6e04b7b01b6fb4bd32d3fca5e31061e141b62a35
[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 "internal.h"
39 #include "dsputil.h"
40 #include "get_bits.h"
41
42 #include "vp3data.h"
43 #include "xiph.h"
44 #include "thread.h"
45
46 #define FRAGMENT_PIXELS 8
47
48 static av_cold int vp3_decode_end(AVCodecContext *avctx);
49 static void vp3_decode_flush(AVCodecContext *avctx);
50
51 //FIXME split things out into their own arrays
52 typedef struct Vp3Fragment {
53 int16_t dc;
54 uint8_t coding_method;
55 uint8_t qpi;
56 } Vp3Fragment;
57
58 #define SB_NOT_CODED 0
59 #define SB_PARTIALLY_CODED 1
60 #define SB_FULLY_CODED 2
61
62 // This is the maximum length of a single long bit run that can be encoded
63 // for superblock coding or block qps. Theora special-cases this to read a
64 // bit instead of flipping the current bit to allow for runs longer than 4129.
65 #define MAXIMUM_LONG_BIT_RUN 4129
66
67 #define MODE_INTER_NO_MV 0
68 #define MODE_INTRA 1
69 #define MODE_INTER_PLUS_MV 2
70 #define MODE_INTER_LAST_MV 3
71 #define MODE_INTER_PRIOR_LAST 4
72 #define MODE_USING_GOLDEN 5
73 #define MODE_GOLDEN_MV 6
74 #define MODE_INTER_FOURMV 7
75 #define CODING_MODE_COUNT 8
76
77 /* special internal mode */
78 #define MODE_COPY 8
79
80 /* There are 6 preset schemes, plus a free-form scheme */
81 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
82 {
83 /* scheme 1: Last motion vector dominates */
84 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
85 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
86 MODE_INTRA, MODE_USING_GOLDEN,
87 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
88
89 /* scheme 2 */
90 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
91 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
92 MODE_INTRA, MODE_USING_GOLDEN,
93 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
94
95 /* scheme 3 */
96 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
97 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
98 MODE_INTRA, MODE_USING_GOLDEN,
99 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
100
101 /* scheme 4 */
102 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
103 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
104 MODE_INTRA, MODE_USING_GOLDEN,
105 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
106
107 /* scheme 5: No motion vector dominates */
108 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
109 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
110 MODE_INTRA, MODE_USING_GOLDEN,
111 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
112
113 /* scheme 6 */
114 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
115 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
116 MODE_INTER_PLUS_MV, MODE_INTRA,
117 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
118
119 };
120
121 static const uint8_t hilbert_offset[16][2] = {
122 {0,0}, {1,0}, {1,1}, {0,1},
123 {0,2}, {0,3}, {1,3}, {1,2},
124 {2,2}, {2,3}, {3,3}, {3,2},
125 {3,1}, {2,1}, {2,0}, {3,0}
126 };
127
128 #define MIN_DEQUANT_VAL 2
129
130 typedef struct Vp3DecodeContext {
131 AVCodecContext *avctx;
132 int theora, theora_tables;
133 int version;
134 int width, height;
135 int chroma_x_shift, chroma_y_shift;
136 AVFrame golden_frame;
137 AVFrame last_frame;
138 AVFrame current_frame;
139 int keyframe;
140 DSPContext dsp;
141 int flipped_image;
142 int last_slice_end;
143 int skip_loop_filter;
144
145 int qps[3];
146 int nqps;
147 int last_qps[3];
148
149 int superblock_count;
150 int y_superblock_width;
151 int y_superblock_height;
152 int y_superblock_count;
153 int c_superblock_width;
154 int c_superblock_height;
155 int c_superblock_count;
156 int u_superblock_start;
157 int v_superblock_start;
158 unsigned char *superblock_coding;
159
160 int macroblock_count;
161 int macroblock_width;
162 int macroblock_height;
163
164 int fragment_count;
165 int fragment_width[2];
166 int fragment_height[2];
167
168 Vp3Fragment *all_fragments;
169 int fragment_start[3];
170 int data_offset[3];
171
172 int8_t (*motion_val[2])[2];
173
174 ScanTable scantable;
175
176 /* tables */
177 uint16_t coded_dc_scale_factor[64];
178 uint32_t coded_ac_scale_factor[64];
179 uint8_t base_matrix[384][64];
180 uint8_t qr_count[2][3];
181 uint8_t qr_size [2][3][64];
182 uint16_t qr_base[2][3][64];
183
184 /**
185 * This is a list of all tokens in bitstream order. Reordering takes place
186 * by pulling from each level during IDCT. As a consequence, IDCT must be
187 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
188 * otherwise. The 32 different tokens with up to 12 bits of extradata are
189 * collapsed into 3 types, packed as follows:
190 * (from the low to high bits)
191 *
192 * 2 bits: type (0,1,2)
193 * 0: EOB run, 14 bits for run length (12 needed)
194 * 1: zero run, 7 bits for run length
195 * 7 bits for the next coefficient (3 needed)
196 * 2: coefficient, 14 bits (11 needed)
197 *
198 * Coefficients are signed, so are packed in the highest bits for automatic
199 * sign extension.
200 */
201 int16_t *dct_tokens[3][64];
202 int16_t *dct_tokens_base;
203 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
204 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
205 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
206
207 /**
208 * number of blocks that contain DCT coefficients at the given level or higher
209 */
210 int num_coded_frags[3][64];
211 int total_num_coded_frags;
212
213 /* this is a list of indexes into the all_fragments array indicating
214 * which of the fragments are coded */
215 int *coded_fragment_list[3];
216
217 VLC dc_vlc[16];
218 VLC ac_vlc_1[16];
219 VLC ac_vlc_2[16];
220 VLC ac_vlc_3[16];
221 VLC ac_vlc_4[16];
222
223 VLC superblock_run_length_vlc;
224 VLC fragment_run_length_vlc;
225 VLC mode_code_vlc;
226 VLC motion_vector_vlc;
227
228 /* these arrays need to be on 16-byte boundaries since SSE2 operations
229 * index into them */
230 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
231
232 /* This table contains superblock_count * 16 entries. Each set of 16
233 * numbers corresponds to the fragment indexes 0..15 of the superblock.
234 * An entry will be -1 to indicate that no entry corresponds to that
235 * index. */
236 int *superblock_fragments;
237
238 /* This is an array that indicates how a particular macroblock
239 * is coded. */
240 unsigned char *macroblock_coding;
241
242 uint8_t *edge_emu_buffer;
243
244 /* Huffman decode */
245 int hti;
246 unsigned int hbits;
247 int entries;
248 int huff_code_size;
249 uint32_t huffman_table[80][32][2];
250
251 uint8_t filter_limit_values[64];
252 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
253 } Vp3DecodeContext;
254
255 /************************************************************************
256 * VP3 specific functions
257 ************************************************************************/
258
259 /*
260 * This function sets up all of the various blocks mappings:
261 * superblocks <-> fragments, macroblocks <-> fragments,
262 * superblocks <-> macroblocks
263 *
264 * @return 0 is successful; returns 1 if *anything* went wrong.
265 */
266 static int init_block_mapping(Vp3DecodeContext *s)
267 {
268 int sb_x, sb_y, plane;
269 int x, y, i, j = 0;
270
271 for (plane = 0; plane < 3; plane++) {
272 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
273 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
274 int frag_width = s->fragment_width[!!plane];
275 int frag_height = s->fragment_height[!!plane];
276
277 for (sb_y = 0; sb_y < sb_height; sb_y++)
278 for (sb_x = 0; sb_x < sb_width; sb_x++)
279 for (i = 0; i < 16; i++) {
280 x = 4*sb_x + hilbert_offset[i][0];
281 y = 4*sb_y + hilbert_offset[i][1];
282
283 if (x < frag_width && y < frag_height)
284 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
285 else
286 s->superblock_fragments[j++] = -1;
287 }
288 }
289
290 return 0; /* successful path out */
291 }
292
293 /*
294 * This function sets up the dequantization tables used for a particular
295 * frame.
296 */
297 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
298 {
299 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
300 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
301 int i, plane, inter, qri, bmi, bmj, qistart;
302
303 for(inter=0; inter<2; inter++){
304 for(plane=0; plane<3; plane++){
305 int sum=0;
306 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
307 sum+= s->qr_size[inter][plane][qri];
308 if(s->qps[qpi] <= sum)
309 break;
310 }
311 qistart= sum - s->qr_size[inter][plane][qri];
312 bmi= s->qr_base[inter][plane][qri ];
313 bmj= s->qr_base[inter][plane][qri+1];
314 for(i=0; i<64; i++){
315 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
316 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
317 + s->qr_size[inter][plane][qri])
318 / (2*s->qr_size[inter][plane][qri]);
319
320 int qmin= 8<<(inter + !i);
321 int qscale= i ? ac_scale_factor : dc_scale_factor;
322
323 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
324 }
325 // all DC coefficients use the same quant so as not to interfere with DC prediction
326 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
327 }
328 }
329 }
330
331 /*
332 * This function initializes the loop filter boundary limits if the frame's
333 * quality index is different from the previous frame's.
334 *
335 * The filter_limit_values may not be larger than 127.
336 */
337 static void init_loop_filter(Vp3DecodeContext *s)
338 {
339 int *bounding_values= s->bounding_values_array+127;
340 int filter_limit;
341 int x;
342 int value;
343
344 filter_limit = s->filter_limit_values[s->qps[0]];
345
346 /* set up the bounding values */
347 memset(s->bounding_values_array, 0, 256 * sizeof(int));
348 for (x = 0; x < filter_limit; x++) {
349 bounding_values[-x] = -x;
350 bounding_values[x] = x;
351 }
352 for (x = value = filter_limit; x < 128 && value; x++, value--) {
353 bounding_values[ x] = value;
354 bounding_values[-x] = -value;
355 }
356 if (value)
357 bounding_values[128] = value;
358 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
359 }
360
361 /*
362 * This function unpacks all of the superblock/macroblock/fragment coding
363 * information from the bitstream.
364 */
365 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
366 {
367 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
368 int bit = 0;
369 int current_superblock = 0;
370 int current_run = 0;
371 int num_partial_superblocks = 0;
372
373 int i, j;
374 int current_fragment;
375 int plane;
376
377 if (s->keyframe) {
378 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
379
380 } else {
381
382 /* unpack the list of partially-coded superblocks */
383 bit = get_bits1(gb) ^ 1;
384 current_run = 0;
385
386 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
387 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
388 bit = get_bits1(gb);
389 else
390 bit ^= 1;
391
392 current_run = get_vlc2(gb,
393 s->superblock_run_length_vlc.table, 6, 2) + 1;
394 if (current_run == 34)
395 current_run += get_bits(gb, 12);
396
397 if (current_superblock + current_run > s->superblock_count) {
398 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
399 return -1;
400 }
401
402 memset(s->superblock_coding + current_superblock, bit, current_run);
403
404 current_superblock += current_run;
405 if (bit)
406 num_partial_superblocks += current_run;
407 }
408
409 /* unpack the list of fully coded superblocks if any of the blocks were
410 * not marked as partially coded in the previous step */
411 if (num_partial_superblocks < s->superblock_count) {
412 int superblocks_decoded = 0;
413
414 current_superblock = 0;
415 bit = get_bits1(gb) ^ 1;
416 current_run = 0;
417
418 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
419 && get_bits_left(gb) > 0) {
420
421 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
422 bit = get_bits1(gb);
423 else
424 bit ^= 1;
425
426 current_run = get_vlc2(gb,
427 s->superblock_run_length_vlc.table, 6, 2) + 1;
428 if (current_run == 34)
429 current_run += get_bits(gb, 12);
430
431 for (j = 0; j < current_run; current_superblock++) {
432 if (current_superblock >= s->superblock_count) {
433 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
434 return -1;
435 }
436
437 /* skip any superblocks already marked as partially coded */
438 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
439 s->superblock_coding[current_superblock] = 2*bit;
440 j++;
441 }
442 }
443 superblocks_decoded += current_run;
444 }
445 }
446
447 /* if there were partial blocks, initialize bitstream for
448 * unpacking fragment codings */
449 if (num_partial_superblocks) {
450
451 current_run = 0;
452 bit = get_bits1(gb);
453 /* toggle the bit because as soon as the first run length is
454 * fetched the bit will be toggled again */
455 bit ^= 1;
456 }
457 }
458
459 /* figure out which fragments are coded; iterate through each
460 * superblock (all planes) */
461 s->total_num_coded_frags = 0;
462 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
463
464 for (plane = 0; plane < 3; plane++) {
465 int sb_start = superblock_starts[plane];
466 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
467 int num_coded_frags = 0;
468
469 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
470
471 /* iterate through all 16 fragments in a superblock */
472 for (j = 0; j < 16; j++) {
473
474 /* if the fragment is in bounds, check its coding status */
475 current_fragment = s->superblock_fragments[i * 16 + j];
476 if (current_fragment != -1) {
477 int coded = s->superblock_coding[i];
478
479 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
480
481 /* fragment may or may not be coded; this is the case
482 * that cares about the fragment coding runs */
483 if (current_run-- == 0) {
484 bit ^= 1;
485 current_run = get_vlc2(gb,
486 s->fragment_run_length_vlc.table, 5, 2);
487 }
488 coded = bit;
489 }
490
491 if (coded) {
492 /* default mode; actual mode will be decoded in
493 * the next phase */
494 s->all_fragments[current_fragment].coding_method =
495 MODE_INTER_NO_MV;
496 s->coded_fragment_list[plane][num_coded_frags++] =
497 current_fragment;
498 } else {
499 /* not coded; copy this fragment from the prior frame */
500 s->all_fragments[current_fragment].coding_method =
501 MODE_COPY;
502 }
503 }
504 }
505 }
506 s->total_num_coded_frags += num_coded_frags;
507 for (i = 0; i < 64; i++)
508 s->num_coded_frags[plane][i] = num_coded_frags;
509 if (plane < 2)
510 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
511 }
512 return 0;
513 }
514
515 /*
516 * This function unpacks all the coding mode data for individual macroblocks
517 * from the bitstream.
518 */
519 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
520 {
521 int i, j, k, sb_x, sb_y;
522 int scheme;
523 int current_macroblock;
524 int current_fragment;
525 int coding_mode;
526 int custom_mode_alphabet[CODING_MODE_COUNT];
527 const int *alphabet;
528 Vp3Fragment *frag;
529
530 if (s->keyframe) {
531 for (i = 0; i < s->fragment_count; i++)
532 s->all_fragments[i].coding_method = MODE_INTRA;
533
534 } else {
535
536 /* fetch the mode coding scheme for this frame */
537 scheme = get_bits(gb, 3);
538
539 /* is it a custom coding scheme? */
540 if (scheme == 0) {
541 for (i = 0; i < 8; i++)
542 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
543 for (i = 0; i < 8; i++)
544 custom_mode_alphabet[get_bits(gb, 3)] = i;
545 alphabet = custom_mode_alphabet;
546 } else
547 alphabet = ModeAlphabet[scheme-1];
548
549 /* iterate through all of the macroblocks that contain 1 or more
550 * coded fragments */
551 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
552 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
553 if (get_bits_left(gb) <= 0)
554 return -1;
555
556 for (j = 0; j < 4; j++) {
557 int mb_x = 2*sb_x + (j>>1);
558 int mb_y = 2*sb_y + (((j>>1)+j)&1);
559 current_macroblock = mb_y * s->macroblock_width + mb_x;
560
561 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
562 continue;
563
564 #define BLOCK_X (2*mb_x + (k&1))
565 #define BLOCK_Y (2*mb_y + (k>>1))
566 /* coding modes are only stored if the macroblock has at least one
567 * luma block coded, otherwise it must be INTER_NO_MV */
568 for (k = 0; k < 4; k++) {
569 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
570 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
571 break;
572 }
573 if (k == 4) {
574 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
575 continue;
576 }
577
578 /* mode 7 means get 3 bits for each coding mode */
579 if (scheme == 7)
580 coding_mode = get_bits(gb, 3);
581 else
582 coding_mode = alphabet
583 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
584
585 s->macroblock_coding[current_macroblock] = coding_mode;
586 for (k = 0; k < 4; k++) {
587 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
588 if (frag->coding_method != MODE_COPY)
589 frag->coding_method = coding_mode;
590 }
591
592 #define SET_CHROMA_MODES \
593 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
594 frag[s->fragment_start[1]].coding_method = coding_mode;\
595 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
596 frag[s->fragment_start[2]].coding_method = coding_mode;
597
598 if (s->chroma_y_shift) {
599 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
600 SET_CHROMA_MODES
601 } else if (s->chroma_x_shift) {
602 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
603 for (k = 0; k < 2; k++) {
604 SET_CHROMA_MODES
605 frag += s->fragment_width[1];
606 }
607 } else {
608 for (k = 0; k < 4; k++) {
609 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
610 SET_CHROMA_MODES
611 }
612 }
613 }
614 }
615 }
616 }
617
618 return 0;
619 }
620
621 /*
622 * This function unpacks all the motion vectors for the individual
623 * macroblocks from the bitstream.
624 */
625 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
626 {
627 int j, k, sb_x, sb_y;
628 int coding_mode;
629 int motion_x[4];
630 int motion_y[4];
631 int last_motion_x = 0;
632 int last_motion_y = 0;
633 int prior_last_motion_x = 0;
634 int prior_last_motion_y = 0;
635 int current_macroblock;
636 int current_fragment;
637 int frag;
638
639 if (s->keyframe)
640 return 0;
641
642 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
643 coding_mode = get_bits1(gb);
644
645 /* iterate through all of the macroblocks that contain 1 or more
646 * coded fragments */
647 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
648 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
649 if (get_bits_left(gb) <= 0)
650 return -1;
651
652 for (j = 0; j < 4; j++) {
653 int mb_x = 2*sb_x + (j>>1);
654 int mb_y = 2*sb_y + (((j>>1)+j)&1);
655 current_macroblock = mb_y * s->macroblock_width + mb_x;
656
657 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
658 (s->macroblock_coding[current_macroblock] == MODE_COPY))
659 continue;
660
661 switch (s->macroblock_coding[current_macroblock]) {
662
663 case MODE_INTER_PLUS_MV:
664 case MODE_GOLDEN_MV:
665 /* all 6 fragments use the same motion vector */
666 if (coding_mode == 0) {
667 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
668 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
669 } else {
670 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
671 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
672 }
673
674 /* vector maintenance, only on MODE_INTER_PLUS_MV */
675 if (s->macroblock_coding[current_macroblock] ==
676 MODE_INTER_PLUS_MV) {
677 prior_last_motion_x = last_motion_x;
678 prior_last_motion_y = last_motion_y;
679 last_motion_x = motion_x[0];
680 last_motion_y = motion_y[0];
681 }
682 break;
683
684 case MODE_INTER_FOURMV:
685 /* vector maintenance */
686 prior_last_motion_x = last_motion_x;
687 prior_last_motion_y = last_motion_y;
688
689 /* fetch 4 vectors from the bitstream, one for each
690 * Y fragment, then average for the C fragment vectors */
691 for (k = 0; k < 4; k++) {
692 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
693 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
694 if (coding_mode == 0) {
695 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
696 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
697 } else {
698 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
699 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
700 }
701 last_motion_x = motion_x[k];
702 last_motion_y = motion_y[k];
703 } else {
704 motion_x[k] = 0;
705 motion_y[k] = 0;
706 }
707 }
708 break;
709
710 case MODE_INTER_LAST_MV:
711 /* all 6 fragments use the last motion vector */
712 motion_x[0] = last_motion_x;
713 motion_y[0] = last_motion_y;
714
715 /* no vector maintenance (last vector remains the
716 * last vector) */
717 break;
718
719 case MODE_INTER_PRIOR_LAST:
720 /* all 6 fragments use the motion vector prior to the
721 * last motion vector */
722 motion_x[0] = prior_last_motion_x;
723 motion_y[0] = prior_last_motion_y;
724
725 /* vector maintenance */
726 prior_last_motion_x = last_motion_x;
727 prior_last_motion_y = last_motion_y;
728 last_motion_x = motion_x[0];
729 last_motion_y = motion_y[0];
730 break;
731
732 default:
733 /* covers intra, inter without MV, golden without MV */
734 motion_x[0] = 0;
735 motion_y[0] = 0;
736
737 /* no vector maintenance */
738 break;
739 }
740
741 /* assign the motion vectors to the correct fragments */
742 for (k = 0; k < 4; k++) {
743 current_fragment =
744 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
745 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
746 s->motion_val[0][current_fragment][0] = motion_x[k];
747 s->motion_val[0][current_fragment][1] = motion_y[k];
748 } else {
749 s->motion_val[0][current_fragment][0] = motion_x[0];
750 s->motion_val[0][current_fragment][1] = motion_y[0];
751 }
752 }
753
754 if (s->chroma_y_shift) {
755 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
756 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
757 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
758 }
759 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
760 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
761 frag = mb_y*s->fragment_width[1] + mb_x;
762 s->motion_val[1][frag][0] = motion_x[0];
763 s->motion_val[1][frag][1] = motion_y[0];
764 } else if (s->chroma_x_shift) {
765 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
766 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
767 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
768 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
769 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
770 } else {
771 motion_x[1] = motion_x[0];
772 motion_y[1] = motion_y[0];
773 }
774 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
775 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
776
777 frag = 2*mb_y*s->fragment_width[1] + mb_x;
778 for (k = 0; k < 2; k++) {
779 s->motion_val[1][frag][0] = motion_x[k];
780 s->motion_val[1][frag][1] = motion_y[k];
781 frag += s->fragment_width[1];
782 }
783 } else {
784 for (k = 0; k < 4; k++) {
785 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
786 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
787 s->motion_val[1][frag][0] = motion_x[k];
788 s->motion_val[1][frag][1] = motion_y[k];
789 } else {
790 s->motion_val[1][frag][0] = motion_x[0];
791 s->motion_val[1][frag][1] = motion_y[0];
792 }
793 }
794 }
795 }
796 }
797 }
798
799 return 0;
800 }
801
802 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
803 {
804 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
805 int num_blocks = s->total_num_coded_frags;
806
807 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
808 i = blocks_decoded = num_blocks_at_qpi = 0;
809
810 bit = get_bits1(gb) ^ 1;
811 run_length = 0;
812
813 do {
814 if (run_length == MAXIMUM_LONG_BIT_RUN)
815 bit = get_bits1(gb);
816 else
817 bit ^= 1;
818
819 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
820 if (run_length == 34)
821 run_length += get_bits(gb, 12);
822 blocks_decoded += run_length;
823
824 if (!bit)
825 num_blocks_at_qpi += run_length;
826
827 for (j = 0; j < run_length; i++) {
828 if (i >= s->total_num_coded_frags)
829 return -1;
830
831 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
832 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
833 j++;
834 }
835 }
836 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
837
838 num_blocks -= num_blocks_at_qpi;
839 }
840
841 return 0;
842 }
843
844 /*
845 * This function is called by unpack_dct_coeffs() to extract the VLCs from
846 * the bitstream. The VLCs encode tokens which are used to unpack DCT
847 * data. This function unpacks all the VLCs for either the Y plane or both
848 * C planes, and is called for DC coefficients or different AC coefficient
849 * levels (since different coefficient types require different VLC tables.
850 *
851 * This function returns a residual eob run. E.g, if a particular token gave
852 * instructions to EOB the next 5 fragments and there were only 2 fragments
853 * left in the current fragment range, 3 would be returned so that it could
854 * be passed into the next call to this same function.
855 */
856 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
857 VLC *table, int coeff_index,
858 int plane,
859 int eob_run)
860 {
861 int i, j = 0;
862 int token;
863 int zero_run = 0;
864 DCTELEM coeff = 0;
865 int bits_to_get;
866 int blocks_ended;
867 int coeff_i = 0;
868 int num_coeffs = s->num_coded_frags[plane][coeff_index];
869 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
870
871 /* local references to structure members to avoid repeated deferences */
872 int *coded_fragment_list = s->coded_fragment_list[plane];
873 Vp3Fragment *all_fragments = s->all_fragments;
874 VLC_TYPE (*vlc_table)[2] = table->table;
875
876 if (num_coeffs < 0)
877 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
878
879 if (eob_run > num_coeffs) {
880 coeff_i = blocks_ended = num_coeffs;
881 eob_run -= num_coeffs;
882 } else {
883 coeff_i = blocks_ended = eob_run;
884 eob_run = 0;
885 }
886
887 // insert fake EOB token to cover the split between planes or zzi
888 if (blocks_ended)
889 dct_tokens[j++] = blocks_ended << 2;
890
891 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
892 /* decode a VLC into a token */
893 token = get_vlc2(gb, vlc_table, 11, 3);
894 /* use the token to get a zero run, a coefficient, and an eob run */
895 if ((unsigned) token <= 6U) {
896 eob_run = eob_run_base[token];
897 if (eob_run_get_bits[token])
898 eob_run += get_bits(gb, eob_run_get_bits[token]);
899
900 // record only the number of blocks ended in this plane,
901 // any spill will be recorded in the next plane.
902 if (eob_run > num_coeffs - coeff_i) {
903 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
904 blocks_ended += num_coeffs - coeff_i;
905 eob_run -= num_coeffs - coeff_i;
906 coeff_i = num_coeffs;
907 } else {
908 dct_tokens[j++] = TOKEN_EOB(eob_run);
909 blocks_ended += eob_run;
910 coeff_i += eob_run;
911 eob_run = 0;
912 }
913 } else if (token >= 0) {
914 bits_to_get = coeff_get_bits[token];
915 if (bits_to_get)
916 bits_to_get = get_bits(gb, bits_to_get);
917 coeff = coeff_tables[token][bits_to_get];
918
919 zero_run = zero_run_base[token];
920 if (zero_run_get_bits[token])
921 zero_run += get_bits(gb, zero_run_get_bits[token]);
922
923 if (zero_run) {
924 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
925 } else {
926 // Save DC into the fragment structure. DC prediction is
927 // done in raster order, so the actual DC can't be in with
928 // other tokens. We still need the token in dct_tokens[]
929 // however, or else the structure collapses on itself.
930 if (!coeff_index)
931 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
932
933 dct_tokens[j++] = TOKEN_COEFF(coeff);
934 }
935
936 if (coeff_index + zero_run > 64) {
937 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
938 " %d coeffs left\n", zero_run, 64-coeff_index);
939 zero_run = 64 - coeff_index;
940 }
941
942 // zero runs code multiple coefficients,
943 // so don't try to decode coeffs for those higher levels
944 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
945 s->num_coded_frags[plane][i]--;
946 coeff_i++;
947 } else {
948 av_log(s->avctx, AV_LOG_ERROR,
949 "Invalid token %d\n", token);
950 return -1;
951 }
952 }
953
954 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
955 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
956
957 // decrement the number of blocks that have higher coeffecients for each
958 // EOB run at this level
959 if (blocks_ended)
960 for (i = coeff_index+1; i < 64; i++)
961 s->num_coded_frags[plane][i] -= blocks_ended;
962
963 // setup the next buffer
964 if (plane < 2)
965 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
966 else if (coeff_index < 63)
967 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
968
969 return eob_run;
970 }
971
972 static void reverse_dc_prediction(Vp3DecodeContext *s,
973 int first_fragment,
974 int fragment_width,
975 int fragment_height);
976 /*
977 * This function unpacks all of the DCT coefficient data from the
978 * bitstream.
979 */
980 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
981 {
982 int i;
983 int dc_y_table;
984 int dc_c_table;
985 int ac_y_table;
986 int ac_c_table;
987 int residual_eob_run = 0;
988 VLC *y_tables[64];
989 VLC *c_tables[64];
990
991 s->dct_tokens[0][0] = s->dct_tokens_base;
992
993 /* fetch the DC table indexes */
994 dc_y_table = get_bits(gb, 4);
995 dc_c_table = get_bits(gb, 4);
996
997 /* unpack the Y plane DC coefficients */
998 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
999 0, residual_eob_run);
1000 if (residual_eob_run < 0)
1001 return residual_eob_run;
1002
1003 /* reverse prediction of the Y-plane DC coefficients */
1004 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1005
1006 /* unpack the C plane DC coefficients */
1007 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1008 1, residual_eob_run);
1009 if (residual_eob_run < 0)
1010 return residual_eob_run;
1011 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1012 2, residual_eob_run);
1013 if (residual_eob_run < 0)
1014 return residual_eob_run;
1015
1016 /* reverse prediction of the C-plane DC coefficients */
1017 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1018 {
1019 reverse_dc_prediction(s, s->fragment_start[1],
1020 s->fragment_width[1], s->fragment_height[1]);
1021 reverse_dc_prediction(s, s->fragment_start[2],
1022 s->fragment_width[1], s->fragment_height[1]);
1023 }
1024
1025 /* fetch the AC table indexes */
1026 ac_y_table = get_bits(gb, 4);
1027 ac_c_table = get_bits(gb, 4);
1028
1029 /* build tables of AC VLC tables */
1030 for (i = 1; i <= 5; i++) {
1031 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1032 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1033 }
1034 for (i = 6; i <= 14; i++) {
1035 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1036 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1037 }
1038 for (i = 15; i <= 27; i++) {
1039 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1040 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1041 }
1042 for (i = 28; i <= 63; i++) {
1043 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1044 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1045 }
1046
1047 /* decode all AC coefficents */
1048 for (i = 1; i <= 63; i++) {
1049 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1050 0, residual_eob_run);
1051 if (residual_eob_run < 0)
1052 return residual_eob_run;
1053
1054 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1055 1, residual_eob_run);
1056 if (residual_eob_run < 0)
1057 return residual_eob_run;
1058 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1059 2, residual_eob_run);
1060 if (residual_eob_run < 0)
1061 return residual_eob_run;
1062 }
1063
1064 return 0;
1065 }
1066
1067 /*
1068 * This function reverses the DC prediction for each coded fragment in
1069 * the frame. Much of this function is adapted directly from the original
1070 * VP3 source code.
1071 */
1072 #define COMPATIBLE_FRAME(x) \
1073 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1074 #define DC_COEFF(u) s->all_fragments[u].dc
1075
1076 static void reverse_dc_prediction(Vp3DecodeContext *s,
1077 int first_fragment,
1078 int fragment_width,
1079 int fragment_height)
1080 {
1081
1082 #define PUL 8
1083 #define PU 4
1084 #define PUR 2
1085 #define PL 1
1086
1087 int x, y;
1088 int i = first_fragment;
1089
1090 int predicted_dc;
1091
1092 /* DC values for the left, up-left, up, and up-right fragments */
1093 int vl, vul, vu, vur;
1094
1095 /* indexes for the left, up-left, up, and up-right fragments */
1096 int l, ul, u, ur;
1097
1098 /*
1099 * The 6 fields mean:
1100 * 0: up-left multiplier
1101 * 1: up multiplier
1102 * 2: up-right multiplier
1103 * 3: left multiplier
1104 */
1105 static const int predictor_transform[16][4] = {
1106 { 0, 0, 0, 0},
1107 { 0, 0, 0,128}, // PL
1108 { 0, 0,128, 0}, // PUR
1109 { 0, 0, 53, 75}, // PUR|PL
1110 { 0,128, 0, 0}, // PU
1111 { 0, 64, 0, 64}, // PU|PL
1112 { 0,128, 0, 0}, // PU|PUR
1113 { 0, 0, 53, 75}, // PU|PUR|PL
1114 {128, 0, 0, 0}, // PUL
1115 { 0, 0, 0,128}, // PUL|PL
1116 { 64, 0, 64, 0}, // PUL|PUR
1117 { 0, 0, 53, 75}, // PUL|PUR|PL
1118 { 0,128, 0, 0}, // PUL|PU
1119 {-104,116, 0,116}, // PUL|PU|PL
1120 { 24, 80, 24, 0}, // PUL|PU|PUR
1121 {-104,116, 0,116} // PUL|PU|PUR|PL
1122 };
1123
1124 /* This table shows which types of blocks can use other blocks for
1125 * prediction. For example, INTRA is the only mode in this table to
1126 * have a frame number of 0. That means INTRA blocks can only predict
1127 * from other INTRA blocks. There are 2 golden frame coding types;
1128 * blocks encoding in these modes can only predict from other blocks
1129 * that were encoded with these 1 of these 2 modes. */
1130 static const unsigned char compatible_frame[9] = {
1131 1, /* MODE_INTER_NO_MV */
1132 0, /* MODE_INTRA */
1133 1, /* MODE_INTER_PLUS_MV */
1134 1, /* MODE_INTER_LAST_MV */
1135 1, /* MODE_INTER_PRIOR_MV */
1136 2, /* MODE_USING_GOLDEN */
1137 2, /* MODE_GOLDEN_MV */
1138 1, /* MODE_INTER_FOUR_MV */
1139 3 /* MODE_COPY */
1140 };
1141 int current_frame_type;
1142
1143 /* there is a last DC predictor for each of the 3 frame types */
1144 short last_dc[3];
1145
1146 int transform = 0;
1147
1148 vul = vu = vur = vl = 0;
1149 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1150
1151 /* for each fragment row... */
1152 for (y = 0; y < fragment_height; y++) {
1153
1154 /* for each fragment in a row... */
1155 for (x = 0; x < fragment_width; x++, i++) {
1156
1157 /* reverse prediction if this block was coded */
1158 if (s->all_fragments[i].coding_method != MODE_COPY) {
1159
1160 current_frame_type =
1161 compatible_frame[s->all_fragments[i].coding_method];
1162
1163 transform= 0;
1164 if(x){
1165 l= i-1;
1166 vl = DC_COEFF(l);
1167 if(COMPATIBLE_FRAME(l))
1168 transform |= PL;
1169 }
1170 if(y){
1171 u= i-fragment_width;
1172 vu = DC_COEFF(u);
1173 if(COMPATIBLE_FRAME(u))
1174 transform |= PU;
1175 if(x){
1176 ul= i-fragment_width-1;
1177 vul = DC_COEFF(ul);
1178 if(COMPATIBLE_FRAME(ul))
1179 transform |= PUL;
1180 }
1181 if(x + 1 < fragment_width){
1182 ur= i-fragment_width+1;
1183 vur = DC_COEFF(ur);
1184 if(COMPATIBLE_FRAME(ur))
1185 transform |= PUR;
1186 }
1187 }
1188
1189 if (transform == 0) {
1190
1191 /* if there were no fragments to predict from, use last
1192 * DC saved */
1193 predicted_dc = last_dc[current_frame_type];
1194 } else {
1195
1196 /* apply the appropriate predictor transform */
1197 predicted_dc =
1198 (predictor_transform[transform][0] * vul) +
1199 (predictor_transform[transform][1] * vu) +
1200 (predictor_transform[transform][2] * vur) +
1201 (predictor_transform[transform][3] * vl);
1202
1203 predicted_dc /= 128;
1204
1205 /* check for outranging on the [ul u l] and
1206 * [ul u ur l] predictors */
1207 if ((transform == 15) || (transform == 13)) {
1208 if (FFABS(predicted_dc - vu) > 128)
1209 predicted_dc = vu;
1210 else if (FFABS(predicted_dc - vl) > 128)
1211 predicted_dc = vl;
1212 else if (FFABS(predicted_dc - vul) > 128)
1213 predicted_dc = vul;
1214 }
1215 }
1216
1217 /* at long last, apply the predictor */
1218 DC_COEFF(i) += predicted_dc;
1219 /* save the DC */
1220 last_dc[current_frame_type] = DC_COEFF(i);
1221 }
1222 }
1223 }
1224 }
1225
1226 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1227 {
1228 int x, y;
1229 int *bounding_values= s->bounding_values_array+127;
1230
1231 int width = s->fragment_width[!!plane];
1232 int height = s->fragment_height[!!plane];
1233 int fragment = s->fragment_start [plane] + ystart * width;
1234 int stride = s->current_frame.linesize[plane];
1235 uint8_t *plane_data = s->current_frame.data [plane];
1236 if (!s->flipped_image) stride = -stride;
1237 plane_data += s->data_offset[plane] + 8*ystart*stride;
1238
1239 for (y = ystart; y < yend; y++) {
1240
1241 for (x = 0; x < width; x++) {
1242 /* This code basically just deblocks on the edges of coded blocks.
1243 * However, it has to be much more complicated because of the
1244 * braindamaged deblock ordering used in VP3/Theora. Order matters
1245 * because some pixels get filtered twice. */
1246 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1247 {
1248 /* do not perform left edge filter for left columns frags */
1249 if (x > 0) {
1250 s->dsp.vp3_h_loop_filter(
1251 plane_data + 8*x,
1252 stride, bounding_values);
1253 }
1254
1255 /* do not perform top edge filter for top row fragments */
1256 if (y > 0) {
1257 s->dsp.vp3_v_loop_filter(
1258 plane_data + 8*x,
1259 stride, bounding_values);
1260 }
1261
1262 /* do not perform right edge filter for right column
1263 * fragments or if right fragment neighbor is also coded
1264 * in this frame (it will be filtered in next iteration) */
1265 if ((x < width - 1) &&
1266 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1267 s->dsp.vp3_h_loop_filter(
1268 plane_data + 8*x + 8,
1269 stride, bounding_values);
1270 }
1271
1272 /* do not perform bottom edge filter for bottom row
1273 * fragments or if bottom fragment neighbor is also coded
1274 * in this frame (it will be filtered in the next row) */
1275 if ((y < height - 1) &&
1276 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1277 s->dsp.vp3_v_loop_filter(
1278 plane_data + 8*x + 8*stride,
1279 stride, bounding_values);
1280 }
1281 }
1282
1283 fragment++;
1284 }
1285 plane_data += 8*stride;
1286 }
1287 }
1288
1289 /**
1290 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1291 * for the next block in coding order
1292 */
1293 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1294 int plane, int inter, DCTELEM block[64])
1295 {
1296 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1297 uint8_t *perm = s->scantable.permutated;
1298 int i = 0;
1299
1300 do {
1301 int token = *s->dct_tokens[plane][i];
1302 switch (token & 3) {
1303 case 0: // EOB
1304 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1305 s->dct_tokens[plane][i]++;
1306 else
1307 *s->dct_tokens[plane][i] = token & ~3;
1308 goto end;
1309 case 1: // zero run
1310 s->dct_tokens[plane][i]++;
1311 i += (token >> 2) & 0x7f;
1312 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1313 i++;
1314 break;
1315 case 2: // coeff
1316 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1317 s->dct_tokens[plane][i++]++;
1318 break;
1319 default: // shouldn't happen
1320 return i;
1321 }
1322 } while (i < 64);
1323 end:
1324 // the actual DC+prediction is in the fragment structure
1325 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1326 return i;
1327 }
1328
1329 /**
1330 * called when all pixels up to row y are complete
1331 */
1332 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1333 {
1334 int h, cy, i;
1335 int offset[AV_NUM_DATA_POINTERS];
1336
1337 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1338 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1339
1340 // At the end of the frame, report INT_MAX instead of the height of the frame.
1341 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1342 // they don't have to clip their values.
1343 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1344 }
1345
1346 if(s->avctx->draw_horiz_band==NULL)
1347 return;
1348
1349 h= y - s->last_slice_end;
1350 s->last_slice_end= y;
1351 y -= h;
1352
1353 if (!s->flipped_image) {
1354 y = s->avctx->height - y - h;
1355 }
1356
1357 cy = y >> s->chroma_y_shift;
1358 offset[0] = s->current_frame.linesize[0]*y;
1359 offset[1] = s->current_frame.linesize[1]*cy;
1360 offset[2] = s->current_frame.linesize[2]*cy;
1361 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1362 offset[i] = 0;
1363
1364 emms_c();
1365 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1366 }
1367
1368 /**
1369 * Wait for the reference frame of the current fragment.
1370 * The progress value is in luma pixel rows.
1371 */
1372 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1373 {
1374 AVFrame *ref_frame;
1375 int ref_row;
1376 int border = motion_y&1;
1377
1378 if (fragment->coding_method == MODE_USING_GOLDEN ||
1379 fragment->coding_method == MODE_GOLDEN_MV)
1380 ref_frame = &s->golden_frame;
1381 else
1382 ref_frame = &s->last_frame;
1383
1384 ref_row = y + (motion_y>>1);
1385 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1386
1387 ff_thread_await_progress(ref_frame, ref_row, 0);
1388 }
1389
1390 /*
1391 * Perform the final rendering for a particular slice of data.
1392 * The slice number ranges from 0..(c_superblock_height - 1).
1393 */
1394 static void render_slice(Vp3DecodeContext *s, int slice)
1395 {
1396 int x, y, i, j, fragment;
1397 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1398 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1399 int motion_halfpel_index;
1400 uint8_t *motion_source;
1401 int plane, first_pixel;
1402
1403 if (slice >= s->c_superblock_height)
1404 return;
1405
1406 for (plane = 0; plane < 3; plane++) {
1407 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1408 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1409 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1410 int stride = s->current_frame.linesize[plane];
1411 int plane_width = s->width >> (plane && s->chroma_x_shift);
1412 int plane_height = s->height >> (plane && s->chroma_y_shift);
1413 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1414
1415 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1416 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1417 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1418
1419 int fragment_width = s->fragment_width[!!plane];
1420 int fragment_height = s->fragment_height[!!plane];
1421 int fragment_start = s->fragment_start[plane];
1422 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1423
1424 if (!s->flipped_image) stride = -stride;
1425 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1426 continue;
1427
1428 /* for each superblock row in the slice (both of them)... */
1429 for (; sb_y < slice_height; sb_y++) {
1430
1431 /* for each superblock in a row... */
1432 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1433
1434 /* for each block in a superblock... */
1435 for (j = 0; j < 16; j++) {
1436 x = 4*sb_x + hilbert_offset[j][0];
1437 y = 4*sb_y + hilbert_offset[j][1];
1438 fragment = y*fragment_width + x;
1439
1440 i = fragment_start + fragment;
1441
1442 // bounds check
1443 if (x >= fragment_width || y >= fragment_height)
1444 continue;
1445
1446 first_pixel = 8*y*stride + 8*x;
1447
1448 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1449 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1450
1451 /* transform if this block was coded */
1452 if (s->all_fragments[i].coding_method != MODE_COPY) {
1453 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1454 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1455 motion_source= golden_plane;
1456 else
1457 motion_source= last_plane;
1458
1459 motion_source += first_pixel;
1460 motion_halfpel_index = 0;
1461
1462 /* sort out the motion vector if this fragment is coded
1463 * using a motion vector method */
1464 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1465 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1466 int src_x, src_y;
1467 motion_x = motion_val[fragment][0];
1468 motion_y = motion_val[fragment][1];
1469
1470 src_x= (motion_x>>1) + 8*x;
1471 src_y= (motion_y>>1) + 8*y;
1472
1473 motion_halfpel_index = motion_x & 0x01;
1474 motion_source += (motion_x >> 1);
1475
1476 motion_halfpel_index |= (motion_y & 0x01) << 1;
1477 motion_source += ((motion_y >> 1) * stride);
1478
1479 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1480 uint8_t *temp= s->edge_emu_buffer;
1481 if(stride<0) temp -= 8*stride;
1482
1483 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1484 motion_source= temp;
1485 }
1486 }
1487
1488
1489 /* first, take care of copying a block from either the
1490 * previous or the golden frame */
1491 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1492 /* Note, it is possible to implement all MC cases with
1493 put_no_rnd_pixels_l2 which would look more like the
1494 VP3 source but this would be slower as
1495 put_no_rnd_pixels_tab is better optimzed */
1496 if(motion_halfpel_index != 3){
1497 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1498 output_plane + first_pixel,
1499 motion_source, stride, 8);
1500 }else{
1501 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1502 s->dsp.put_no_rnd_pixels_l2[1](
1503 output_plane + first_pixel,
1504 motion_source - d,
1505 motion_source + stride + 1 + d,
1506 stride, 8);
1507 }
1508 }
1509
1510 s->dsp.clear_block(block);
1511
1512 /* invert DCT and place (or add) in final output */
1513
1514 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1515 vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1516 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1517 block[0] += 128<<3;
1518 s->dsp.idct_put(
1519 output_plane + first_pixel,
1520 stride,
1521 block);
1522 } else {
1523 if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
1524 s->dsp.idct_add(
1525 output_plane + first_pixel,
1526 stride,
1527 block);
1528 } else {
1529 s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1530 }
1531 }
1532 } else {
1533
1534 /* copy directly from the previous frame */
1535 s->dsp.put_pixels_tab[1][0](
1536 output_plane + first_pixel,
1537 last_plane + first_pixel,
1538 stride, 8);
1539
1540 }
1541 }
1542 }
1543
1544 // Filter up to the last row in the superblock row
1545 if (!s->skip_loop_filter)
1546 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1547 }
1548 }
1549
1550 /* this looks like a good place for slice dispatch... */
1551 /* algorithm:
1552 * if (slice == s->macroblock_height - 1)
1553 * dispatch (both last slice & 2nd-to-last slice);
1554 * else if (slice > 0)
1555 * dispatch (slice - 1);
1556 */
1557
1558 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1559 }
1560
1561 /// Allocate tables for per-frame data in Vp3DecodeContext
1562 static av_cold int allocate_tables(AVCodecContext *avctx)
1563 {
1564 Vp3DecodeContext *s = avctx->priv_data;
1565 int y_fragment_count, c_fragment_count;
1566
1567 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1568 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1569
1570 s->superblock_coding = av_malloc(s->superblock_count);
1571 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1572 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1573 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1574 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1575 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1576
1577 /* work out the block mapping tables */
1578 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1579 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1580
1581 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1582 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1583 !s->motion_val[0] || !s->motion_val[1]) {
1584 vp3_decode_end(avctx);
1585 return -1;
1586 }
1587
1588 init_block_mapping(s);
1589
1590 return 0;
1591 }
1592
1593 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1594 {
1595 Vp3DecodeContext *s = avctx->priv_data;
1596 int i, inter, plane;
1597 int c_width;
1598 int c_height;
1599 int y_fragment_count, c_fragment_count;
1600
1601 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1602 s->version = 0;
1603 else
1604 s->version = 1;
1605
1606 s->avctx = avctx;
1607 s->width = FFALIGN(avctx->width, 16);
1608 s->height = FFALIGN(avctx->height, 16);
1609 if (avctx->pix_fmt == PIX_FMT_NONE)
1610 avctx->pix_fmt = PIX_FMT_YUV420P;
1611 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1612 if(avctx->idct_algo==FF_IDCT_AUTO)
1613 avctx->idct_algo=FF_IDCT_VP3;
1614 dsputil_init(&s->dsp, avctx);
1615
1616 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1617
1618 /* initialize to an impossible value which will force a recalculation
1619 * in the first frame decode */
1620 for (i = 0; i < 3; i++)
1621 s->qps[i] = -1;
1622
1623 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1624
1625 s->y_superblock_width = (s->width + 31) / 32;
1626 s->y_superblock_height = (s->height + 31) / 32;
1627 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1628
1629 /* work out the dimensions for the C planes */
1630 c_width = s->width >> s->chroma_x_shift;
1631 c_height = s->height >> s->chroma_y_shift;
1632 s->c_superblock_width = (c_width + 31) / 32;
1633 s->c_superblock_height = (c_height + 31) / 32;
1634 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1635
1636 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1637 s->u_superblock_start = s->y_superblock_count;
1638 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1639
1640 s->macroblock_width = (s->width + 15) / 16;
1641 s->macroblock_height = (s->height + 15) / 16;
1642 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1643
1644 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1645 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1646 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1647 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1648
1649 /* fragment count covers all 8x8 blocks for all 3 planes */
1650 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1651 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1652 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1653 s->fragment_start[1] = y_fragment_count;
1654 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1655
1656 if (!s->theora_tables)
1657 {
1658 for (i = 0; i < 64; i++) {
1659 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1660 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1661 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1662 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1663 s->base_matrix[2][i] = vp31_inter_dequant[i];
1664 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1665 }
1666
1667 for(inter=0; inter<2; inter++){
1668 for(plane=0; plane<3; plane++){
1669 s->qr_count[inter][plane]= 1;
1670 s->qr_size [inter][plane][0]= 63;
1671 s->qr_base [inter][plane][0]=
1672 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1673 }
1674 }
1675
1676 /* init VLC tables */
1677 for (i = 0; i < 16; i++) {
1678
1679 /* DC histograms */
1680 init_vlc(&s->dc_vlc[i], 11, 32,
1681 &dc_bias[i][0][1], 4, 2,
1682 &dc_bias[i][0][0], 4, 2, 0);
1683
1684 /* group 1 AC histograms */
1685 init_vlc(&s->ac_vlc_1[i], 11, 32,
1686 &ac_bias_0[i][0][1], 4, 2,
1687 &ac_bias_0[i][0][0], 4, 2, 0);
1688
1689 /* group 2 AC histograms */
1690 init_vlc(&s->ac_vlc_2[i], 11, 32,
1691 &ac_bias_1[i][0][1], 4, 2,
1692 &ac_bias_1[i][0][0], 4, 2, 0);
1693
1694 /* group 3 AC histograms */
1695 init_vlc(&s->ac_vlc_3[i], 11, 32,
1696 &ac_bias_2[i][0][1], 4, 2,
1697 &ac_bias_2[i][0][0], 4, 2, 0);
1698
1699 /* group 4 AC histograms */
1700 init_vlc(&s->ac_vlc_4[i], 11, 32,
1701 &ac_bias_3[i][0][1], 4, 2,
1702 &ac_bias_3[i][0][0], 4, 2, 0);
1703 }
1704 } else {
1705
1706 for (i = 0; i < 16; i++) {
1707 /* DC histograms */
1708 if (init_vlc(&s->dc_vlc[i], 11, 32,
1709 &s->huffman_table[i][0][1], 8, 4,
1710 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1711 goto vlc_fail;
1712
1713 /* group 1 AC histograms */
1714 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1715 &s->huffman_table[i+16][0][1], 8, 4,
1716 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1717 goto vlc_fail;
1718
1719 /* group 2 AC histograms */
1720 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1721 &s->huffman_table[i+16*2][0][1], 8, 4,
1722 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1723 goto vlc_fail;
1724
1725 /* group 3 AC histograms */
1726 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1727 &s->huffman_table[i+16*3][0][1], 8, 4,
1728 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1729 goto vlc_fail;
1730
1731 /* group 4 AC histograms */
1732 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1733 &s->huffman_table[i+16*4][0][1], 8, 4,
1734 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1735 goto vlc_fail;
1736 }
1737 }
1738
1739 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1740 &superblock_run_length_vlc_table[0][1], 4, 2,
1741 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1742
1743 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1744 &fragment_run_length_vlc_table[0][1], 4, 2,
1745 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1746
1747 init_vlc(&s->mode_code_vlc, 3, 8,
1748 &mode_code_vlc_table[0][1], 2, 1,
1749 &mode_code_vlc_table[0][0], 2, 1, 0);
1750
1751 init_vlc(&s->motion_vector_vlc, 6, 63,
1752 &motion_vector_vlc_table[0][1], 2, 1,
1753 &motion_vector_vlc_table[0][0], 2, 1, 0);
1754
1755 for (i = 0; i < 3; i++) {
1756 s->current_frame.data[i] = NULL;
1757 s->last_frame.data[i] = NULL;
1758 s->golden_frame.data[i] = NULL;
1759 }
1760
1761 return allocate_tables(avctx);
1762
1763 vlc_fail:
1764 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1765 return -1;
1766 }
1767
1768 /// Release and shuffle frames after decode finishes
1769 static void update_frames(AVCodecContext *avctx)
1770 {
1771 Vp3DecodeContext *s = avctx->priv_data;
1772
1773 /* release the last frame, if it is allocated and if it is not the
1774 * golden frame */
1775 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1776 ff_thread_release_buffer(avctx, &s->last_frame);
1777
1778 /* shuffle frames (last = current) */
1779 s->last_frame= s->current_frame;
1780
1781 if (s->keyframe) {
1782 if (s->golden_frame.data[0])
1783 ff_thread_release_buffer(avctx, &s->golden_frame);
1784 s->golden_frame = s->current_frame;
1785 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1786 }
1787
1788 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1789 }
1790
1791 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1792 {
1793 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1794 int qps_changed = 0, i, err;
1795
1796 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1797
1798 if (!s1->current_frame.data[0]
1799 ||s->width != s1->width
1800 ||s->height!= s1->height) {
1801 if (s != s1)
1802 copy_fields(s, s1, golden_frame, current_frame);
1803 return -1;
1804 }
1805
1806 if (s != s1) {
1807 // init tables if the first frame hasn't been decoded
1808 if (!s->current_frame.data[0]) {
1809 int y_fragment_count, c_fragment_count;
1810 s->avctx = dst;
1811 err = allocate_tables(dst);
1812 if (err)
1813 return err;
1814 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1815 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1816 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1817 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1818 }
1819
1820 // copy previous frame data
1821 copy_fields(s, s1, golden_frame, dsp);
1822
1823 // copy qscale data if necessary
1824 for (i = 0; i < 3; i++) {
1825 if (s->qps[i] != s1->qps[1]) {
1826 qps_changed = 1;
1827 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1828 }
1829 }
1830
1831 if (s->qps[0] != s1->qps[0])
1832 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1833
1834 if (qps_changed)
1835 copy_fields(s, s1, qps, superblock_count);
1836 #undef copy_fields
1837 }
1838
1839 update_frames(dst);
1840
1841 return 0;
1842 }
1843
1844 static int vp3_decode_frame(AVCodecContext *avctx,
1845 void *data, int *data_size,
1846 AVPacket *avpkt)
1847 {
1848 const uint8_t *buf = avpkt->data;
1849 int buf_size = avpkt->size;
1850 Vp3DecodeContext *s = avctx->priv_data;
1851 GetBitContext gb;
1852 int i;
1853
1854 init_get_bits(&gb, buf, buf_size * 8);
1855
1856 if (s->theora && get_bits1(&gb))
1857 {
1858 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1859 return -1;
1860 }
1861
1862 s->keyframe = !get_bits1(&gb);
1863 if (!s->theora)
1864 skip_bits(&gb, 1);
1865 for (i = 0; i < 3; i++)
1866 s->last_qps[i] = s->qps[i];
1867
1868 s->nqps=0;
1869 do{
1870 s->qps[s->nqps++]= get_bits(&gb, 6);
1871 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1872 for (i = s->nqps; i < 3; i++)
1873 s->qps[i] = -1;
1874
1875 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1876 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1877 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1878
1879 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1880 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1881
1882 if (s->qps[0] != s->last_qps[0])
1883 init_loop_filter(s);
1884
1885 for (i = 0; i < s->nqps; i++)
1886 // reinit all dequantizers if the first one changed, because
1887 // the DC of the first quantizer must be used for all matrices
1888 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1889 init_dequantizer(s, i);
1890
1891 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1892 return buf_size;
1893
1894 s->current_frame.reference = 3;
1895 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1896 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1897 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1898 goto error;
1899 }
1900
1901 if (!s->edge_emu_buffer)
1902 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1903
1904 if (s->keyframe) {
1905 if (!s->theora)
1906 {
1907 skip_bits(&gb, 4); /* width code */
1908 skip_bits(&gb, 4); /* height code */
1909 if (s->version)
1910 {
1911 s->version = get_bits(&gb, 5);
1912 if (avctx->frame_number == 0)
1913 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1914 }
1915 }
1916 if (s->version || s->theora)
1917 {
1918 if (get_bits1(&gb))
1919 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1920 skip_bits(&gb, 2); /* reserved? */
1921 }
1922 } else {
1923 if (!s->golden_frame.data[0]) {
1924 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1925
1926 s->golden_frame.reference = 3;
1927 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1928 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1929 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1930 goto error;
1931 }
1932 s->last_frame = s->golden_frame;
1933 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1934 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
1935 }
1936 }
1937
1938 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1939 ff_thread_finish_setup(avctx);
1940
1941 if (unpack_superblocks(s, &gb)){
1942 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1943 goto error;
1944 }
1945 if (unpack_modes(s, &gb)){
1946 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1947 goto error;
1948 }
1949 if (unpack_vectors(s, &gb)){
1950 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1951 goto error;
1952 }
1953 if (unpack_block_qpis(s, &gb)){
1954 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1955 goto error;
1956 }
1957 if (unpack_dct_coeffs(s, &gb)){
1958 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1959 goto error;
1960 }
1961
1962 for (i = 0; i < 3; i++) {
1963 int height = s->height >> (i && s->chroma_y_shift);
1964 if (s->flipped_image)
1965 s->data_offset[i] = 0;
1966 else
1967 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1968 }
1969
1970 s->last_slice_end = 0;
1971 for (i = 0; i < s->c_superblock_height; i++)
1972 render_slice(s, i);
1973
1974 // filter the last row
1975 for (i = 0; i < 3; i++) {
1976 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1977 apply_loop_filter(s, i, row, row+1);
1978 }
1979 vp3_draw_horiz_band(s, s->avctx->height);
1980
1981 *data_size=sizeof(AVFrame);
1982 *(AVFrame*)data= s->current_frame;
1983
1984 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1985 update_frames(avctx);
1986
1987 return buf_size;
1988
1989 error:
1990 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
1991
1992 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1993 avctx->release_buffer(avctx, &s->current_frame);
1994
1995 return -1;
1996 }
1997
1998 static av_cold int vp3_decode_end(AVCodecContext *avctx)
1999 {
2000 Vp3DecodeContext *s = avctx->priv_data;
2001 int i;
2002
2003 av_free(s->superblock_coding);
2004 av_free(s->all_fragments);
2005 av_free(s->coded_fragment_list[0]);
2006 av_free(s->dct_tokens_base);
2007 av_free(s->superblock_fragments);
2008 av_free(s->macroblock_coding);
2009 av_free(s->motion_val[0]);
2010 av_free(s->motion_val[1]);
2011 av_free(s->edge_emu_buffer);
2012
2013 if (avctx->internal->is_copy)
2014 return 0;
2015
2016 for (i = 0; i < 16; i++) {
2017 free_vlc(&s->dc_vlc[i]);
2018 free_vlc(&s->ac_vlc_1[i]);
2019 free_vlc(&s->ac_vlc_2[i]);
2020 free_vlc(&s->ac_vlc_3[i]);
2021 free_vlc(&s->ac_vlc_4[i]);
2022 }
2023
2024 free_vlc(&s->superblock_run_length_vlc);
2025 free_vlc(&s->fragment_run_length_vlc);
2026 free_vlc(&s->mode_code_vlc);
2027 free_vlc(&s->motion_vector_vlc);
2028
2029 /* release all frames */
2030 vp3_decode_flush(avctx);
2031
2032 return 0;
2033 }
2034
2035 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2036 {
2037 Vp3DecodeContext *s = avctx->priv_data;
2038
2039 if (get_bits1(gb)) {
2040 int token;
2041 if (s->entries >= 32) { /* overflow */
2042 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2043 return -1;
2044 }
2045 token = get_bits(gb, 5);
2046 //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);
2047 s->huffman_table[s->hti][token][0] = s->hbits;
2048 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2049 s->entries++;
2050 }
2051 else {
2052 if (s->huff_code_size >= 32) {/* overflow */
2053 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2054 return -1;
2055 }
2056 s->huff_code_size++;
2057 s->hbits <<= 1;
2058 if (read_huffman_tree(avctx, gb))
2059 return -1;
2060 s->hbits |= 1;
2061 if (read_huffman_tree(avctx, gb))
2062 return -1;
2063 s->hbits >>= 1;
2064 s->huff_code_size--;
2065 }
2066 return 0;
2067 }
2068
2069 #if CONFIG_THEORA_DECODER
2070 static const enum PixelFormat theora_pix_fmts[4] = {
2071 PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
2072 };
2073
2074 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2075 {
2076 Vp3DecodeContext *s = avctx->priv_data;
2077 int visible_width, visible_height, colorspace;
2078 int offset_x = 0, offset_y = 0;
2079 AVRational fps, aspect;
2080
2081 s->theora = get_bits_long(gb, 24);
2082 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2083
2084 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2085 /* but previous versions have the image flipped relative to vp3 */
2086 if (s->theora < 0x030200)
2087 {
2088 s->flipped_image = 1;
2089 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2090 }
2091
2092 visible_width = s->width = get_bits(gb, 16) << 4;
2093 visible_height = s->height = get_bits(gb, 16) << 4;
2094
2095 if(av_image_check_size(s->width, s->height, 0, avctx)){
2096 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2097 s->width= s->height= 0;
2098 return -1;
2099 }
2100
2101 if (s->theora >= 0x030200) {
2102 visible_width = get_bits_long(gb, 24);
2103 visible_height = get_bits_long(gb, 24);
2104
2105 offset_x = get_bits(gb, 8); /* offset x */
2106 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2107 }
2108
2109 fps.num = get_bits_long(gb, 32);
2110 fps.den = get_bits_long(gb, 32);
2111 if (fps.num && fps.den) {
2112 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2113 fps.den, fps.num, 1<<30);
2114 }
2115
2116 aspect.num = get_bits_long(gb, 24);
2117 aspect.den = get_bits_long(gb, 24);
2118 if (aspect.num && aspect.den) {
2119 av_reduce(&avctx->sample_aspect_ratio.num,
2120 &avctx->sample_aspect_ratio.den,
2121 aspect.num, aspect.den, 1<<30);
2122 }
2123
2124 if (s->theora < 0x030200)
2125 skip_bits(gb, 5); /* keyframe frequency force */
2126 colorspace = get_bits(gb, 8);
2127 skip_bits(gb, 24); /* bitrate */
2128
2129 skip_bits(gb, 6); /* quality hint */
2130
2131 if (s->theora >= 0x030200)
2132 {
2133 skip_bits(gb, 5); /* keyframe frequency force */
2134 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2135 skip_bits(gb, 3); /* reserved */
2136 }
2137
2138 // align_get_bits(gb);
2139
2140 if ( visible_width <= s->width && visible_width > s->width-16
2141 && visible_height <= s->height && visible_height > s->height-16
2142 && !offset_x && (offset_y == s->height - visible_height))
2143 avcodec_set_dimensions(avctx, visible_width, visible_height);
2144 else
2145 avcodec_set_dimensions(avctx, s->width, s->height);
2146
2147 if (colorspace == 1) {
2148 avctx->color_primaries = AVCOL_PRI_BT470M;
2149 } else if (colorspace == 2) {
2150 avctx->color_primaries = AVCOL_PRI_BT470BG;
2151 }
2152 if (colorspace == 1 || colorspace == 2) {
2153 avctx->colorspace = AVCOL_SPC_BT470BG;
2154 avctx->color_trc = AVCOL_TRC_BT709;
2155 }
2156
2157 return 0;
2158 }
2159
2160 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2161 {
2162 Vp3DecodeContext *s = avctx->priv_data;
2163 int i, n, matrices, inter, plane;
2164
2165 if (s->theora >= 0x030200) {
2166 n = get_bits(gb, 3);
2167 /* loop filter limit values table */
2168 if (n)
2169 for (i = 0; i < 64; i++)
2170 s->filter_limit_values[i] = get_bits(gb, n);
2171 }
2172
2173 if (s->theora >= 0x030200)
2174 n = get_bits(gb, 4) + 1;
2175 else
2176 n = 16;
2177 /* quality threshold table */
2178 for (i = 0; i < 64; i++)
2179 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2180
2181 if (s->theora >= 0x030200)
2182 n = get_bits(gb, 4) + 1;
2183 else
2184 n = 16;
2185 /* dc scale factor table */
2186 for (i = 0; i < 64; i++)
2187 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2188
2189 if (s->theora >= 0x030200)
2190 matrices = get_bits(gb, 9) + 1;
2191 else
2192 matrices = 3;
2193
2194 if(matrices > 384){
2195 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2196 return -1;
2197 }
2198
2199 for(n=0; n<matrices; n++){
2200 for (i = 0; i < 64; i++)
2201 s->base_matrix[n][i]= get_bits(gb, 8);
2202 }
2203
2204 for (inter = 0; inter <= 1; inter++) {
2205 for (plane = 0; plane <= 2; plane++) {
2206 int newqr= 1;
2207 if (inter || plane > 0)
2208 newqr = get_bits1(gb);
2209 if (!newqr) {
2210 int qtj, plj;
2211 if(inter && get_bits1(gb)){
2212 qtj = 0;
2213 plj = plane;
2214 }else{
2215 qtj= (3*inter + plane - 1) / 3;
2216 plj= (plane + 2) % 3;
2217 }
2218 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2219 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2220 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2221 } else {
2222 int qri= 0;
2223 int qi = 0;
2224
2225 for(;;){
2226 i= get_bits(gb, av_log2(matrices-1)+1);
2227 if(i>= matrices){
2228 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2229 return -1;
2230 }
2231 s->qr_base[inter][plane][qri]= i;
2232 if(qi >= 63)
2233 break;
2234 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2235 s->qr_size[inter][plane][qri++]= i;
2236 qi += i;
2237 }
2238
2239 if (qi > 63) {
2240 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2241 return -1;
2242 }
2243 s->qr_count[inter][plane]= qri;
2244 }
2245 }
2246 }
2247
2248 /* Huffman tables */
2249 for (s->hti = 0; s->hti < 80; s->hti++) {
2250 s->entries = 0;
2251 s->huff_code_size = 1;
2252 if (!get_bits1(gb)) {
2253 s->hbits = 0;
2254 if(read_huffman_tree(avctx, gb))
2255 return -1;
2256 s->hbits = 1;
2257 if(read_huffman_tree(avctx, gb))
2258 return -1;
2259 }
2260 }
2261
2262 s->theora_tables = 1;
2263
2264 return 0;
2265 }
2266
2267 static av_cold int theora_decode_init(AVCodecContext *avctx)
2268 {
2269 Vp3DecodeContext *s = avctx->priv_data;
2270 GetBitContext gb;
2271 int ptype;
2272 uint8_t *header_start[3];
2273 int header_len[3];
2274 int i;
2275
2276 s->theora = 1;
2277
2278 if (!avctx->extradata_size)
2279 {
2280 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2281 return -1;
2282 }
2283
2284 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2285 42, header_start, header_len) < 0) {
2286 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2287 return -1;
2288 }
2289
2290 for(i=0;i<3;i++) {
2291 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2292
2293 ptype = get_bits(&gb, 8);
2294
2295 if (!(ptype & 0x80))
2296 {
2297 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2298 // return -1;
2299 }
2300
2301 // FIXME: Check for this as well.
2302 skip_bits_long(&gb, 6*8); /* "theora" */
2303
2304 switch(ptype)
2305 {
2306 case 0x80:
2307 theora_decode_header(avctx, &gb);
2308 break;
2309 case 0x81:
2310 // FIXME: is this needed? it breaks sometimes
2311 // theora_decode_comments(avctx, gb);
2312 break;
2313 case 0x82:
2314 if (theora_decode_tables(avctx, &gb))
2315 return -1;
2316 break;
2317 default:
2318 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2319 break;
2320 }
2321 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2322 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2323 if (s->theora < 0x030200)
2324 break;
2325 }
2326
2327 return vp3_decode_init(avctx);
2328 }
2329
2330 static void vp3_decode_flush(AVCodecContext *avctx)
2331 {
2332 Vp3DecodeContext *s = avctx->priv_data;
2333
2334 if (s->golden_frame.data[0]) {
2335 if (s->golden_frame.data[0] == s->last_frame.data[0])
2336 memset(&s->last_frame, 0, sizeof(AVFrame));
2337 if (s->current_frame.data[0] == s->golden_frame.data[0])
2338 memset(&s->current_frame, 0, sizeof(AVFrame));
2339 ff_thread_release_buffer(avctx, &s->golden_frame);
2340 }
2341 if (s->last_frame.data[0]) {
2342 if (s->current_frame.data[0] == s->last_frame.data[0])
2343 memset(&s->current_frame, 0, sizeof(AVFrame));
2344 ff_thread_release_buffer(avctx, &s->last_frame);
2345 }
2346 if (s->current_frame.data[0])
2347 ff_thread_release_buffer(avctx, &s->current_frame);
2348 }
2349
2350 static int vp3_init_thread_copy(AVCodecContext *avctx)
2351 {
2352 Vp3DecodeContext *s = avctx->priv_data;
2353
2354 s->superblock_coding = NULL;
2355 s->all_fragments = NULL;
2356 s->coded_fragment_list[0] = NULL;
2357 s->dct_tokens_base = NULL;
2358 s->superblock_fragments = NULL;
2359 s->macroblock_coding = NULL;
2360 s->motion_val[0] = NULL;
2361 s->motion_val[1] = NULL;
2362 s->edge_emu_buffer = NULL;
2363
2364 return 0;
2365 }
2366
2367 AVCodec ff_theora_decoder = {
2368 .name = "theora",
2369 .type = AVMEDIA_TYPE_VIDEO,
2370 .id = CODEC_ID_THEORA,
2371 .priv_data_size = sizeof(Vp3DecodeContext),
2372 .init = theora_decode_init,
2373 .close = vp3_decode_end,
2374 .decode = vp3_decode_frame,
2375 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2376 .flush = vp3_decode_flush,
2377 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2378 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2379 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2380 };
2381 #endif
2382
2383 AVCodec ff_vp3_decoder = {
2384 .name = "vp3",
2385 .type = AVMEDIA_TYPE_VIDEO,
2386 .id = CODEC_ID_VP3,
2387 .priv_data_size = sizeof(Vp3DecodeContext),
2388 .init = vp3_decode_init,
2389 .close = vp3_decode_end,
2390 .decode = vp3_decode_frame,
2391 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2392 .flush = vp3_decode_flush,
2393 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2394 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2395 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2396 };