 }
1030 for (i = 28; i <= 63; i++) {
1031 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1032 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1033 }
1034
1035 /* decode all AC coefficents */
1036 for (i = 1; i <= 63; i++) {
1037 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1038 0, residual_eob_run);
1039
1040 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1041 1, residual_eob_run);
1042 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1043 2, residual_eob_run);
1044 }
1045
1046 return 0;
1047 }
1048
1049 /*
1050 * This function reverses the DC prediction for each coded fragment in
1051 * the frame. Much of this function is adapted directly from the original
1052 * VP3 source code.
1053 */
1054 #define COMPATIBLE_FRAME(x) \
1055 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1056 #define DC_COEFF(u) s->all_fragments[u].dc
1057
1058 static void reverse_dc_prediction(Vp3DecodeContext *s,
1059 int first_fragment,
1060 int fragment_width,
1061 int fragment_height)
1062 {
1063
1064 #define PUL 8
1065 #define PU 4
1066 #define PUR 2
1067 #define PL 1
1068
1069 int x, y;
1070 int i = first_fragment;
1071
1072 int predicted_dc;
1073
1074 /* DC values for the left, up-left, up, and up-right fragments */
1075 int vl, vul, vu, vur;
1076
1077 /* indexes for the left, up-left, up, and up-right fragments */
1078 int l, ul, u, ur;
1079
1080 /*
1081 * The 6 fields mean:
1082 * 0: up-left multiplier
1083 * 1: up multiplier
1084 * 2: up-right multiplier
1085 * 3: left multiplier
1086 */
1087 static const int predictor_transform[16][4] = {
1088 { 0, 0, 0, 0},
1089 { 0, 0, 0,128}, // PL
1090 { 0, 0,128, 0}, // PUR
1091 { 0, 0, 53, 75}, // PUR|PL
1092 { 0,128, 0, 0}, // PU
1093 { 0, 64, 0, 64}, // PU|PL
1094 { 0,128, 0, 0}, // PU|PUR
1095 { 0, 0, 53, 75}, // PU|PUR|PL
1096 {128, 0, 0, 0}, // PUL
1097 { 0, 0, 0,128}, // PUL|PL
1098 { 64, 0, 64, 0}, // PUL|PUR
1099 { 0, 0, 53, 75}, // PUL|PUR|PL
1100 { 0,128, 0, 0}, // PUL|PU
1101 {-104,116, 0,116}, // PUL|PU|PL
1102 { 24, 80, 24, 0}, // PUL|PU|PUR
1103 {-104,116, 0,116} // PUL|PU|PUR|PL
1104 };
1105
1106 /* This table shows which types of blocks can use other blocks for
1107 * prediction. For example, INTRA is the only mode in this table to
1108 * have a frame number of 0. That means INTRA blocks can only predict
1109 * from other INTRA blocks. There are 2 golden frame coding types;
1110 * blocks encoding in these modes can only predict from other blocks
1111 * that were encoded with these 1 of these 2 modes. */
1112 static const unsigned char compatible_frame[9] = {
1113 1, /* MODE_INTER_NO_MV */
1114 0, /* MODE_INTRA */
1115 1, /* MODE_INTER_PLUS_MV */
1116 1, /* MODE_INTER_LAST_MV */
1117 1, /* MODE_INTER_PRIOR_MV */
1118 2, /* MODE_USING_GOLDEN */
1119 2, /* MODE_GOLDEN_MV */
1120 1, /* MODE_INTER_FOUR_MV */
1121 3 /* MODE_COPY */
1122 };
1123 int current_frame_type;
1124
1125 /* there is a last DC predictor for each of the 3 frame types */
1126 short last_dc[3];
1127
1128 int transform = 0;
1129
1130 vul = vu = vur = vl = 0;
1131 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1132
1133 /* for each fragment row... */
1134 for (y = 0; y < fragment_height; y++) {
1135
1136 /* for each fragment in a row... */
1137 for (x = 0; x < fragment_width; x++, i++) {
1138
1139 /* reverse prediction if this block was coded */
1140 if (s->all_fragments[i].coding_method != MODE_COPY) {
1141
1142 current_frame_type =
1143 compatible_frame[s->all_fragments[i].coding_method];
1144
1145 transform= 0;
1146 if(x){
1147 l= i-1;
1148 vl = DC_COEFF(l);
1149 if(COMPATIBLE_FRAME(l))
1150 transform |= PL;
1151 }
1152 if(y){
1153 u= i-fragment_width;
1154 vu = DC_COEFF(u);
1155 if(COMPATIBLE_FRAME(u))
1156 transform |= PU;
1157 if(x){
1158 ul= i-fragment_width-1;
1159 vul = DC_COEFF(ul);
1160 if(COMPATIBLE_FRAME(ul))
1161 transform |= PUL;
1162 }
1163 if(x + 1 < fragment_width){
1164 ur= i-fragment_width+1;
1165 vur = DC_COEFF(ur);
1166 if(COMPATIBLE_FRAME(ur))
1167 transform |= PUR;
1168 }
1169 }
1170
1171 if (transform == 0) {
1172
1173 /* if there were no fragments to predict from, use last
1174 * DC saved */
1175 predicted_dc = last_dc[current_frame_type];
1176 } else {
1177
1178 /* apply the appropriate predictor transform */
1179 predicted_dc =
1180 (predictor_transform[transform][0] * vul) +
1181 (predictor_transform[transform][1] * vu) +
1182 (predictor_transform[transform][2] * vur) +
1183 (predictor_transform[transform][3] * vl);
1184
1185 predicted_dc /= 128;
1186
1187 /* check for outranging on the [ul u l] and
1188 * [ul u ur l] predictors */
1189 if ((transform == 15) || (transform == 13)) {
1190 if (FFABS(predicted_dc - vu) > 128)
1191 predicted_dc = vu;
1192 else if (FFABS(predicted_dc - vl) > 128)
1193 predicted_dc = vl;
1194 else if (FFABS(predicted_dc - vul) > 128)
1195 predicted_dc = vul;
1196 }
1197 }
1198
1199 /* at long last, apply the predictor */
1200 DC_COEFF(i) += predicted_dc;
1201 /* save the DC */
1202 last_dc[current_frame_type] = DC_COEFF(i);
1203 }
1204 }
1205 }
1206 }
1207
1208 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1209 {
1210 int x, y;
1211 int *bounding_values= s->bounding_values_array+127;
1212
1213 int width = s->fragment_width[!!plane];
1214 int height = s->fragment_height[!!plane];
1215 int fragment = s->fragment_start [plane] + ystart * width;
1216 int stride = s->current_frame.linesize[plane];
1217 uint8_t *plane_data = s->current_frame.data [plane];
1218 if (!s->flipped_image) stride = -stride;
1219 plane_data += s->data_offset[plane] + 8*ystart*stride;
1220
1221 for (y = ystart; y < yend; y++) {
1222
1223 for (x = 0; x < width; x++) {
1224 /* This code basically just deblocks on the edges of coded blocks.
1225 * However, it has to be much more complicated because of the
1226 * braindamaged deblock ordering used in VP3/Theora. Order matters
1227 * because some pixels get filtered twice. */
1228 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1229 {
1230 /* do not perform left edge filter for left columns frags */
1231 if (x > 0) {
1232 s->dsp.vp3_h_loop_filter(
1233 plane_data + 8*x,
1234 stride, bounding_values);
1235 }
1236
1237 /* do not perform top edge filter for top row fragments */
1238 if (y > 0) {
1239 s->dsp.vp3_v_loop_filter(
1240 plane_data + 8*x,
1241 stride, bounding_values);
1242 }
1243
1244 /* do not perform right edge filter for right column
1245 * fragments or if right fragment neighbor is also coded
1246 * in this frame (it will be filtered in next iteration) */
1247 if ((x < width - 1) &&
1248 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1249 s->dsp.vp3_h_loop_filter(
1250 plane_data + 8*x + 8,
1251 stride, bounding_values);
1252 }
1253
1254 /* do not perform bottom edge filter for bottom row
1255 * fragments or if bottom fragment neighbor is also coded
1256 * in this frame (it will be filtered in the next row) */
1257 if ((y < height - 1) &&
1258 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1259 s->dsp.vp3_v_loop_filter(
1260 plane_data + 8*x + 8*stride,
1261 stride, bounding_values);
1262 }
1263 }
1264
1265 fragment++;
1266 }
1267 plane_data += 8*stride;
1268 }
1269 }
1270
1271 /**
1272 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1273 * for the next block in coding order
1274 */
1275 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1276 int plane, int inter, DCTELEM block[64])
1277 {
1278 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1279 uint8_t *perm = s->scantable.permutated;
1280 int i = 0;
1281
1282 do {
1283 int token = *s->dct_tokens[plane][i];
1284 switch (token & 3) {
1285 case 0: // EOB
1286 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1287 s->dct_tokens[plane][i]++;
1288 else
1289 *s->dct_tokens[plane][i] = token & ~3;
1290 goto end;
1291 case 1: // zero run
1292 s->dct_tokens[plane][i]++;
1293 i += (token >> 2) & 0x7f;
1294 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1295 i++;
1296 break;
1297 case 2: // coeff
1298 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1299 s->dct_tokens[plane][i++]++;
1300 break;
1301 default: // shouldn't happen
1302 return i;
1303 }
1304 } while (i < 64);
1305 end:
1306 // the actual DC+prediction is in the fragment structure
1307 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1308 return i;
1309 }
1310
1311 /**
1312 * called when all pixels up to row y are complete
1313 */
1314 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1315 {
1316 int h, cy;
1317 int offset[4];
1318
1319 if (HAVE_PTHREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1320 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1321
1322 // At the end of the frame, report INT_MAX instead of the height of the frame.
1323 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1324 // they don't have to clip their values.
1325 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1326 }
1327
1328 if(s->avctx->draw_horiz_band==NULL)
1329 return;
1330
1331 h= y - s->last_slice_end;
1332 s->last_slice_end= y;
1333 y -= h;
1334
1335 if (!s->flipped_image) {
1336 y = s->avctx->height - y - h;
1337 }
1338
1339 cy = y >> s->chroma_y_shift;
1340 offset[0] = s->current_frame.linesize[0]*y;
1341 offset[1] = s->current_frame.linesize[1]*cy;
1342 offset[2] = s->current_frame.linesize[2]*cy;
1343 offset[3] = 0;
1344
1345 emms_c();
1346 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1347 }
1348
1349 /**
1350 * Wait for the reference frame of the current fragment.
1351 * The progress value is in luma pixel rows.
1352 */
1353 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1354 {
1355 AVFrame *ref_frame;
1356 int ref_row;
1357 int border = motion_y&1;
1358
1359 if (fragment->coding_method == MODE_USING_GOLDEN ||
1360 fragment->coding_method == MODE_GOLDEN_MV)
1361 ref_frame = &s->golden_frame;
1362 else
1363 ref_frame = &s->last_frame;
1364
1365 ref_row = y + (motion_y>>1);
1366 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1367
1368 ff_thread_await_progress(ref_frame, ref_row, 0);
1369 }
1370
1371 /*
1372 * Perform the final rendering for a particular slice of data.
1373 * The slice number ranges from 0..(c_superblock_height - 1).
1374 */
1375 static void render_slice(Vp3DecodeContext *s, int slice)
1376 {
1377 int x, y, i, j, fragment;
1378 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1379 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1380 int motion_halfpel_index;
1381 uint8_t *motion_source;
1382 int plane, first_pixel;
1383
1384 if (slice >= s->c_superblock_height)
1385 return;
1386
1387 for (plane = 0; plane < 3; plane++) {
1388 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1389 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1390 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1391 int stride = s->current_frame.linesize[plane];
1392 int plane_width = s->width >> (plane && s->chroma_x_shift);
1393 int plane_height = s->height >> (plane && s->chroma_y_shift);
1394 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1395
1396 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1397 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1398 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1399
1400 int fragment_width = s->fragment_width[!!plane];
1401 int fragment_height = s->fragment_height[!!plane];
1402 int fragment_start = s->fragment_start[plane];
1403 int do_await = !plane && HAVE_PTHREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1404
1405 if (!s->flipped_image) stride = -stride;
1406 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1407 continue;
1408
1409 /* for each superblock row in the slice (both of them)... */
1410 for (; sb_y < slice_height; sb_y++) {
1411
1412 /* for each superblock in a row... */
1413 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1414
1415 /* for each block in a superblock... */
1416 for (j = 0; j < 16; j++) {
1417 x = 4*sb_x + hilbert_offset[j][0];
1418 y = 4*sb_y + hilbert_offset[j][1];
1419 fragment = y*fragment_width + x;
1420
1421 i = fragment_start + fragment;
1422
1423 // bounds check
1424 if (x >= fragment_width || y >= fragment_height)
1425 continue;
1426
1427 first_pixel = 8*y*stride + 8*x;
1428
1429 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1430 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1431
1432 /* transform if this block was coded */
1433 if (s->all_fragments[i].coding_method != MODE_COPY) {
1434 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1435 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1436 motion_source= golden_plane;
1437 else
1438 motion_source= last_plane;
1439
1440 motion_source += first_pixel;
1441 motion_halfpel_index = 0;
1442
1443 /* sort out the motion vector if this fragment is coded
1444 * using a motion vector method */
1445 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1446 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1447 int src_x, src_y;
1448 motion_x = motion_val[fragment][0];
1449 motion_y = motion_val[fragment][1];
1450
1451 src_x= (motion_x>>1) + 8*x;
1452 src_y= (motion_y>>1) + 8*y;
1453
1454 motion_halfpel_index = motion_x & 0x01;
1455 motion_source += (motion_x >> 1);
1456
1457 motion_halfpel_index |= (motion_y & 0x01) << 1;
1458 motion_source += ((motion_y >> 1) * stride);
1459
1460 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1461 uint8_t *temp= s->edge_emu_buffer;
1462 if(stride<0) temp -= 8*stride;
1463
1464 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1465 motion_source= temp;
1466 }
1467 }
1468
1469
1470 /* first, take care of copying a block from either the
1471 * previous or the golden frame */
1472 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1473 /* Note, it is possible to implement all MC cases with
1474 put_no_rnd_pixels_l2 which would look more like the
1475 VP3 source but this would be slower as
1476 put_no_rnd_pixels_tab is better optimzed */
1477 if(motion_halfpel_index != 3){
1478 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1479 output_plane + first_pixel,
1480 motion_source, stride, 8);
1481 }else{
1482 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1483 s->dsp.put_no_rnd_pixels_l2[1](
1484 output_plane + first_pixel,
1485 motion_source - d,
1486 motion_source + stride + 1 + d,
1487 stride, 8);
1488 }
1489 }
1490
1491 s->dsp.clear_block(block);
1492
1493 /* invert DCT and place (or add) in final output */
1494
1495 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1496 vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1497 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1498 block[0] += 128<<3;
1499 s->dsp.idct_put(
1500 output_plane + first_pixel,
1501 stride,
1502 block);
1503 } else {
1504 if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
1505 s->dsp.idct_add(
1506 output_plane + first_pixel,
1507 stride,
1508 block);
1509 } else {
1510 s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1511 }
1512 }
1513 } else {
1514
1515 /* copy directly from the previous frame */
1516 s->dsp.put_pixels_tab[1][0](
1517 output_plane + first_pixel,
1518 last_plane + first_pixel,
1519 stride, 8);
1520
1521 }
1522 }
1523 }
1524
1525 // Filter up to the last row in the superblock row
1526 if (!s->skip_loop_filter)
1527 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1528 }
1529 }
1530
1531 /* this looks like a good place for slice dispatch... */
1532 /* algorithm:
1533 * if (slice == s->macroblock_height - 1)
1534 * dispatch (both last slice & 2nd-to-last slice);
1535 * else if (slice > 0)
1536 * dispatch (slice - 1);
1537 */
1538
1539 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1540 }
1541
1542 /// Allocate tables for per-frame data in Vp3DecodeContext
1543 static av_cold int allocate_tables(AVCodecContext *avctx)
1544 {
1545 Vp3DecodeContext *s = avctx->priv_data;
1546 int y_fragment_count, c_fragment_count;
1547
1548 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1549 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1550
1551 s->superblock_coding = av_malloc(s->superblock_count);
1552 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1553 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1554 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1555 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1556 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1557
1558 /* work out the block mapping tables */
1559 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1560 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1561
1562 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1563 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1564 !s->motion_val[0] || !s->motion_val[1]) {
1565 vp3_decode_end(avctx);
1566 return -1;
1567 }
1568
1569 init_block_mapping(s);
1570
1571 return 0;
1572 }
1573
1574 /*
1575 * This is the ffmpeg/libavcodec API init function.
1576 */
1577 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1578 {
1579 Vp3DecodeContext *s = avctx->priv_data;
1580 int i, inter, plane;
1581 int c_width;
1582 int c_height;
1583 int y_fragment_count, c_fragment_count;
1584
1585 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1586 s->version = 0;
1587 else
1588 s->version = 1;
1589
1590 s->avctx = avctx;
1591 s->width = FFALIGN(avctx->width, 16);
1592 s->height = FFALIGN(avctx->height, 16);
1593 if (avctx->pix_fmt == PIX_FMT_NONE)
1594 avctx->pix_fmt = PIX_FMT_YUV420P;
1595 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1596 if(avctx->idct_algo==FF_IDCT_AUTO)
1597 avctx->idct_algo=FF_IDCT_VP3;
1598 dsputil_init(&s->dsp, avctx);
1599
1600 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1601
1602 /* initialize to an impossible value which will force a recalculation
1603 * in the first frame decode */
1604 for (i = 0; i < 3; i++)
1605 s->qps[i] = -1;
1606
1607 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1608
1609 s->y_superblock_width = (s->width + 31) / 32;
1610 s->y_superblock_height = (s->height + 31) / 32;
1611 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1612
1613 /* work out the dimensions for the C planes */
1614 c_width = s->width >> s->chroma_x_shift;
1615 c_height = s->height >> s->chroma_y_shift;
1616 s->c_superblock_width = (c_width + 31) / 32;
1617 s->c_superblock_height = (c_height + 31) / 32;
1618 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1619
1620 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1621 s->u_superblock_start = s->y_superblock_count;
1622 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1623
1624 s->macroblock_width = (s->width + 15) / 16;
1625 s->macroblock_height = (s->height + 15) / 16;
1626 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1627
1628 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1629 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1630 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1631 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1632
1633 /* fragment count covers all 8x8 blocks for all 3 planes */
1634 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1635 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1636 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1637 s->fragment_start[1] = y_fragment_count;
1638 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1639
1640 if (!s->theora_tables)
1641 {
1642 for (i = 0; i < 64; i++) {
1643 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1644 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1645 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1646 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1647 s->base_matrix[2][i] = vp31_inter_dequant[i];
1648 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1649 }
1650
1651 for(inter=0; inter<2; inter++){
1652 for(plane=0; plane<3; plane++){
1653 s->qr_count[inter][plane]= 1;
1654 s->qr_size [inter][plane][0]= 63;
1655 s->qr_base [inter][plane][0]=
1656 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1657 }
1658 }
1659
1660 /* init VLC tables */
1661 for (i = 0; i < 16; i++) {
1662
1663 /* DC histograms */
1664 init_vlc(&s->dc_vlc[i], 11, 32,
1665 &dc_bias[i][0][1], 4, 2,
1666 &dc_bias[i][0][0], 4, 2, 0);
1667
1668 /* group 1 AC histograms */
1669 init_vlc(&s->ac_vlc_1[i], 11, 32,
1670 &ac_bias_0[i][0][1], 4, 2,
1671 &ac_bias_0[i][0][0], 4, 2, 0);
1672
1673 /* group 2 AC histograms */
1674 init_vlc(&s->ac_vlc_2[i], 11, 32,
1675 &ac_bias_1[i][0][1], 4, 2,
1676 &ac_bias_1[i][0][0], 4, 2, 0);
1677
1678 /* group 3 AC histograms */
1679 init_vlc(&s->ac_vlc_3[i], 11, 32,
1680 &ac_bias_2[i][0][1], 4, 2,
1681 &ac_bias_2[i][0][0], 4, 2, 0);
1682
1683 /* group 4 AC histograms */
1684 init_vlc(&s->ac_vlc_4[i], 11, 32,
1685 &ac_bias_3[i][0][1], 4, 2,
1686 &ac_bias_3[i][0][0], 4, 2, 0);
1687 }
1688 } else {
1689
1690 for (i = 0; i < 16; i++) {
1691 /* DC histograms */
1692 if (init_vlc(&s->dc_vlc[i], 11, 32,
1693 &s->huffman_table[i][0][1], 8, 4,
1694 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1695 goto vlc_fail;
1696
1697 /* group 1 AC histograms */
1698 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1699 &s->huffman_table[i+16][0][1], 8, 4,
1700 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1701 goto vlc_fail;
1702
1703 /* group 2 AC histograms */
1704 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1705 &s->huffman_table[i+16*2][0][1], 8, 4,
1706 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1707 goto vlc_fail;
1708
1709 /* group 3 AC histograms */
1710 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1711 &s->huffman_table[i+16*3][0][1], 8, 4,
1712 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1713 goto vlc_fail;
1714
1715 /* group 4 AC histograms */
1716 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1717 &s->huffman_table[i+16*4][0][1], 8, 4,
1718 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1719 goto vlc_fail;
1720 }
1721 }
1722
1723 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1724 &superblock_run_length_vlc_table[0][1], 4, 2,
1725 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1726
1727 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1728 &fragment_run_length_vlc_table[0][1], 4, 2,
1729 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1730
1731 init_vlc(&s->mode_code_vlc, 3, 8,
1732 &mode_code_vlc_table[0][1], 2, 1,
1733 &mode_code_vlc_table[0][0], 2, 1, 0);
1734
1735 init_vlc(&s->motion_vector_vlc, 6, 63,
1736 &motion_vector_vlc_table[0][1], 2, 1,
1737 &motion_vector_vlc_table[0][0], 2, 1, 0);
1738
1739 for (i = 0; i < 3; i++) {
1740 s->current_frame.data[i] = NULL;
1741 s->last_frame.data[i] = NULL;
1742 s->golden_frame.data[i] = NULL;
1743 }
1744
1745 return allocate_tables(avctx);
1746
1747 vlc_fail:
1748 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1749 return -1;
1750 }
1751
1752 /// Release and shuffle frames after decode finishes
1753 static void update_frames(AVCodecContext *avctx)
1754 {
1755 Vp3DecodeContext *s = avctx->priv_data;
1756
1757 /* release the last frame, if it is allocated and if it is not the
1758 * golden frame */
1759 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1760 ff_thread_release_buffer(avctx, &s->last_frame);
1761
1762 /* shuffle frames (last = current) */
1763 s->last_frame= s->current_frame;
1764
1765 if (s->keyframe) {
1766 if (s->golden_frame.data[0])
1767 ff_thread_release_buffer(avctx, &s->golden_frame);
1768 s->golden_frame = s->current_frame;
1769 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1770 }
1771
1772 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1773 }
1774
1775 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1776 {
1777 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1778 int qps_changed = 0, i, err;
1779
1780 if (!s1->current_frame.data[0]
1781 ||s->width != s1->width
1782 ||s->height!= s1->height)
1783 return -1;
1784
1785 if (s != s1) {
1786 // init tables if the first frame hasn't been decoded
1787 if (!s->current_frame.data[0]) {
1788 int y_fragment_count, c_fragment_count;
1789 s->avctx = dst;
1790 err = allocate_tables(dst);
1791 if (err)
1792 return err;
1793 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1794 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1795 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1796 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1797 }
1798
1799 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1800
1801 // copy previous frame data
1802 copy_fields(s, s1, golden_frame, dsp);
1803
1804 // copy qscale data if necessary
1805 for (i = 0; i < 3; i++) {
1806 if (s->qps[i] != s1->qps[1]) {
1807 qps_changed = 1;
1808 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1809 }
1810 }
1811
1812 if (s->qps[0] != s1->qps[0])
1813 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1814
1815 if (qps_changed)
1816 copy_fields(s, s1, qps, superblock_count);
1817 #undef copy_fields
1818 }
1819
1820 update_frames(dst);
1821
1822 return 0;
1823 }
1824
1825 /*
1826 * This is the ffmpeg/libavcodec API frame decode function.
1827 */
1828 static int vp3_decode_frame(AVCodecContext *avctx,
1829 void *data, int *data_size,
1830 AVPacket *avpkt)
1831 {
1832 const uint8_t *buf = avpkt->data;
1833 int buf_size = avpkt->size;
1834 Vp3DecodeContext *s = avctx->priv_data;
1835 GetBitContext gb;
1836 int i;
1837
1838 init_get_bits(&gb, buf, buf_size * 8);
1839
1840 if (s->theora && get_bits1(&gb))
1841 {
1842 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1843 return -1;
1844 }
1845
1846 s->keyframe = !get_bits1(&gb);
1847 if (!s->theora)
1848 skip_bits(&gb, 1);
1849 for (i = 0; i < 3; i++)
1850 s->last_qps[i] = s->qps[i];
1851
1852 s->nqps=0;
1853 do{
1854 s->qps[s->nqps++]= get_bits(&gb, 6);
1855 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1856 for (i = s->nqps; i < 3; i++)
1857 s->qps[i] = -1;
1858
1859 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1860 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1861 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1862
1863 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1864 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1865
1866 if (s->qps[0] != s->last_qps[0])
1867 init_loop_filter(s);
1868
1869 for (i = 0; i < s->nqps; i++)
1870 // reinit all dequantizers if the first one changed, because
1871 // the DC of the first quantizer must be used for all matrices
1872 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1873 init_dequantizer(s, i);
1874
1875 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1876 return buf_size;
1877
1878 s->current_frame.reference = 3;
1879 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1880 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1881 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1882 goto error;
1883 }
1884
1885 if (!s->edge_emu_buffer)
1886 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1887
1888 if (s->keyframe) {
1889 if (!s->theora)
1890 {
1891 skip_bits(&gb, 4); /* width code */
1892 skip_bits(&gb, 4); /* height code */
1893 if (s->version)
1894 {
1895 s->version = get_bits(&gb, 5);
1896 if (avctx->frame_number == 0)
1897 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1898 }
1899 }
1900 if (s->version || s->theora)
1901 {
1902 if (get_bits1(&gb))
1903 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1904 skip_bits(&gb, 2); /* reserved? */
1905 }
1906 } else {
1907 if (!s->golden_frame.data[0]) {
1908 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1909
1910 s->golden_frame.reference = 3;
1911 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1912 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1913 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1914 goto error;
1915 }
1916 s->last_frame = s->golden_frame;
1917 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1918 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
1919 }
1920 }
1921
1922 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1923 ff_thread_finish_setup(avctx);
1924
1925 if (unpack_superblocks(s, &gb)){
1926 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1927 goto error;
1928 }
1929 if (unpack_modes(s, &gb)){
1930 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1931 goto error;
1932 }
1933 if (unpack_vectors(s, &gb)){
1934 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1935 goto error;
1936 }
1937 if (unpack_block_qpis(s, &gb)){
1938 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1939 goto error;
1940 }
1941 if (unpack_dct_coeffs(s, &gb)){
1942 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1943 goto error;
1944 }
1945
1946 for (i = 0; i < 3; i++) {
1947 int height = s->height >> (i && s->chroma_y_shift);
1948 if (s->flipped_image)
1949 s->data_offset[i] = 0;
1950 else
1951 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1952 }
1953
1954 s->last_slice_end = 0;
1955 for (i = 0; i < s->c_superblock_height; i++)
1956 render_slice(s, i);
1957
1958 // filter the last row
1959 for (i = 0; i < 3; i++) {
1960 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1961 apply_loop_filter(s, i, row, row+1);
1962 }
1963 vp3_draw_horiz_band(s, s->avctx->height);
1964
1965 *data_size=sizeof(AVFrame);
1966 *(AVFrame*)data= s->current_frame;
1967
1968 if (!HAVE_PTHREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1969 update_frames(avctx);
1970
1971 return buf_size;
1972
1973 error:
1974 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
1975
1976 if (!HAVE_PTHREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1977 avctx->release_buffer(avctx, &s->current_frame);
1978
1979 return -1;
1980 }
1981
1982 /*
1983 * This is the ffmpeg/libavcodec API module cleanup function.
1984 */
1985 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1986 {
1987 Vp3DecodeContext *s = avctx->priv_data;
1988 int i;
1989
1990 if (avctx->is_copy && !s->current_frame.data[0])
1991 return 0;
1992
1993 av_free(s->superblock_coding);
1994 av_free(s->all_fragments);
1995 av_free(s->coded_fragment_list[0]);
1996 av_free(s->dct_tokens_base);
1997 av_free(s->superblock_fragments);
1998 av_free(s->macroblock_coding);
1999 av_free(s->motion_val[0]);
2000 av_free(s->motion_val[1]);
2001 av_free(s->edge_emu_buffer);
2002
2003 if (avctx->is_copy) return 0;
2004
2005 for (i = 0; i < 16; i++) {
2006 free_vlc(&s->dc_vlc[i]);
2007 free_vlc(&s->ac_vlc_1[i]);
2008 free_vlc(&s->ac_vlc_2[i]);
2009 free_vlc(&s->ac_vlc_3[i]);
2010 free_vlc(&s->ac_vlc_4[i]);
2011 }
2012
2013 free_vlc(&s->superblock_run_length_vlc);
2014 free_vlc(&s->fragment_run_length_vlc);
2015 free_vlc(&s->mode_code_vlc);
2016 free_vlc(&s->motion_vector_vlc);
2017
2018 /* release all frames */
2019 if (s->golden_frame.data[0])
2020 ff_thread_release_buffer(avctx, &s->golden_frame);
2021 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
2022 ff_thread_release_buffer(avctx, &s->last_frame);
2023 /* no need to release the current_frame since it will always be pointing
2024 * to the same frame as either the golden or last frame */
2025
2026 return 0;
2027 }
2028
2029 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2030 {
2031 Vp3DecodeContext *s = avctx->priv_data;
2032
2033 if (get_bits1(gb)) {
2034 int token;
2035 if (s->entries >= 32) { /* overflow */
2036 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2037 return -1;
2038 }
2039 token = get_bits(gb, 5);
2040 //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);
2041 s->huffman_table[s->hti][token][0] = s->hbits;
2042 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2043 s->entries++;
2044 }
2045 else {
2046 if (s->huff_code_size >= 32) {/* overflow */
2047 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2048 return -1;
2049 }
2050 s->huff_code_size++;
2051 s->hbits <<= 1;
2052 if (read_huffman_tree(avctx, gb))
2053 return -1;
2054 s->hbits |= 1;
2055 if (read_huffman_tree(avctx, gb))
2056 return -1;
2057 s->hbits >>= 1;
2058 s->huff_code_size--;
2059 }
2060 return 0;
2061 }
2062
2063 #if CONFIG_THEORA_DECODER
2064 static const enum PixelFormat theora_pix_fmts[4] = {
2065 PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
2066 };
2067
2068 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2069 {
2070 Vp3DecodeContext *s = avctx->priv_data;
2071 int visible_width, visible_height, colorspace;
2072 int offset_x = 0, offset_y = 0;
2073 AVRational fps, aspect;
2074
2075 s->theora = get_bits_long(gb, 24);
2076 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2077
2078 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2079 /* but previous versions have the image flipped relative to vp3 */
2080 if (s->theora < 0x030200)
2081 {
2082 s->flipped_image = 1;
2083 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2084 }
2085
2086 visible_width = s->width = get_bits(gb, 16) << 4;
2087 visible_height = s->height = get_bits(gb, 16) << 4;
2088
2089 if(av_image_check_size(s->width, s->height, 0, avctx)){
2090 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2091 s->width= s->height= 0;
2092 return -1;
2093 }
2094
2095 if (s->theora >= 0x030200) {
2096 visible_width = get_bits_long(gb, 24);
2097 visible_height = get_bits_long(gb, 24);
2098
2099 offset_x = get_bits(gb, 8); /* offset x */
2100 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2101 }
2102
2103 fps.num = get_bits_long(gb, 32);
2104 fps.den = get_bits_long(gb, 32);
2105 if (fps.num && fps.den) {
2106 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2107 fps.den, fps.num, 1<<30);
2108 }
2109
2110 aspect.num = get_bits_long(gb, 24);
2111 aspect.den = get_bits_long(gb, 24);
2112 if (aspect.num && aspect.den) {
2113 av_reduce(&avctx->sample_aspect_ratio.num,
2114 &avctx->sample_aspect_ratio.den,
2115 aspect.num, aspect.den, 1<<30);
2116 }
2117
2118 if (s->theora < 0x030200)
2119 skip_bits(gb, 5); /* keyframe frequency force */
2120 colorspace = get_bits(gb, 8);
2121 skip_bits(gb, 24); /* bitrate */
2122
2123 skip_bits(gb, 6); /* quality hint */
2124
2125 if (s->theora >= 0x030200)
2126 {
2127 skip_bits(gb, 5); /* keyframe frequency force */
2128 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2129 skip_bits(gb, 3); /* reserved */
2130 }
2131
2132 // align_get_bits(gb);
2133
2134 if ( visible_width <= s->width && visible_width > s->width-16
2135 && visible_height <= s->height && visible_height > s->height-16
2136 && !offset_x && (offset_y == s->height - visible_height))
2137 avcodec_set_dimensions(avctx, visible_width, visible_height);
2138 else
2139 avcodec_set_dimensions(avctx, s->width, s->height);
2140
2141 if (colorspace == 1) {
2142 avctx->color_primaries = AVCOL_PRI_BT470M;
2143 } else if (colorspace == 2) {
2144 avctx->color_primaries = AVCOL_PRI_BT470BG;
2145 }
2146 if (colorspace == 1 || colorspace == 2) {
2147 avctx->colorspace = AVCOL_SPC_BT470BG;
2148 avctx->color_trc = AVCOL_TRC_BT709;
2149 }
2150
2151 return 0;
2152 }
2153
2154 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2155 {
2156 Vp3DecodeContext *s = avctx->priv_data;
2157 int i, n, matrices, inter, plane;
2158
2159 if (s->theora >= 0x030200) {
2160 n = get_bits(gb, 3);
2161 /* loop filter limit values table */
2162 if (n)
2163 for (i = 0; i < 64; i++)
2164 s->filter_limit_values[i] = get_bits(gb, n);
2165 }
2166
2167 if (s->theora >= 0x030200)
2168 n = get_bits(gb, 4) + 1;
2169 else
2170 n = 16;
2171 /* quality threshold table */
2172 for (i = 0; i < 64; i++)
2173 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2174
2175 if (s->theora >= 0x030200)
2176 n = get_bits(gb, 4) + 1;
2177 else
2178 n = 16;
2179 /* dc scale factor table */
2180 for (i = 0; i < 64; i++)
2181 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2182
2183 if (s->theora >= 0x030200)
2184 matrices = get_bits(gb, 9) + 1;
2185 else
2186 matrices = 3;
2187
2188 if(matrices > 384){
2189 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2190 return -1;
2191 }
2192
2193 for(n=0; n<matrices; n++){
2194 for (i = 0; i < 64; i++)
2195 s->base_matrix[n][i]= get_bits(gb, 8);
2196 }
2197
2198 for (inter = 0; inter <= 1; inter++) {
2199 for (plane = 0; plane <= 2; plane++) {
2200 int newqr= 1;
2201 if (inter || plane > 0)
2202 newqr = get_bits1(gb);
2203 if (!newqr) {
2204 int qtj, plj;
2205 if(inter && get_bits1(gb)){
2206 qtj = 0;
2207 plj = plane;
2208 }else{
2209 qtj= (3*inter + plane - 1) / 3;
2210 plj= (plane + 2) % 3;
2211 }
2212 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2213 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2214 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2215 } else {
2216 int qri= 0;
2217 int qi = 0;
2218
2219 for(;;){
2220 i= get_bits(gb, av_log2(matrices-1)+1);
2221 if(i>= matrices){
2222 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2223 return -1;
2224 }
2225 s->qr_base[inter][plane][qri]= i;
2226 if(qi >= 63)
2227 break;
2228 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2229 s->qr_size[inter][plane][qri++]= i;
2230 qi += i;
2231 }
2232
2233 if (qi > 63) {
2234 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2235 return -1;
2236 }
2237 s->qr_count[inter][plane]= qri;
2238 }
2239 }
2240 }
2241
2242 /* Huffman tables */
2243 for (s->hti = 0; s->hti < 80; s->hti++) {
2244 s->entries = 0;
2245 s->huff_code_size = 1;
2246 if (!get_bits1(gb)) {
2247 s->hbits = 0;
2248 if(read_huffman_tree(avctx, gb))
2249 return -1;
2250 s->hbits = 1;
2251 if(read_huffman_tree(avctx, gb))
2252 return -1;
2253 }
2254 }
2255
2256 s->theora_tables = 1;
2257
2258 return 0;
2259 }
2260
2261 static av_cold int theora_decode_init(AVCodecContext *avctx)
2262 {
2263 Vp3DecodeContext *s = avctx->priv_data;
2264 GetBitContext gb;
2265 int ptype;
2266 uint8_t *header_start[3];
2267 int header_len[3];
2268 int i;
2269
2270 s->theora = 1;
2271
2272 if (!avctx->extradata_size)
2273 {
2274 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2275 return -1;
2276 }
2277
2278 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2279 42, header_start, header_len) < 0) {
2280 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2281 return -1;
2282 }
2283
2284 for(i=0;i<3;i++) {
2285 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2286
2287 ptype = get_bits(&gb, 8);
2288
2289 if (!(ptype & 0x80))
2290 {
2291 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2292 // return -1;
2293 }
2294
2295 // FIXME: Check for this as well.
2296 skip_bits_long(&gb, 6*8); /* "theora" */
2297
2298 switch(ptype)
2299 {
2300 case 0x80:
2301 theora_decode_header(avctx, &gb);
2302 break;
2303 case 0x81:
2304 // FIXME: is this needed? it breaks sometimes
2305 // theora_decode_comments(avctx, gb);
2306 break;
2307 case 0x82:
2308 if (theora_decode_tables(avctx, &gb))
2309 return -1;
2310 break;
2311 default:
2312 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2313 break;
2314 }
2315 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2316 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2317 if (s->theora < 0x030200)
2318 break;
2319 }
2320
2321 return vp3_decode_init(avctx);
2322 }
2323
2324 static void vp3_decode_flush(AVCodecContext *avctx)
2325 {
2326 Vp3DecodeContext *s = avctx->priv_data;
2327
2328 if (s->golden_frame.data[0]) {
2329 if (s->golden_frame.data[0] == s->last_frame.data[0])
2330 memset(&s->last_frame, 0, sizeof(AVFrame));
2331 if (s->current_frame.data[0] == s->golden_frame.data[0])
2332 memset(&s->current_frame, 0, sizeof(AVFrame));
2333 ff_thread_release_buffer(avctx, &s->golden_frame);
2334 }
2335 if (s->last_frame.data[0]) {
2336 if (s->current_frame.data[0] == s->last_frame.data[0])
2337 memset(&s->current_frame, 0, sizeof(AVFrame));
2338 ff_thread_release_buffer(avctx, &s->last_frame);
2339 }
2340 if (s->current_frame.data[0])
2341 ff_thread_release_buffer(avctx, &s->current_frame);
2342 }
2343
2344 AVCodec ff_theora_decoder = {
2345 .name = "theora",
2346 .type = AVMEDIA_TYPE_VIDEO,
2347 .id = CODEC_ID_THEORA,
2348 .priv_data_size = sizeof(Vp3DecodeContext),
2349 .init = theora_decode_init,
2350 .close = vp3_decode_end,
2351 .decode = vp3_decode_frame,
2352 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2353 .flush = vp3_decode_flush,
2354 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2355 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2356 };
2357 #endif
2358
2359 AVCodec ff_vp3_decoder = {
2360 .name = "vp3",
2361 .type = AVMEDIA_TYPE_VIDEO,
2362 .id = CODEC_ID_VP3,
2363 .priv_data_size = sizeof(Vp3DecodeContext),
2364 .init = vp3_decode_init,
2365 .close = vp3_decode_end,
2366 .decode = vp3_decode_frame,
2367 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2368 .flush = vp3_decode_flush,
2369 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2370 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2371 };