lavc: fix decode_frame() third parameter semantics for video decoders
[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 "vp3dsp.h"
44 #include "xiph.h"
45 #include "thread.h"
46
47 #define FRAGMENT_PIXELS 8
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 VP3DSPContext vp3dsp;
140 int flipped_image;
141 int last_slice_end;
142 int skip_loop_filter;
143
144 int qps[3];
145 int nqps;
146 int last_qps[3];
147
148 int superblock_count;
149 int y_superblock_width;
150 int y_superblock_height;
151 int y_superblock_count;
152 int c_superblock_width;
153 int c_superblock_height;
154 int c_superblock_count;
155 int u_superblock_start;
156 int v_superblock_start;
157 unsigned char *superblock_coding;
158
159 int macroblock_count;
160 int macroblock_width;
161 int macroblock_height;
162
163 int fragment_count;
164 int fragment_width[2];
165 int fragment_height[2];
166
167 Vp3Fragment *all_fragments;
168 int fragment_start[3];
169 int data_offset[3];
170
171 int8_t (*motion_val[2])[2];
172
173 ScanTable scantable;
174
175 /* tables */
176 uint16_t coded_dc_scale_factor[64];
177 uint32_t coded_ac_scale_factor[64];
178 uint8_t base_matrix[384][64];
179 uint8_t qr_count[2][3];
180 uint8_t qr_size [2][3][64];
181 uint16_t qr_base[2][3][64];
182
183 /**
184 * This is a list of all tokens in bitstream order. Reordering takes place
185 * by pulling from each level during IDCT. As a consequence, IDCT must be
186 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
187 * otherwise. The 32 different tokens with up to 12 bits of extradata are
188 * collapsed into 3 types, packed as follows:
189 * (from the low to high bits)
190 *
191 * 2 bits: type (0,1,2)
192 * 0: EOB run, 14 bits for run length (12 needed)
193 * 1: zero run, 7 bits for run length
194 * 7 bits for the next coefficient (3 needed)
195 * 2: coefficient, 14 bits (11 needed)
196 *
197 * Coefficients are signed, so are packed in the highest bits for automatic
198 * sign extension.
199 */
200 int16_t *dct_tokens[3][64];
201 int16_t *dct_tokens_base;
202 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
203 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
204 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
205
206 /**
207 * number of blocks that contain DCT coefficients at the given level or higher
208 */
209 int num_coded_frags[3][64];
210 int total_num_coded_frags;
211
212 /* this is a list of indexes into the all_fragments array indicating
213 * which of the fragments are coded */
214 int *coded_fragment_list[3];
215
216 VLC dc_vlc[16];
217 VLC ac_vlc_1[16];
218 VLC ac_vlc_2[16];
219 VLC ac_vlc_3[16];
220 VLC ac_vlc_4[16];
221
222 VLC superblock_run_length_vlc;
223 VLC fragment_run_length_vlc;
224 VLC mode_code_vlc;
225 VLC motion_vector_vlc;
226
227 /* these arrays need to be on 16-byte boundaries since SSE2 operations
228 * index into them */
229 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
230
231 /* This table contains superblock_count * 16 entries. Each set of 16
232 * numbers corresponds to the fragment indexes 0..15 of the superblock.
233 * An entry will be -1 to indicate that no entry corresponds to that
234 * index. */
235 int *superblock_fragments;
236
237 /* This is an array that indicates how a particular macroblock
238 * is coded. */
239 unsigned char *macroblock_coding;
240
241 uint8_t *edge_emu_buffer;
242
243 /* Huffman decode */
244 int hti;
245 unsigned int hbits;
246 int entries;
247 int huff_code_size;
248 uint32_t huffman_table[80][32][2];
249
250 uint8_t filter_limit_values[64];
251 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
252 } Vp3DecodeContext;
253
254 /************************************************************************
255 * VP3 specific functions
256 ************************************************************************/
257
258 static void vp3_decode_flush(AVCodecContext *avctx)
259 {
260 Vp3DecodeContext *s = avctx->priv_data;
261
262 if (s->golden_frame.data[0]) {
263 if (s->golden_frame.data[0] == s->last_frame.data[0])
264 memset(&s->last_frame, 0, sizeof(AVFrame));
265 if (s->current_frame.data[0] == s->golden_frame.data[0])
266 memset(&s->current_frame, 0, sizeof(AVFrame));
267 ff_thread_release_buffer(avctx, &s->golden_frame);
268 }
269 if (s->last_frame.data[0]) {
270 if (s->current_frame.data[0] == s->last_frame.data[0])
271 memset(&s->current_frame, 0, sizeof(AVFrame));
272 ff_thread_release_buffer(avctx, &s->last_frame);
273 }
274 if (s->current_frame.data[0])
275 ff_thread_release_buffer(avctx, &s->current_frame);
276 }
277
278 static av_cold int vp3_decode_end(AVCodecContext *avctx)
279 {
280 Vp3DecodeContext *s = avctx->priv_data;
281 int i;
282
283 av_free(s->superblock_coding);
284 av_free(s->all_fragments);
285 av_free(s->coded_fragment_list[0]);
286 av_free(s->dct_tokens_base);
287 av_free(s->superblock_fragments);
288 av_free(s->macroblock_coding);
289 av_free(s->motion_val[0]);
290 av_free(s->motion_val[1]);
291 av_free(s->edge_emu_buffer);
292
293 if (avctx->internal->is_copy)
294 return 0;
295
296 for (i = 0; i < 16; i++) {
297 ff_free_vlc(&s->dc_vlc[i]);
298 ff_free_vlc(&s->ac_vlc_1[i]);
299 ff_free_vlc(&s->ac_vlc_2[i]);
300 ff_free_vlc(&s->ac_vlc_3[i]);
301 ff_free_vlc(&s->ac_vlc_4[i]);
302 }
303
304 ff_free_vlc(&s->superblock_run_length_vlc);
305 ff_free_vlc(&s->fragment_run_length_vlc);
306 ff_free_vlc(&s->mode_code_vlc);
307 ff_free_vlc(&s->motion_vector_vlc);
308
309 /* release all frames */
310 vp3_decode_flush(avctx);
311
312 return 0;
313 }
314
315 /*
316 * This function sets up all of the various blocks mappings:
317 * superblocks <-> fragments, macroblocks <-> fragments,
318 * superblocks <-> macroblocks
319 *
320 * @return 0 is successful; returns 1 if *anything* went wrong.
321 */
322 static int init_block_mapping(Vp3DecodeContext *s)
323 {
324 int sb_x, sb_y, plane;
325 int x, y, i, j = 0;
326
327 for (plane = 0; plane < 3; plane++) {
328 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
329 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
330 int frag_width = s->fragment_width[!!plane];
331 int frag_height = s->fragment_height[!!plane];
332
333 for (sb_y = 0; sb_y < sb_height; sb_y++)
334 for (sb_x = 0; sb_x < sb_width; sb_x++)
335 for (i = 0; i < 16; i++) {
336 x = 4*sb_x + hilbert_offset[i][0];
337 y = 4*sb_y + hilbert_offset[i][1];
338
339 if (x < frag_width && y < frag_height)
340 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
341 else
342 s->superblock_fragments[j++] = -1;
343 }
344 }
345
346 return 0; /* successful path out */
347 }
348
349 /*
350 * This function sets up the dequantization tables used for a particular
351 * frame.
352 */
353 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
354 {
355 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
356 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
357 int i, plane, inter, qri, bmi, bmj, qistart;
358
359 for(inter=0; inter<2; inter++){
360 for(plane=0; plane<3; plane++){
361 int sum=0;
362 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
363 sum+= s->qr_size[inter][plane][qri];
364 if(s->qps[qpi] <= sum)
365 break;
366 }
367 qistart= sum - s->qr_size[inter][plane][qri];
368 bmi= s->qr_base[inter][plane][qri ];
369 bmj= s->qr_base[inter][plane][qri+1];
370 for(i=0; i<64; i++){
371 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
372 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
373 + s->qr_size[inter][plane][qri])
374 / (2*s->qr_size[inter][plane][qri]);
375
376 int qmin= 8<<(inter + !i);
377 int qscale= i ? ac_scale_factor : dc_scale_factor;
378
379 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
380 }
381 // all DC coefficients use the same quant so as not to interfere with DC prediction
382 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
383 }
384 }
385 }
386
387 /*
388 * This function initializes the loop filter boundary limits if the frame's
389 * quality index is different from the previous frame's.
390 *
391 * The filter_limit_values may not be larger than 127.
392 */
393 static void init_loop_filter(Vp3DecodeContext *s)
394 {
395 int *bounding_values= s->bounding_values_array+127;
396 int filter_limit;
397 int x;
398 int value;
399
400 filter_limit = s->filter_limit_values[s->qps[0]];
401 assert(filter_limit < 128);
402
403 /* set up the bounding values */
404 memset(s->bounding_values_array, 0, 256 * sizeof(int));
405 for (x = 0; x < filter_limit; x++) {
406 bounding_values[-x] = -x;
407 bounding_values[x] = x;
408 }
409 for (x = value = filter_limit; x < 128 && value; x++, value--) {
410 bounding_values[ x] = value;
411 bounding_values[-x] = -value;
412 }
413 if (value)
414 bounding_values[128] = value;
415 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
416 }
417
418 /*
419 * This function unpacks all of the superblock/macroblock/fragment coding
420 * information from the bitstream.
421 */
422 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
423 {
424 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
425 int bit = 0;
426 int current_superblock = 0;
427 int current_run = 0;
428 int num_partial_superblocks = 0;
429
430 int i, j;
431 int current_fragment;
432 int plane;
433
434 if (s->keyframe) {
435 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
436
437 } else {
438
439 /* unpack the list of partially-coded superblocks */
440 bit = get_bits1(gb) ^ 1;
441 current_run = 0;
442
443 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
444 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
445 bit = get_bits1(gb);
446 else
447 bit ^= 1;
448
449 current_run = get_vlc2(gb,
450 s->superblock_run_length_vlc.table, 6, 2) + 1;
451 if (current_run == 34)
452 current_run += get_bits(gb, 12);
453
454 if (current_superblock + current_run > s->superblock_count) {
455 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
456 return -1;
457 }
458
459 memset(s->superblock_coding + current_superblock, bit, current_run);
460
461 current_superblock += current_run;
462 if (bit)
463 num_partial_superblocks += current_run;
464 }
465
466 /* unpack the list of fully coded superblocks if any of the blocks were
467 * not marked as partially coded in the previous step */
468 if (num_partial_superblocks < s->superblock_count) {
469 int superblocks_decoded = 0;
470
471 current_superblock = 0;
472 bit = get_bits1(gb) ^ 1;
473 current_run = 0;
474
475 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
476 && get_bits_left(gb) > 0) {
477
478 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
479 bit = get_bits1(gb);
480 else
481 bit ^= 1;
482
483 current_run = get_vlc2(gb,
484 s->superblock_run_length_vlc.table, 6, 2) + 1;
485 if (current_run == 34)
486 current_run += get_bits(gb, 12);
487
488 for (j = 0; j < current_run; current_superblock++) {
489 if (current_superblock >= s->superblock_count) {
490 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
491 return -1;
492 }
493
494 /* skip any superblocks already marked as partially coded */
495 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
496 s->superblock_coding[current_superblock] = 2*bit;
497 j++;
498 }
499 }
500 superblocks_decoded += current_run;
501 }
502 }
503
504 /* if there were partial blocks, initialize bitstream for
505 * unpacking fragment codings */
506 if (num_partial_superblocks) {
507
508 current_run = 0;
509 bit = get_bits1(gb);
510 /* toggle the bit because as soon as the first run length is
511 * fetched the bit will be toggled again */
512 bit ^= 1;
513 }
514 }
515
516 /* figure out which fragments are coded; iterate through each
517 * superblock (all planes) */
518 s->total_num_coded_frags = 0;
519 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
520
521 for (plane = 0; plane < 3; plane++) {
522 int sb_start = superblock_starts[plane];
523 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
524 int num_coded_frags = 0;
525
526 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
527
528 /* iterate through all 16 fragments in a superblock */
529 for (j = 0; j < 16; j++) {
530
531 /* if the fragment is in bounds, check its coding status */
532 current_fragment = s->superblock_fragments[i * 16 + j];
533 if (current_fragment != -1) {
534 int coded = s->superblock_coding[i];
535
536 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
537
538 /* fragment may or may not be coded; this is the case
539 * that cares about the fragment coding runs */
540 if (current_run-- == 0) {
541 bit ^= 1;
542 current_run = get_vlc2(gb,
543 s->fragment_run_length_vlc.table, 5, 2);
544 }
545 coded = bit;
546 }
547
548 if (coded) {
549 /* default mode; actual mode will be decoded in
550 * the next phase */
551 s->all_fragments[current_fragment].coding_method =
552 MODE_INTER_NO_MV;
553 s->coded_fragment_list[plane][num_coded_frags++] =
554 current_fragment;
555 } else {
556 /* not coded; copy this fragment from the prior frame */
557 s->all_fragments[current_fragment].coding_method =
558 MODE_COPY;
559 }
560 }
561 }
562 }
563 s->total_num_coded_frags += num_coded_frags;
564 for (i = 0; i < 64; i++)
565 s->num_coded_frags[plane][i] = num_coded_frags;
566 if (plane < 2)
567 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
568 }
569 return 0;
570 }
571
572 /*
573 * This function unpacks all the coding mode data for individual macroblocks
574 * from the bitstream.
575 */
576 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
577 {
578 int i, j, k, sb_x, sb_y;
579 int scheme;
580 int current_macroblock;
581 int current_fragment;
582 int coding_mode;
583 int custom_mode_alphabet[CODING_MODE_COUNT];
584 const int *alphabet;
585 Vp3Fragment *frag;
586
587 if (s->keyframe) {
588 for (i = 0; i < s->fragment_count; i++)
589 s->all_fragments[i].coding_method = MODE_INTRA;
590
591 } else {
592
593 /* fetch the mode coding scheme for this frame */
594 scheme = get_bits(gb, 3);
595
596 /* is it a custom coding scheme? */
597 if (scheme == 0) {
598 for (i = 0; i < 8; i++)
599 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
600 for (i = 0; i < 8; i++)
601 custom_mode_alphabet[get_bits(gb, 3)] = i;
602 alphabet = custom_mode_alphabet;
603 } else
604 alphabet = ModeAlphabet[scheme-1];
605
606 /* iterate through all of the macroblocks that contain 1 or more
607 * coded fragments */
608 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
609 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
610 if (get_bits_left(gb) <= 0)
611 return -1;
612
613 for (j = 0; j < 4; j++) {
614 int mb_x = 2*sb_x + (j>>1);
615 int mb_y = 2*sb_y + (((j>>1)+j)&1);
616 current_macroblock = mb_y * s->macroblock_width + mb_x;
617
618 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
619 continue;
620
621 #define BLOCK_X (2*mb_x + (k&1))
622 #define BLOCK_Y (2*mb_y + (k>>1))
623 /* coding modes are only stored if the macroblock has at least one
624 * luma block coded, otherwise it must be INTER_NO_MV */
625 for (k = 0; k < 4; k++) {
626 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
627 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
628 break;
629 }
630 if (k == 4) {
631 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
632 continue;
633 }
634
635 /* mode 7 means get 3 bits for each coding mode */
636 if (scheme == 7)
637 coding_mode = get_bits(gb, 3);
638 else
639 coding_mode = alphabet
640 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
641
642 s->macroblock_coding[current_macroblock] = coding_mode;
643 for (k = 0; k < 4; k++) {
644 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
645 if (frag->coding_method != MODE_COPY)
646 frag->coding_method = coding_mode;
647 }
648
649 #define SET_CHROMA_MODES \
650 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
651 frag[s->fragment_start[1]].coding_method = coding_mode;\
652 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
653 frag[s->fragment_start[2]].coding_method = coding_mode;
654
655 if (s->chroma_y_shift) {
656 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
657 SET_CHROMA_MODES
658 } else if (s->chroma_x_shift) {
659 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
660 for (k = 0; k < 2; k++) {
661 SET_CHROMA_MODES
662 frag += s->fragment_width[1];
663 }
664 } else {
665 for (k = 0; k < 4; k++) {
666 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
667 SET_CHROMA_MODES
668 }
669 }
670 }
671 }
672 }
673 }
674
675 return 0;
676 }
677
678 /*
679 * This function unpacks all the motion vectors for the individual
680 * macroblocks from the bitstream.
681 */
682 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
683 {
684 int j, k, sb_x, sb_y;
685 int coding_mode;
686 int motion_x[4];
687 int motion_y[4];
688 int last_motion_x = 0;
689 int last_motion_y = 0;
690 int prior_last_motion_x = 0;
691 int prior_last_motion_y = 0;
692 int current_macroblock;
693 int current_fragment;
694 int frag;
695
696 if (s->keyframe)
697 return 0;
698
699 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
700 coding_mode = get_bits1(gb);
701
702 /* iterate through all of the macroblocks that contain 1 or more
703 * coded fragments */
704 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
705 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
706 if (get_bits_left(gb) <= 0)
707 return -1;
708
709 for (j = 0; j < 4; j++) {
710 int mb_x = 2*sb_x + (j>>1);
711 int mb_y = 2*sb_y + (((j>>1)+j)&1);
712 current_macroblock = mb_y * s->macroblock_width + mb_x;
713
714 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
715 (s->macroblock_coding[current_macroblock] == MODE_COPY))
716 continue;
717
718 switch (s->macroblock_coding[current_macroblock]) {
719
720 case MODE_INTER_PLUS_MV:
721 case MODE_GOLDEN_MV:
722 /* all 6 fragments use the same motion vector */
723 if (coding_mode == 0) {
724 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
725 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
726 } else {
727 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
728 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
729 }
730
731 /* vector maintenance, only on MODE_INTER_PLUS_MV */
732 if (s->macroblock_coding[current_macroblock] ==
733 MODE_INTER_PLUS_MV) {
734 prior_last_motion_x = last_motion_x;
735 prior_last_motion_y = last_motion_y;
736 last_motion_x = motion_x[0];
737 last_motion_y = motion_y[0];
738 }
739 break;
740
741 case MODE_INTER_FOURMV:
742 /* vector maintenance */
743 prior_last_motion_x = last_motion_x;
744 prior_last_motion_y = last_motion_y;
745
746 /* fetch 4 vectors from the bitstream, one for each
747 * Y fragment, then average for the C fragment vectors */
748 for (k = 0; k < 4; k++) {
749 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
750 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
751 if (coding_mode == 0) {
752 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
753 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
754 } else {
755 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
756 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
757 }
758 last_motion_x = motion_x[k];
759 last_motion_y = motion_y[k];
760 } else {
761 motion_x[k] = 0;
762 motion_y[k] = 0;
763 }
764 }
765 break;
766
767 case MODE_INTER_LAST_MV:
768 /* all 6 fragments use the last motion vector */
769 motion_x[0] = last_motion_x;
770 motion_y[0] = last_motion_y;
771
772 /* no vector maintenance (last vector remains the
773 * last vector) */
774 break;
775
776 case MODE_INTER_PRIOR_LAST:
777 /* all 6 fragments use the motion vector prior to the
778 * last motion vector */
779 motion_x[0] = prior_last_motion_x;
780 motion_y[0] = prior_last_motion_y;
781
782 /* vector maintenance */
783 prior_last_motion_x = last_motion_x;
784 prior_last_motion_y = last_motion_y;
785 last_motion_x = motion_x[0];
786 last_motion_y = motion_y[0];
787 break;
788
789 default:
790 /* covers intra, inter without MV, golden without MV */
791 motion_x[0] = 0;
792 motion_y[0] = 0;
793
794 /* no vector maintenance */
795 break;
796 }
797
798 /* assign the motion vectors to the correct fragments */
799 for (k = 0; k < 4; k++) {
800 current_fragment =
801 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
802 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
803 s->motion_val[0][current_fragment][0] = motion_x[k];
804 s->motion_val[0][current_fragment][1] = motion_y[k];
805 } else {
806 s->motion_val[0][current_fragment][0] = motion_x[0];
807 s->motion_val[0][current_fragment][1] = motion_y[0];
808 }
809 }
810
811 if (s->chroma_y_shift) {
812 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
813 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
814 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
815 }
816 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
817 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
818 frag = mb_y*s->fragment_width[1] + mb_x;
819 s->motion_val[1][frag][0] = motion_x[0];
820 s->motion_val[1][frag][1] = motion_y[0];
821 } else if (s->chroma_x_shift) {
822 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
823 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
824 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
825 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
826 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
827 } else {
828 motion_x[1] = motion_x[0];
829 motion_y[1] = motion_y[0];
830 }
831 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
832 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
833
834 frag = 2*mb_y*s->fragment_width[1] + mb_x;
835 for (k = 0; k < 2; k++) {
836 s->motion_val[1][frag][0] = motion_x[k];
837 s->motion_val[1][frag][1] = motion_y[k];
838 frag += s->fragment_width[1];
839 }
840 } else {
841 for (k = 0; k < 4; k++) {
842 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
843 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
844 s->motion_val[1][frag][0] = motion_x[k];
845 s->motion_val[1][frag][1] = motion_y[k];
846 } else {
847 s->motion_val[1][frag][0] = motion_x[0];
848 s->motion_val[1][frag][1] = motion_y[0];
849 }
850 }
851 }
852 }
853 }
854 }
855
856 return 0;
857 }
858
859 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
860 {
861 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
862 int num_blocks = s->total_num_coded_frags;
863
864 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
865 i = blocks_decoded = num_blocks_at_qpi = 0;
866
867 bit = get_bits1(gb) ^ 1;
868 run_length = 0;
869
870 do {
871 if (run_length == MAXIMUM_LONG_BIT_RUN)
872 bit = get_bits1(gb);
873 else
874 bit ^= 1;
875
876 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
877 if (run_length == 34)
878 run_length += get_bits(gb, 12);
879 blocks_decoded += run_length;
880
881 if (!bit)
882 num_blocks_at_qpi += run_length;
883
884 for (j = 0; j < run_length; i++) {
885 if (i >= s->total_num_coded_frags)
886 return -1;
887
888 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
889 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
890 j++;
891 }
892 }
893 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
894
895 num_blocks -= num_blocks_at_qpi;
896 }
897
898 return 0;
899 }
900
901 /*
902 * This function is called by unpack_dct_coeffs() to extract the VLCs from
903 * the bitstream. The VLCs encode tokens which are used to unpack DCT
904 * data. This function unpacks all the VLCs for either the Y plane or both
905 * C planes, and is called for DC coefficients or different AC coefficient
906 * levels (since different coefficient types require different VLC tables.
907 *
908 * This function returns a residual eob run. E.g, if a particular token gave
909 * instructions to EOB the next 5 fragments and there were only 2 fragments
910 * left in the current fragment range, 3 would be returned so that it could
911 * be passed into the next call to this same function.
912 */
913 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
914 VLC *table, int coeff_index,
915 int plane,
916 int eob_run)
917 {
918 int i, j = 0;
919 int token;
920 int zero_run = 0;
921 DCTELEM coeff = 0;
922 int bits_to_get;
923 int blocks_ended;
924 int coeff_i = 0;
925 int num_coeffs = s->num_coded_frags[plane][coeff_index];
926 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
927
928 /* local references to structure members to avoid repeated deferences */
929 int *coded_fragment_list = s->coded_fragment_list[plane];
930 Vp3Fragment *all_fragments = s->all_fragments;
931 VLC_TYPE (*vlc_table)[2] = table->table;
932
933 if (num_coeffs < 0)
934 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
935
936 if (eob_run > num_coeffs) {
937 coeff_i = blocks_ended = num_coeffs;
938 eob_run -= num_coeffs;
939 } else {
940 coeff_i = blocks_ended = eob_run;
941 eob_run = 0;
942 }
943
944 // insert fake EOB token to cover the split between planes or zzi
945 if (blocks_ended)
946 dct_tokens[j++] = blocks_ended << 2;
947
948 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
949 /* decode a VLC into a token */
950 token = get_vlc2(gb, vlc_table, 11, 3);
951 /* use the token to get a zero run, a coefficient, and an eob run */
952 if ((unsigned) token <= 6U) {
953 eob_run = eob_run_base[token];
954 if (eob_run_get_bits[token])
955 eob_run += get_bits(gb, eob_run_get_bits[token]);
956
957 // record only the number of blocks ended in this plane,
958 // any spill will be recorded in the next plane.
959 if (eob_run > num_coeffs - coeff_i) {
960 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
961 blocks_ended += num_coeffs - coeff_i;
962 eob_run -= num_coeffs - coeff_i;
963 coeff_i = num_coeffs;
964 } else {
965 dct_tokens[j++] = TOKEN_EOB(eob_run);
966 blocks_ended += eob_run;
967 coeff_i += eob_run;
968 eob_run = 0;
969 }
970 } else if (token >= 0) {
971 bits_to_get = coeff_get_bits[token];
972 if (bits_to_get)
973 bits_to_get = get_bits(gb, bits_to_get);
974 coeff = coeff_tables[token][bits_to_get];
975
976 zero_run = zero_run_base[token];
977 if (zero_run_get_bits[token])
978 zero_run += get_bits(gb, zero_run_get_bits[token]);
979
980 if (zero_run) {
981 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
982 } else {
983 // Save DC into the fragment structure. DC prediction is
984 // done in raster order, so the actual DC can't be in with
985 // other tokens. We still need the token in dct_tokens[]
986 // however, or else the structure collapses on itself.
987 if (!coeff_index)
988 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
989
990 dct_tokens[j++] = TOKEN_COEFF(coeff);
991 }
992
993 if (coeff_index + zero_run > 64) {
994 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
995 " %d coeffs left\n", zero_run, 64-coeff_index);
996 zero_run = 64 - coeff_index;
997 }
998
999 // zero runs code multiple coefficients,
1000 // so don't try to decode coeffs for those higher levels
1001 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
1002 s->num_coded_frags[plane][i]--;
1003 coeff_i++;
1004 } else {
1005 av_log(s->avctx, AV_LOG_ERROR,
1006 "Invalid token %d\n", token);
1007 return -1;
1008 }
1009 }
1010
1011 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1012 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1013
1014 // decrement the number of blocks that have higher coeffecients for each
1015 // EOB run at this level
1016 if (blocks_ended)
1017 for (i = coeff_index+1; i < 64; i++)
1018 s->num_coded_frags[plane][i] -= blocks_ended;
1019
1020 // setup the next buffer
1021 if (plane < 2)
1022 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
1023 else if (coeff_index < 63)
1024 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
1025
1026 return eob_run;
1027 }
1028
1029 static void reverse_dc_prediction(Vp3DecodeContext *s,
1030 int first_fragment,
1031 int fragment_width,
1032 int fragment_height);
1033 /*
1034 * This function unpacks all of the DCT coefficient data from the
1035 * bitstream.
1036 */
1037 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1038 {
1039 int i;
1040 int dc_y_table;
1041 int dc_c_table;
1042 int ac_y_table;
1043 int ac_c_table;
1044 int residual_eob_run = 0;
1045 VLC *y_tables[64];
1046 VLC *c_tables[64];
1047
1048 s->dct_tokens[0][0] = s->dct_tokens_base;
1049
1050 /* fetch the DC table indexes */
1051 dc_y_table = get_bits(gb, 4);
1052 dc_c_table = get_bits(gb, 4);
1053
1054 /* unpack the Y plane DC coefficients */
1055 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1056 0, residual_eob_run);
1057 if (residual_eob_run < 0)
1058 return residual_eob_run;
1059
1060 /* reverse prediction of the Y-plane DC coefficients */
1061 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1062
1063 /* unpack the C plane DC coefficients */
1064 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1065 1, residual_eob_run);
1066 if (residual_eob_run < 0)
1067 return residual_eob_run;
1068 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1069 2, residual_eob_run);
1070 if (residual_eob_run < 0)
1071 return residual_eob_run;
1072
1073 /* reverse prediction of the C-plane DC coefficients */
1074 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1075 {
1076 reverse_dc_prediction(s, s->fragment_start[1],
1077 s->fragment_width[1], s->fragment_height[1]);
1078 reverse_dc_prediction(s, s->fragment_start[2],
1079 s->fragment_width[1], s->fragment_height[1]);
1080 }
1081
1082 /* fetch the AC table indexes */
1083 ac_y_table = get_bits(gb, 4);
1084 ac_c_table = get_bits(gb, 4);
1085
1086 /* build tables of AC VLC tables */
1087 for (i = 1; i <= 5; i++) {
1088 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1089 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1090 }
1091 for (i = 6; i <= 14; i++) {
1092 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1093 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1094 }
1095 for (i = 15; i <= 27; i++) {
1096 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1097 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1098 }
1099 for (i = 28; i <= 63; i++) {
1100 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1101 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1102 }
1103
1104 /* decode all AC coefficents */
1105 for (i = 1; i <= 63; i++) {
1106 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1107 0, residual_eob_run);
1108 if (residual_eob_run < 0)
1109 return residual_eob_run;
1110
1111 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1112 1, residual_eob_run);
1113 if (residual_eob_run < 0)
1114 return residual_eob_run;
1115 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1116 2, residual_eob_run);
1117 if (residual_eob_run < 0)
1118 return residual_eob_run;
1119 }
1120
1121 return 0;
1122 }
1123
1124 /*
1125 * This function reverses the DC prediction for each coded fragment in
1126 * the frame. Much of this function is adapted directly from the original
1127 * VP3 source code.
1128 */
1129 #define COMPATIBLE_FRAME(x) \
1130 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1131 #define DC_COEFF(u) s->all_fragments[u].dc
1132
1133 static void reverse_dc_prediction(Vp3DecodeContext *s,
1134 int first_fragment,
1135 int fragment_width,
1136 int fragment_height)
1137 {
1138
1139 #define PUL 8
1140 #define PU 4
1141 #define PUR 2
1142 #define PL 1
1143
1144 int x, y;
1145 int i = first_fragment;
1146
1147 int predicted_dc;
1148
1149 /* DC values for the left, up-left, up, and up-right fragments */
1150 int vl, vul, vu, vur;
1151
1152 /* indexes for the left, up-left, up, and up-right fragments */
1153 int l, ul, u, ur;
1154
1155 /*
1156 * The 6 fields mean:
1157 * 0: up-left multiplier
1158 * 1: up multiplier
1159 * 2: up-right multiplier
1160 * 3: left multiplier
1161 */
1162 static const int predictor_transform[16][4] = {
1163 { 0, 0, 0, 0},
1164 { 0, 0, 0,128}, // PL
1165 { 0, 0,128, 0}, // PUR
1166 { 0, 0, 53, 75}, // PUR|PL
1167 { 0,128, 0, 0}, // PU
1168 { 0, 64, 0, 64}, // PU|PL
1169 { 0,128, 0, 0}, // PU|PUR
1170 { 0, 0, 53, 75}, // PU|PUR|PL
1171 {128, 0, 0, 0}, // PUL
1172 { 0, 0, 0,128}, // PUL|PL
1173 { 64, 0, 64, 0}, // PUL|PUR
1174 { 0, 0, 53, 75}, // PUL|PUR|PL
1175 { 0,128, 0, 0}, // PUL|PU
1176 {-104,116, 0,116}, // PUL|PU|PL
1177 { 24, 80, 24, 0}, // PUL|PU|PUR
1178 {-104,116, 0,116} // PUL|PU|PUR|PL
1179 };
1180
1181 /* This table shows which types of blocks can use other blocks for
1182 * prediction. For example, INTRA is the only mode in this table to
1183 * have a frame number of 0. That means INTRA blocks can only predict
1184 * from other INTRA blocks. There are 2 golden frame coding types;
1185 * blocks encoding in these modes can only predict from other blocks
1186 * that were encoded with these 1 of these 2 modes. */
1187 static const unsigned char compatible_frame[9] = {
1188 1, /* MODE_INTER_NO_MV */
1189 0, /* MODE_INTRA */
1190 1, /* MODE_INTER_PLUS_MV */
1191 1, /* MODE_INTER_LAST_MV */
1192 1, /* MODE_INTER_PRIOR_MV */
1193 2, /* MODE_USING_GOLDEN */
1194 2, /* MODE_GOLDEN_MV */
1195 1, /* MODE_INTER_FOUR_MV */
1196 3 /* MODE_COPY */
1197 };
1198 int current_frame_type;
1199
1200 /* there is a last DC predictor for each of the 3 frame types */
1201 short last_dc[3];
1202
1203 int transform = 0;
1204
1205 vul = vu = vur = vl = 0;
1206 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1207
1208 /* for each fragment row... */
1209 for (y = 0; y < fragment_height; y++) {
1210
1211 /* for each fragment in a row... */
1212 for (x = 0; x < fragment_width; x++, i++) {
1213
1214 /* reverse prediction if this block was coded */
1215 if (s->all_fragments[i].coding_method != MODE_COPY) {
1216
1217 current_frame_type =
1218 compatible_frame[s->all_fragments[i].coding_method];
1219
1220 transform= 0;
1221 if(x){
1222 l= i-1;
1223 vl = DC_COEFF(l);
1224 if(COMPATIBLE_FRAME(l))
1225 transform |= PL;
1226 }
1227 if(y){
1228 u= i-fragment_width;
1229 vu = DC_COEFF(u);
1230 if(COMPATIBLE_FRAME(u))
1231 transform |= PU;
1232 if(x){
1233 ul= i-fragment_width-1;
1234 vul = DC_COEFF(ul);
1235 if(COMPATIBLE_FRAME(ul))
1236 transform |= PUL;
1237 }
1238 if(x + 1 < fragment_width){
1239 ur= i-fragment_width+1;
1240 vur = DC_COEFF(ur);
1241 if(COMPATIBLE_FRAME(ur))
1242 transform |= PUR;
1243 }
1244 }
1245
1246 if (transform == 0) {
1247
1248 /* if there were no fragments to predict from, use last
1249 * DC saved */
1250 predicted_dc = last_dc[current_frame_type];
1251 } else {
1252
1253 /* apply the appropriate predictor transform */
1254 predicted_dc =
1255 (predictor_transform[transform][0] * vul) +
1256 (predictor_transform[transform][1] * vu) +
1257 (predictor_transform[transform][2] * vur) +
1258 (predictor_transform[transform][3] * vl);
1259
1260 predicted_dc /= 128;
1261
1262 /* check for outranging on the [ul u l] and
1263 * [ul u ur l] predictors */
1264 if ((transform == 15) || (transform == 13)) {
1265 if (FFABS(predicted_dc - vu) > 128)
1266 predicted_dc = vu;
1267 else if (FFABS(predicted_dc - vl) > 128)
1268 predicted_dc = vl;
1269 else if (FFABS(predicted_dc - vul) > 128)
1270 predicted_dc = vul;
1271 }
1272 }
1273
1274 /* at long last, apply the predictor */
1275 DC_COEFF(i) += predicted_dc;
1276 /* save the DC */
1277 last_dc[current_frame_type] = DC_COEFF(i);
1278 }
1279 }
1280 }
1281 }
1282
1283 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1284 {
1285 int x, y;
1286 int *bounding_values= s->bounding_values_array+127;
1287
1288 int width = s->fragment_width[!!plane];
1289 int height = s->fragment_height[!!plane];
1290 int fragment = s->fragment_start [plane] + ystart * width;
1291 int stride = s->current_frame.linesize[plane];
1292 uint8_t *plane_data = s->current_frame.data [plane];
1293 if (!s->flipped_image) stride = -stride;
1294 plane_data += s->data_offset[plane] + 8*ystart*stride;
1295
1296 for (y = ystart; y < yend; y++) {
1297
1298 for (x = 0; x < width; x++) {
1299 /* This code basically just deblocks on the edges of coded blocks.
1300 * However, it has to be much more complicated because of the
1301 * braindamaged deblock ordering used in VP3/Theora. Order matters
1302 * because some pixels get filtered twice. */
1303 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1304 {
1305 /* do not perform left edge filter for left columns frags */
1306 if (x > 0) {
1307 s->vp3dsp.h_loop_filter(
1308 plane_data + 8*x,
1309 stride, bounding_values);
1310 }
1311
1312 /* do not perform top edge filter for top row fragments */
1313 if (y > 0) {
1314 s->vp3dsp.v_loop_filter(
1315 plane_data + 8*x,
1316 stride, bounding_values);
1317 }
1318
1319 /* do not perform right edge filter for right column
1320 * fragments or if right fragment neighbor is also coded
1321 * in this frame (it will be filtered in next iteration) */
1322 if ((x < width - 1) &&
1323 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1324 s->vp3dsp.h_loop_filter(
1325 plane_data + 8*x + 8,
1326 stride, bounding_values);
1327 }
1328
1329 /* do not perform bottom edge filter for bottom row
1330 * fragments or if bottom fragment neighbor is also coded
1331 * in this frame (it will be filtered in the next row) */
1332 if ((y < height - 1) &&
1333 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1334 s->vp3dsp.v_loop_filter(
1335 plane_data + 8*x + 8*stride,
1336 stride, bounding_values);
1337 }
1338 }
1339
1340 fragment++;
1341 }
1342 plane_data += 8*stride;
1343 }
1344 }
1345
1346 /**
1347 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1348 * for the next block in coding order
1349 */
1350 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1351 int plane, int inter, DCTELEM block[64])
1352 {
1353 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1354 uint8_t *perm = s->scantable.permutated;
1355 int i = 0;
1356
1357 do {
1358 int token = *s->dct_tokens[plane][i];
1359 switch (token & 3) {
1360 case 0: // EOB
1361 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1362 s->dct_tokens[plane][i]++;
1363 else
1364 *s->dct_tokens[plane][i] = token & ~3;
1365 goto end;
1366 case 1: // zero run
1367 s->dct_tokens[plane][i]++;
1368 i += (token >> 2) & 0x7f;
1369 if (i > 63) {
1370 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1371 return i;
1372 }
1373 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1374 i++;
1375 break;
1376 case 2: // coeff
1377 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1378 s->dct_tokens[plane][i++]++;
1379 break;
1380 default: // shouldn't happen
1381 return i;
1382 }
1383 } while (i < 64);
1384 // return value is expected to be a valid level
1385 i--;
1386 end:
1387 // the actual DC+prediction is in the fragment structure
1388 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1389 return i;
1390 }
1391
1392 /**
1393 * called when all pixels up to row y are complete
1394 */
1395 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1396 {
1397 int h, cy, i;
1398 int offset[AV_NUM_DATA_POINTERS];
1399
1400 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1401 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1402
1403 // At the end of the frame, report INT_MAX instead of the height of the frame.
1404 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1405 // they don't have to clip their values.
1406 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1407 }
1408
1409 if(s->avctx->draw_horiz_band==NULL)
1410 return;
1411
1412 h= y - s->last_slice_end;
1413 s->last_slice_end= y;
1414 y -= h;
1415
1416 if (!s->flipped_image) {
1417 y = s->avctx->height - y - h;
1418 }
1419
1420 cy = y >> s->chroma_y_shift;
1421 offset[0] = s->current_frame.linesize[0]*y;
1422 offset[1] = s->current_frame.linesize[1]*cy;
1423 offset[2] = s->current_frame.linesize[2]*cy;
1424 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1425 offset[i] = 0;
1426
1427 emms_c();
1428 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1429 }
1430
1431 /**
1432 * Wait for the reference frame of the current fragment.
1433 * The progress value is in luma pixel rows.
1434 */
1435 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1436 {
1437 AVFrame *ref_frame;
1438 int ref_row;
1439 int border = motion_y&1;
1440
1441 if (fragment->coding_method == MODE_USING_GOLDEN ||
1442 fragment->coding_method == MODE_GOLDEN_MV)
1443 ref_frame = &s->golden_frame;
1444 else
1445 ref_frame = &s->last_frame;
1446
1447 ref_row = y + (motion_y>>1);
1448 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1449
1450 ff_thread_await_progress(ref_frame, ref_row, 0);
1451 }
1452
1453 /*
1454 * Perform the final rendering for a particular slice of data.
1455 * The slice number ranges from 0..(c_superblock_height - 1).
1456 */
1457 static void render_slice(Vp3DecodeContext *s, int slice)
1458 {
1459 int x, y, i, j, fragment;
1460 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1461 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1462 int motion_halfpel_index;
1463 uint8_t *motion_source;
1464 int plane, first_pixel;
1465
1466 if (slice >= s->c_superblock_height)
1467 return;
1468
1469 for (plane = 0; plane < 3; plane++) {
1470 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1471 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1472 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1473 int stride = s->current_frame.linesize[plane];
1474 int plane_width = s->width >> (plane && s->chroma_x_shift);
1475 int plane_height = s->height >> (plane && s->chroma_y_shift);
1476 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1477
1478 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1479 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1480 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1481
1482 int fragment_width = s->fragment_width[!!plane];
1483 int fragment_height = s->fragment_height[!!plane];
1484 int fragment_start = s->fragment_start[plane];
1485 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1486
1487 if (!s->flipped_image) stride = -stride;
1488 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1489 continue;
1490
1491 /* for each superblock row in the slice (both of them)... */
1492 for (; sb_y < slice_height; sb_y++) {
1493
1494 /* for each superblock in a row... */
1495 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1496
1497 /* for each block in a superblock... */
1498 for (j = 0; j < 16; j++) {
1499 x = 4*sb_x + hilbert_offset[j][0];
1500 y = 4*sb_y + hilbert_offset[j][1];
1501 fragment = y*fragment_width + x;
1502
1503 i = fragment_start + fragment;
1504
1505 // bounds check
1506 if (x >= fragment_width || y >= fragment_height)
1507 continue;
1508
1509 first_pixel = 8*y*stride + 8*x;
1510
1511 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1512 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1513
1514 /* transform if this block was coded */
1515 if (s->all_fragments[i].coding_method != MODE_COPY) {
1516 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1517 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1518 motion_source= golden_plane;
1519 else
1520 motion_source= last_plane;
1521
1522 motion_source += first_pixel;
1523 motion_halfpel_index = 0;
1524
1525 /* sort out the motion vector if this fragment is coded
1526 * using a motion vector method */
1527 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1528 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1529 int src_x, src_y;
1530 motion_x = motion_val[fragment][0];
1531 motion_y = motion_val[fragment][1];
1532
1533 src_x= (motion_x>>1) + 8*x;
1534 src_y= (motion_y>>1) + 8*y;
1535
1536 motion_halfpel_index = motion_x & 0x01;
1537 motion_source += (motion_x >> 1);
1538
1539 motion_halfpel_index |= (motion_y & 0x01) << 1;
1540 motion_source += ((motion_y >> 1) * stride);
1541
1542 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1543 uint8_t *temp= s->edge_emu_buffer;
1544 if(stride<0) temp -= 8*stride;
1545
1546 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1547 motion_source= temp;
1548 }
1549 }
1550
1551
1552 /* first, take care of copying a block from either the
1553 * previous or the golden frame */
1554 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1555 /* Note, it is possible to implement all MC cases with
1556 put_no_rnd_pixels_l2 which would look more like the
1557 VP3 source but this would be slower as
1558 put_no_rnd_pixels_tab is better optimzed */
1559 if(motion_halfpel_index != 3){
1560 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1561 output_plane + first_pixel,
1562 motion_source, stride, 8);
1563 }else{
1564 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1565 s->dsp.put_no_rnd_pixels_l2[1](
1566 output_plane + first_pixel,
1567 motion_source - d,
1568 motion_source + stride + 1 + d,
1569 stride, 8);
1570 }
1571 }
1572
1573 s->dsp.clear_block(block);
1574
1575 /* invert DCT and place (or add) in final output */
1576
1577 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1578 int index;
1579 index = vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1580 if (index > 63)
1581 continue;
1582 s->vp3dsp.idct_put(
1583 output_plane + first_pixel,
1584 stride,
1585 block);
1586 } else {
1587 int index = vp3_dequant(s, s->all_fragments + i, plane, 1, block);
1588 if (index > 63)
1589 continue;
1590 if (index > 0) {
1591 s->vp3dsp.idct_add(
1592 output_plane + first_pixel,
1593 stride,
1594 block);
1595 } else {
1596 s->vp3dsp.idct_dc_add(output_plane + first_pixel, stride, block);
1597 }
1598 }
1599 } else {
1600
1601 /* copy directly from the previous frame */
1602 s->dsp.put_pixels_tab[1][0](
1603 output_plane + first_pixel,
1604 last_plane + first_pixel,
1605 stride, 8);
1606
1607 }
1608 }
1609 }
1610
1611 // Filter up to the last row in the superblock row
1612 if (!s->skip_loop_filter)
1613 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1614 }
1615 }
1616
1617 /* this looks like a good place for slice dispatch... */
1618 /* algorithm:
1619 * if (slice == s->macroblock_height - 1)
1620 * dispatch (both last slice & 2nd-to-last slice);
1621 * else if (slice > 0)
1622 * dispatch (slice - 1);
1623 */
1624
1625 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1626 }
1627
1628 /// Allocate tables for per-frame data in Vp3DecodeContext
1629 static av_cold int allocate_tables(AVCodecContext *avctx)
1630 {
1631 Vp3DecodeContext *s = avctx->priv_data;
1632 int y_fragment_count, c_fragment_count;
1633
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
1637 s->superblock_coding = av_malloc(s->superblock_count);
1638 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1639 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1640 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1641 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1642 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1643
1644 /* work out the block mapping tables */
1645 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1646 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1647
1648 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1649 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1650 !s->motion_val[0] || !s->motion_val[1]) {
1651 vp3_decode_end(avctx);
1652 return -1;
1653 }
1654
1655 init_block_mapping(s);
1656
1657 return 0;
1658 }
1659
1660 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1661 {
1662 Vp3DecodeContext *s = avctx->priv_data;
1663 int i, inter, plane;
1664 int c_width;
1665 int c_height;
1666 int y_fragment_count, c_fragment_count;
1667
1668 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1669 s->version = 0;
1670 else
1671 s->version = 1;
1672
1673 s->avctx = avctx;
1674 s->width = FFALIGN(avctx->width, 16);
1675 s->height = FFALIGN(avctx->height, 16);
1676 if (avctx->pix_fmt == AV_PIX_FMT_NONE)
1677 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1678 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1679 ff_dsputil_init(&s->dsp, avctx);
1680 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
1681
1682 ff_init_scantable_permutation(s->dsp.idct_permutation, s->vp3dsp.idct_perm);
1683 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1684
1685 /* initialize to an impossible value which will force a recalculation
1686 * in the first frame decode */
1687 for (i = 0; i < 3; i++)
1688 s->qps[i] = -1;
1689
1690 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift,
1691 &s->chroma_y_shift);
1692
1693 s->y_superblock_width = (s->width + 31) / 32;
1694 s->y_superblock_height = (s->height + 31) / 32;
1695 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1696
1697 /* work out the dimensions for the C planes */
1698 c_width = s->width >> s->chroma_x_shift;
1699 c_height = s->height >> s->chroma_y_shift;
1700 s->c_superblock_width = (c_width + 31) / 32;
1701 s->c_superblock_height = (c_height + 31) / 32;
1702 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1703
1704 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1705 s->u_superblock_start = s->y_superblock_count;
1706 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1707
1708 s->macroblock_width = (s->width + 15) / 16;
1709 s->macroblock_height = (s->height + 15) / 16;
1710 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1711
1712 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1713 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1714 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1715 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1716
1717 /* fragment count covers all 8x8 blocks for all 3 planes */
1718 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1719 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1720 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1721 s->fragment_start[1] = y_fragment_count;
1722 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1723
1724 if (!s->theora_tables)
1725 {
1726 for (i = 0; i < 64; i++) {
1727 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1728 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1729 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1730 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1731 s->base_matrix[2][i] = vp31_inter_dequant[i];
1732 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1733 }
1734
1735 for(inter=0; inter<2; inter++){
1736 for(plane=0; plane<3; plane++){
1737 s->qr_count[inter][plane]= 1;
1738 s->qr_size [inter][plane][0]= 63;
1739 s->qr_base [inter][plane][0]=
1740 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1741 }
1742 }
1743
1744 /* init VLC tables */
1745 for (i = 0; i < 16; i++) {
1746
1747 /* DC histograms */
1748 init_vlc(&s->dc_vlc[i], 11, 32,
1749 &dc_bias[i][0][1], 4, 2,
1750 &dc_bias[i][0][0], 4, 2, 0);
1751
1752 /* group 1 AC histograms */
1753 init_vlc(&s->ac_vlc_1[i], 11, 32,
1754 &ac_bias_0[i][0][1], 4, 2,
1755 &ac_bias_0[i][0][0], 4, 2, 0);
1756
1757 /* group 2 AC histograms */
1758 init_vlc(&s->ac_vlc_2[i], 11, 32,
1759 &ac_bias_1[i][0][1], 4, 2,
1760 &ac_bias_1[i][0][0], 4, 2, 0);
1761
1762 /* group 3 AC histograms */
1763 init_vlc(&s->ac_vlc_3[i], 11, 32,
1764 &ac_bias_2[i][0][1], 4, 2,
1765 &ac_bias_2[i][0][0], 4, 2, 0);
1766
1767 /* group 4 AC histograms */
1768 init_vlc(&s->ac_vlc_4[i], 11, 32,
1769 &ac_bias_3[i][0][1], 4, 2,
1770 &ac_bias_3[i][0][0], 4, 2, 0);
1771 }
1772 } else {
1773
1774 for (i = 0; i < 16; i++) {
1775 /* DC histograms */
1776 if (init_vlc(&s->dc_vlc[i], 11, 32,
1777 &s->huffman_table[i][0][1], 8, 4,
1778 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1779 goto vlc_fail;
1780
1781 /* group 1 AC histograms */
1782 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1783 &s->huffman_table[i+16][0][1], 8, 4,
1784 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1785 goto vlc_fail;
1786
1787 /* group 2 AC histograms */
1788 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1789 &s->huffman_table[i+16*2][0][1], 8, 4,
1790 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1791 goto vlc_fail;
1792
1793 /* group 3 AC histograms */
1794 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1795 &s->huffman_table[i+16*3][0][1], 8, 4,
1796 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1797 goto vlc_fail;
1798
1799 /* group 4 AC histograms */
1800 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1801 &s->huffman_table[i+16*4][0][1], 8, 4,
1802 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1803 goto vlc_fail;
1804 }
1805 }
1806
1807 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1808 &superblock_run_length_vlc_table[0][1], 4, 2,
1809 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1810
1811 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1812 &fragment_run_length_vlc_table[0][1], 4, 2,
1813 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1814
1815 init_vlc(&s->mode_code_vlc, 3, 8,
1816 &mode_code_vlc_table[0][1], 2, 1,
1817 &mode_code_vlc_table[0][0], 2, 1, 0);
1818
1819 init_vlc(&s->motion_vector_vlc, 6, 63,
1820 &motion_vector_vlc_table[0][1], 2, 1,
1821 &motion_vector_vlc_table[0][0], 2, 1, 0);
1822
1823 for (i = 0; i < 3; i++) {
1824 s->current_frame.data[i] = NULL;
1825 s->last_frame.data[i] = NULL;
1826 s->golden_frame.data[i] = NULL;
1827 }
1828
1829 return allocate_tables(avctx);
1830
1831 vlc_fail:
1832 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1833 return -1;
1834 }
1835
1836 /// Release and shuffle frames after decode finishes
1837 static void update_frames(AVCodecContext *avctx)
1838 {
1839 Vp3DecodeContext *s = avctx->priv_data;
1840
1841 /* release the last frame, if it is allocated and if it is not the
1842 * golden frame */
1843 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1844 ff_thread_release_buffer(avctx, &s->last_frame);
1845
1846 /* shuffle frames (last = current) */
1847 s->last_frame= s->current_frame;
1848
1849 if (s->keyframe) {
1850 if (s->golden_frame.data[0])
1851 ff_thread_release_buffer(avctx, &s->golden_frame);
1852 s->golden_frame = s->current_frame;
1853 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1854 }
1855
1856 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1857 }
1858
1859 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1860 {
1861 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1862 int qps_changed = 0, i, err;
1863
1864 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1865
1866 if (!s1->current_frame.data[0]
1867 ||s->width != s1->width
1868 ||s->height!= s1->height) {
1869 if (s != s1)
1870 copy_fields(s, s1, golden_frame, current_frame);
1871 return -1;
1872 }
1873
1874 if (s != s1) {
1875 // init tables if the first frame hasn't been decoded
1876 if (!s->current_frame.data[0]) {
1877 int y_fragment_count, c_fragment_count;
1878 s->avctx = dst;
1879 err = allocate_tables(dst);
1880 if (err)
1881 return err;
1882 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1883 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1884 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1885 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1886 }
1887
1888 // copy previous frame data
1889 copy_fields(s, s1, golden_frame, dsp);
1890
1891 // copy qscale data if necessary
1892 for (i = 0; i < 3; i++) {
1893 if (s->qps[i] != s1->qps[1]) {
1894 qps_changed = 1;
1895 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1896 }
1897 }
1898
1899 if (s->qps[0] != s1->qps[0])
1900 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1901
1902 if (qps_changed)
1903 copy_fields(s, s1, qps, superblock_count);
1904 #undef copy_fields
1905 }
1906
1907 update_frames(dst);
1908
1909 return 0;
1910 }
1911
1912 static int vp3_decode_frame(AVCodecContext *avctx,
1913 void *data, int *got_frame,
1914 AVPacket *avpkt)
1915 {
1916 const uint8_t *buf = avpkt->data;
1917 int buf_size = avpkt->size;
1918 Vp3DecodeContext *s = avctx->priv_data;
1919 GetBitContext gb;
1920 int i;
1921
1922 init_get_bits(&gb, buf, buf_size * 8);
1923
1924 if (s->theora && get_bits1(&gb))
1925 {
1926 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1927 return -1;
1928 }
1929
1930 s->keyframe = !get_bits1(&gb);
1931 if (!s->theora)
1932 skip_bits(&gb, 1);
1933 for (i = 0; i < 3; i++)
1934 s->last_qps[i] = s->qps[i];
1935
1936 s->nqps=0;
1937 do{
1938 s->qps[s->nqps++]= get_bits(&gb, 6);
1939 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1940 for (i = s->nqps; i < 3; i++)
1941 s->qps[i] = -1;
1942
1943 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1944 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1945 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1946
1947 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1948 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1949
1950 if (s->qps[0] != s->last_qps[0])
1951 init_loop_filter(s);
1952
1953 for (i = 0; i < s->nqps; i++)
1954 // reinit all dequantizers if the first one changed, because
1955 // the DC of the first quantizer must be used for all matrices
1956 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1957 init_dequantizer(s, i);
1958
1959 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1960 return buf_size;
1961
1962 s->current_frame.reference = 3;
1963 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1964 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1965 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1966 goto error;
1967 }
1968
1969 if (!s->edge_emu_buffer)
1970 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1971
1972 if (s->keyframe) {
1973 if (!s->theora)
1974 {
1975 skip_bits(&gb, 4); /* width code */
1976 skip_bits(&gb, 4); /* height code */
1977 if (s->version)
1978 {
1979 s->version = get_bits(&gb, 5);
1980 if (avctx->frame_number == 0)
1981 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1982 }
1983 }
1984 if (s->version || s->theora)
1985 {
1986 if (get_bits1(&gb))
1987 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1988 skip_bits(&gb, 2); /* reserved? */
1989 }
1990 } else {
1991 if (!s->golden_frame.data[0]) {
1992 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1993
1994 s->golden_frame.reference = 3;
1995 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1996 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1997 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1998 goto error;
1999 }
2000 s->last_frame = s->golden_frame;
2001 s->last_frame.type = FF_BUFFER_TYPE_COPY;
2002 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2003 }
2004 }
2005
2006 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2007 ff_thread_finish_setup(avctx);
2008
2009 if (unpack_superblocks(s, &gb)){
2010 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2011 goto error;
2012 }
2013 if (unpack_modes(s, &gb)){
2014 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2015 goto error;
2016 }
2017 if (unpack_vectors(s, &gb)){
2018 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2019 goto error;
2020 }
2021 if (unpack_block_qpis(s, &gb)){
2022 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2023 goto error;
2024 }
2025 if (unpack_dct_coeffs(s, &gb)){
2026 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2027 goto error;
2028 }
2029
2030 for (i = 0; i < 3; i++) {
2031 int height = s->height >> (i && s->chroma_y_shift);
2032 if (s->flipped_image)
2033 s->data_offset[i] = 0;
2034 else
2035 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
2036 }
2037
2038 s->last_slice_end = 0;
2039 for (i = 0; i < s->c_superblock_height; i++)
2040 render_slice(s, i);
2041
2042 // filter the last row
2043 for (i = 0; i < 3; i++) {
2044 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
2045 apply_loop_filter(s, i, row, row+1);
2046 }
2047 vp3_draw_horiz_band(s, s->avctx->height);
2048
2049 *got_frame = 1;
2050 *(AVFrame*)data= s->current_frame;
2051
2052 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2053 update_frames(avctx);
2054
2055 return buf_size;
2056
2057 error:
2058 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2059
2060 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2061 avctx->release_buffer(avctx, &s->current_frame);
2062
2063 return -1;
2064 }
2065
2066 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2067 {
2068 Vp3DecodeContext *s = avctx->priv_data;
2069
2070 if (get_bits1(gb)) {
2071 int token;
2072 if (s->entries >= 32) { /* overflow */
2073 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2074 return -1;
2075 }
2076 token = get_bits(gb, 5);
2077 av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2078 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2079 s->huffman_table[s->hti][token][0] = s->hbits;
2080 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2081 s->entries++;
2082 }
2083 else {
2084 if (s->huff_code_size >= 32) {/* overflow */
2085 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2086 return -1;
2087 }
2088 s->huff_code_size++;
2089 s->hbits <<= 1;
2090 if (read_huffman_tree(avctx, gb))
2091 return -1;
2092 s->hbits |= 1;
2093 if (read_huffman_tree(avctx, gb))
2094 return -1;
2095 s->hbits >>= 1;
2096 s->huff_code_size--;
2097 }
2098 return 0;
2099 }
2100
2101 static int vp3_init_thread_copy(AVCodecContext *avctx)
2102 {
2103 Vp3DecodeContext *s = avctx->priv_data;
2104
2105 s->superblock_coding = NULL;
2106 s->all_fragments = NULL;
2107 s->coded_fragment_list[0] = NULL;
2108 s->dct_tokens_base = NULL;
2109 s->superblock_fragments = NULL;
2110 s->macroblock_coding = NULL;
2111 s->motion_val[0] = NULL;
2112 s->motion_val[1] = NULL;
2113 s->edge_emu_buffer = NULL;
2114
2115 return 0;
2116 }
2117
2118 #if CONFIG_THEORA_DECODER
2119 static const enum AVPixelFormat theora_pix_fmts[4] = {
2120 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2121 };
2122
2123 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2124 {
2125 Vp3DecodeContext *s = avctx->priv_data;
2126 int visible_width, visible_height, colorspace;
2127 int offset_x = 0, offset_y = 0;
2128 AVRational fps, aspect;
2129
2130 s->theora = get_bits_long(gb, 24);
2131 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2132
2133 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2134 /* but previous versions have the image flipped relative to vp3 */
2135 if (s->theora < 0x030200)
2136 {
2137 s->flipped_image = 1;
2138 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2139 }
2140
2141 visible_width = s->width = get_bits(gb, 16) << 4;
2142 visible_height = s->height = get_bits(gb, 16) << 4;
2143
2144 if(av_image_check_size(s->width, s->height, 0, avctx)){
2145 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2146 s->width= s->height= 0;
2147 return -1;
2148 }
2149
2150 if (s->theora >= 0x030200) {
2151 visible_width = get_bits_long(gb, 24);
2152 visible_height = get_bits_long(gb, 24);
2153
2154 offset_x = get_bits(gb, 8); /* offset x */
2155 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2156 }
2157
2158 fps.num = get_bits_long(gb, 32);
2159 fps.den = get_bits_long(gb, 32);
2160 if (fps.num && fps.den) {
2161 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2162 fps.den, fps.num, 1<<30);
2163 }
2164
2165 aspect.num = get_bits_long(gb, 24);
2166 aspect.den = get_bits_long(gb, 24);
2167 if (aspect.num && aspect.den) {
2168 av_reduce(&avctx->sample_aspect_ratio.num,
2169 &avctx->sample_aspect_ratio.den,
2170 aspect.num, aspect.den, 1<<30);
2171 }
2172
2173 if (s->theora < 0x030200)
2174 skip_bits(gb, 5); /* keyframe frequency force */
2175 colorspace = get_bits(gb, 8);
2176 skip_bits(gb, 24); /* bitrate */
2177
2178 skip_bits(gb, 6); /* quality hint */
2179
2180 if (s->theora >= 0x030200)
2181 {
2182 skip_bits(gb, 5); /* keyframe frequency force */
2183 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2184 skip_bits(gb, 3); /* reserved */
2185 }
2186
2187 // align_get_bits(gb);
2188
2189 if ( visible_width <= s->width && visible_width > s->width-16
2190 && visible_height <= s->height && visible_height > s->height-16
2191 && !offset_x && (offset_y == s->height - visible_height))
2192 avcodec_set_dimensions(avctx, visible_width, visible_height);
2193 else
2194 avcodec_set_dimensions(avctx, s->width, s->height);
2195
2196 if (colorspace == 1) {
2197 avctx->color_primaries = AVCOL_PRI_BT470M;
2198 } else if (colorspace == 2) {
2199 avctx->color_primaries = AVCOL_PRI_BT470BG;
2200 }
2201 if (colorspace == 1 || colorspace == 2) {
2202 avctx->colorspace = AVCOL_SPC_BT470BG;
2203 avctx->color_trc = AVCOL_TRC_BT709;
2204 }
2205
2206 return 0;
2207 }
2208
2209 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2210 {
2211 Vp3DecodeContext *s = avctx->priv_data;
2212 int i, n, matrices, inter, plane;
2213
2214 if (s->theora >= 0x030200) {
2215 n = get_bits(gb, 3);
2216 /* loop filter limit values table */
2217 if (n)
2218 for (i = 0; i < 64; i++)
2219 s->filter_limit_values[i] = get_bits(gb, n);
2220 }
2221
2222 if (s->theora >= 0x030200)
2223 n = get_bits(gb, 4) + 1;
2224 else
2225 n = 16;
2226 /* quality threshold table */
2227 for (i = 0; i < 64; i++)
2228 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2229
2230 if (s->theora >= 0x030200)
2231 n = get_bits(gb, 4) + 1;
2232 else
2233 n = 16;
2234 /* dc scale factor table */
2235 for (i = 0; i < 64; i++)
2236 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2237
2238 if (s->theora >= 0x030200)
2239 matrices = get_bits(gb, 9) + 1;
2240 else
2241 matrices = 3;
2242
2243 if(matrices > 384){
2244 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2245 return -1;
2246 }
2247
2248 for(n=0; n<matrices; n++){
2249 for (i = 0; i < 64; i++)
2250 s->base_matrix[n][i]= get_bits(gb, 8);
2251 }
2252
2253 for (inter = 0; inter <= 1; inter++) {
2254 for (plane = 0; plane <= 2; plane++) {
2255 int newqr= 1;
2256 if (inter || plane > 0)
2257 newqr = get_bits1(gb);
2258 if (!newqr) {
2259 int qtj, plj;
2260 if(inter && get_bits1(gb)){
2261 qtj = 0;
2262 plj = plane;
2263 }else{
2264 qtj= (3*inter + plane - 1) / 3;
2265 plj= (plane + 2) % 3;
2266 }
2267 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2268 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2269 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2270 } else {
2271 int qri= 0;
2272 int qi = 0;
2273
2274 for(;;){
2275 i= get_bits(gb, av_log2(matrices-1)+1);
2276 if(i>= matrices){
2277 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2278 return -1;
2279 }
2280 s->qr_base[inter][plane][qri]= i;
2281 if(qi >= 63)
2282 break;
2283 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2284 s->qr_size[inter][plane][qri++]= i;
2285 qi += i;
2286 }
2287
2288 if (qi > 63) {
2289 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2290 return -1;
2291 }
2292 s->qr_count[inter][plane]= qri;
2293 }
2294 }
2295 }
2296
2297 /* Huffman tables */
2298 for (s->hti = 0; s->hti < 80; s->hti++) {
2299 s->entries = 0;
2300 s->huff_code_size = 1;
2301 if (!get_bits1(gb)) {
2302 s->hbits = 0;
2303 if(read_huffman_tree(avctx, gb))
2304 return -1;
2305 s->hbits = 1;
2306 if(read_huffman_tree(avctx, gb))
2307 return -1;
2308 }
2309 }
2310
2311 s->theora_tables = 1;
2312
2313 return 0;
2314 }
2315
2316 static av_cold int theora_decode_init(AVCodecContext *avctx)
2317 {
2318 Vp3DecodeContext *s = avctx->priv_data;
2319 GetBitContext gb;
2320 int ptype;
2321 uint8_t *header_start[3];
2322 int header_len[3];
2323 int i;
2324
2325 s->theora = 1;
2326
2327 if (!avctx->extradata_size)
2328 {
2329 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2330 return -1;
2331 }
2332
2333 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2334 42, header_start, header_len) < 0) {
2335 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2336 return -1;
2337 }
2338
2339 for(i=0;i<3;i++) {
2340 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2341
2342 ptype = get_bits(&gb, 8);
2343
2344 if (!(ptype & 0x80))
2345 {
2346 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2347 // return -1;
2348 }
2349
2350 // FIXME: Check for this as well.
2351 skip_bits_long(&gb, 6*8); /* "theora" */
2352
2353 switch(ptype)
2354 {
2355 case 0x80:
2356 theora_decode_header(avctx, &gb);
2357 break;
2358 case 0x81:
2359 // FIXME: is this needed? it breaks sometimes
2360 // theora_decode_comments(avctx, gb);
2361 break;
2362 case 0x82:
2363 if (theora_decode_tables(avctx, &gb))
2364 return -1;
2365 break;
2366 default:
2367 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2368 break;
2369 }
2370 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2371 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2372 if (s->theora < 0x030200)
2373 break;
2374 }
2375
2376 return vp3_decode_init(avctx);
2377 }
2378
2379 AVCodec ff_theora_decoder = {
2380 .name = "theora",
2381 .type = AVMEDIA_TYPE_VIDEO,
2382 .id = AV_CODEC_ID_THEORA,
2383 .priv_data_size = sizeof(Vp3DecodeContext),
2384 .init = theora_decode_init,
2385 .close = vp3_decode_end,
2386 .decode = vp3_decode_frame,
2387 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2388 CODEC_CAP_FRAME_THREADS,
2389 .flush = vp3_decode_flush,
2390 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2391 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2392 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2393 };
2394 #endif
2395
2396 AVCodec ff_vp3_decoder = {
2397 .name = "vp3",
2398 .type = AVMEDIA_TYPE_VIDEO,
2399 .id = AV_CODEC_ID_VP3,
2400 .priv_data_size = sizeof(Vp3DecodeContext),
2401 .init = vp3_decode_init,
2402 .close = vp3_decode_end,
2403 .decode = vp3_decode_frame,
2404 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2405 CODEC_CAP_FRAME_THREADS,
2406 .flush = vp3_decode_flush,
2407 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2408 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2409 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
2410 };