vp3dec: Check coefficient index in vp3_dequant()
[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 //FIXME split things out into their own arrays
49 typedef struct Vp3Fragment {
50 int16_t dc;
51 uint8_t coding_method;
52 uint8_t qpi;
53 } Vp3Fragment;
54
55 #define SB_NOT_CODED 0
56 #define SB_PARTIALLY_CODED 1
57 #define SB_FULLY_CODED 2
58
59 // This is the maximum length of a single long bit run that can be encoded
60 // for superblock coding or block qps. Theora special-cases this to read a
61 // bit instead of flipping the current bit to allow for runs longer than 4129.
62 #define MAXIMUM_LONG_BIT_RUN 4129
63
64 #define MODE_INTER_NO_MV 0
65 #define MODE_INTRA 1
66 #define MODE_INTER_PLUS_MV 2
67 #define MODE_INTER_LAST_MV 3
68 #define MODE_INTER_PRIOR_LAST 4
69 #define MODE_USING_GOLDEN 5
70 #define MODE_GOLDEN_MV 6
71 #define MODE_INTER_FOURMV 7
72 #define CODING_MODE_COUNT 8
73
74 /* special internal mode */
75 #define MODE_COPY 8
76
77 /* There are 6 preset schemes, plus a free-form scheme */
78 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
79 {
80 /* scheme 1: Last motion vector dominates */
81 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
82 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
83 MODE_INTRA, MODE_USING_GOLDEN,
84 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
85
86 /* scheme 2 */
87 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
88 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
89 MODE_INTRA, MODE_USING_GOLDEN,
90 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
91
92 /* scheme 3 */
93 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
94 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
95 MODE_INTRA, MODE_USING_GOLDEN,
96 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
97
98 /* scheme 4 */
99 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
100 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
101 MODE_INTRA, MODE_USING_GOLDEN,
102 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
103
104 /* scheme 5: No motion vector dominates */
105 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
106 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
107 MODE_INTRA, MODE_USING_GOLDEN,
108 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
109
110 /* scheme 6 */
111 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
112 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
113 MODE_INTER_PLUS_MV, MODE_INTRA,
114 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
115
116 };
117
118 static const uint8_t hilbert_offset[16][2] = {
119 {0,0}, {1,0}, {1,1}, {0,1},
120 {0,2}, {0,3}, {1,3}, {1,2},
121 {2,2}, {2,3}, {3,3}, {3,2},
122 {3,1}, {2,1}, {2,0}, {3,0}
123 };
124
125 #define MIN_DEQUANT_VAL 2
126
127 typedef struct Vp3DecodeContext {
128 AVCodecContext *avctx;
129 int theora, theora_tables;
130 int version;
131 int width, height;
132 int chroma_x_shift, chroma_y_shift;
133 AVFrame golden_frame;
134 AVFrame last_frame;
135 AVFrame current_frame;
136 int keyframe;
137 DSPContext dsp;
138 int flipped_image;
139 int last_slice_end;
140 int skip_loop_filter;
141
142 int qps[3];
143 int nqps;
144 int last_qps[3];
145
146 int superblock_count;
147 int y_superblock_width;
148 int y_superblock_height;
149 int y_superblock_count;
150 int c_superblock_width;
151 int c_superblock_height;
152 int c_superblock_count;
153 int u_superblock_start;
154 int v_superblock_start;
155 unsigned char *superblock_coding;
156
157 int macroblock_count;
158 int macroblock_width;
159 int macroblock_height;
160
161 int fragment_count;
162 int fragment_width[2];
163 int fragment_height[2];
164
165 Vp3Fragment *all_fragments;
166 int fragment_start[3];
167 int data_offset[3];
168
169 int8_t (*motion_val[2])[2];
170
171 ScanTable scantable;
172
173 /* tables */
174 uint16_t coded_dc_scale_factor[64];
175 uint32_t coded_ac_scale_factor[64];
176 uint8_t base_matrix[384][64];
177 uint8_t qr_count[2][3];
178 uint8_t qr_size [2][3][64];
179 uint16_t qr_base[2][3][64];
180
181 /**
182 * This is a list of all tokens in bitstream order. Reordering takes place
183 * by pulling from each level during IDCT. As a consequence, IDCT must be
184 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
185 * otherwise. The 32 different tokens with up to 12 bits of extradata are
186 * collapsed into 3 types, packed as follows:
187 * (from the low to high bits)
188 *
189 * 2 bits: type (0,1,2)
190 * 0: EOB run, 14 bits for run length (12 needed)
191 * 1: zero run, 7 bits for run length
192 * 7 bits for the next coefficient (3 needed)
193 * 2: coefficient, 14 bits (11 needed)
194 *
195 * Coefficients are signed, so are packed in the highest bits for automatic
196 * sign extension.
197 */
198 int16_t *dct_tokens[3][64];
199 int16_t *dct_tokens_base;
200 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
201 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
202 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
203
204 /**
205 * number of blocks that contain DCT coefficients at the given level or higher
206 */
207 int num_coded_frags[3][64];
208 int total_num_coded_frags;
209
210 /* this is a list of indexes into the all_fragments array indicating
211 * which of the fragments are coded */
212 int *coded_fragment_list[3];
213
214 VLC dc_vlc[16];
215 VLC ac_vlc_1[16];
216 VLC ac_vlc_2[16];
217 VLC ac_vlc_3[16];
218 VLC ac_vlc_4[16];
219
220 VLC superblock_run_length_vlc;
221 VLC fragment_run_length_vlc;
222 VLC mode_code_vlc;
223 VLC motion_vector_vlc;
224
225 /* these arrays need to be on 16-byte boundaries since SSE2 operations
226 * index into them */
227 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
228
229 /* This table contains superblock_count * 16 entries. Each set of 16
230 * numbers corresponds to the fragment indexes 0..15 of the superblock.
231 * An entry will be -1 to indicate that no entry corresponds to that
232 * index. */
233 int *superblock_fragments;
234
235 /* This is an array that indicates how a particular macroblock
236 * is coded. */
237 unsigned char *macroblock_coding;
238
239 uint8_t *edge_emu_buffer;
240
241 /* Huffman decode */
242 int hti;
243 unsigned int hbits;
244 int entries;
245 int huff_code_size;
246 uint32_t huffman_table[80][32][2];
247
248 uint8_t filter_limit_values[64];
249 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
250 } Vp3DecodeContext;
251
252 /************************************************************************
253 * VP3 specific functions
254 ************************************************************************/
255
256 static void vp3_decode_flush(AVCodecContext *avctx)
257 {
258 Vp3DecodeContext *s = avctx->priv_data;
259
260 if (s->golden_frame.data[0]) {
261 if (s->golden_frame.data[0] == s->last_frame.data[0])
262 memset(&s->last_frame, 0, sizeof(AVFrame));
263 if (s->current_frame.data[0] == s->golden_frame.data[0])
264 memset(&s->current_frame, 0, sizeof(AVFrame));
265 ff_thread_release_buffer(avctx, &s->golden_frame);
266 }
267 if (s->last_frame.data[0]) {
268 if (s->current_frame.data[0] == s->last_frame.data[0])
269 memset(&s->current_frame, 0, sizeof(AVFrame));
270 ff_thread_release_buffer(avctx, &s->last_frame);
271 }
272 if (s->current_frame.data[0])
273 ff_thread_release_buffer(avctx, &s->current_frame);
274 }
275
276 static av_cold int vp3_decode_end(AVCodecContext *avctx)
277 {
278 Vp3DecodeContext *s = avctx->priv_data;
279 int i;
280
281 av_free(s->superblock_coding);
282 av_free(s->all_fragments);
283 av_free(s->coded_fragment_list[0]);
284 av_free(s->dct_tokens_base);
285 av_free(s->superblock_fragments);
286 av_free(s->macroblock_coding);
287 av_free(s->motion_val[0]);
288 av_free(s->motion_val[1]);
289 av_free(s->edge_emu_buffer);
290
291 if (avctx->internal->is_copy)
292 return 0;
293
294 for (i = 0; i < 16; i++) {
295 free_vlc(&s->dc_vlc[i]);
296 free_vlc(&s->ac_vlc_1[i]);
297 free_vlc(&s->ac_vlc_2[i]);
298 free_vlc(&s->ac_vlc_3[i]);
299 free_vlc(&s->ac_vlc_4[i]);
300 }
301
302 free_vlc(&s->superblock_run_length_vlc);
303 free_vlc(&s->fragment_run_length_vlc);
304 free_vlc(&s->mode_code_vlc);
305 free_vlc(&s->motion_vector_vlc);
306
307 /* release all frames */
308 vp3_decode_flush(avctx);
309
310 return 0;
311 }
312
313 /*
314 * This function sets up all of the various blocks mappings:
315 * superblocks <-> fragments, macroblocks <-> fragments,
316 * superblocks <-> macroblocks
317 *
318 * @return 0 is successful; returns 1 if *anything* went wrong.
319 */
320 static int init_block_mapping(Vp3DecodeContext *s)
321 {
322 int sb_x, sb_y, plane;
323 int x, y, i, j = 0;
324
325 for (plane = 0; plane < 3; plane++) {
326 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
327 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
328 int frag_width = s->fragment_width[!!plane];
329 int frag_height = s->fragment_height[!!plane];
330
331 for (sb_y = 0; sb_y < sb_height; sb_y++)
332 for (sb_x = 0; sb_x < sb_width; sb_x++)
333 for (i = 0; i < 16; i++) {
334 x = 4*sb_x + hilbert_offset[i][0];
335 y = 4*sb_y + hilbert_offset[i][1];
336
337 if (x < frag_width && y < frag_height)
338 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
339 else
340 s->superblock_fragments[j++] = -1;
341 }
342 }
343
344 return 0; /* successful path out */
345 }
346
347 /*
348 * This function sets up the dequantization tables used for a particular
349 * frame.
350 */
351 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
352 {
353 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
354 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
355 int i, plane, inter, qri, bmi, bmj, qistart;
356
357 for(inter=0; inter<2; inter++){
358 for(plane=0; plane<3; plane++){
359 int sum=0;
360 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
361 sum+= s->qr_size[inter][plane][qri];
362 if(s->qps[qpi] <= sum)
363 break;
364 }
365 qistart= sum - s->qr_size[inter][plane][qri];
366 bmi= s->qr_base[inter][plane][qri ];
367 bmj= s->qr_base[inter][plane][qri+1];
368 for(i=0; i<64; i++){
369 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
370 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
371 + s->qr_size[inter][plane][qri])
372 / (2*s->qr_size[inter][plane][qri]);
373
374 int qmin= 8<<(inter + !i);
375 int qscale= i ? ac_scale_factor : dc_scale_factor;
376
377 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
378 }
379 // all DC coefficients use the same quant so as not to interfere with DC prediction
380 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
381 }
382 }
383 }
384
385 /*
386 * This function initializes the loop filter boundary limits if the frame's
387 * quality index is different from the previous frame's.
388 *
389 * The filter_limit_values may not be larger than 127.
390 */
391 static void init_loop_filter(Vp3DecodeContext *s)
392 {
393 int *bounding_values= s->bounding_values_array+127;
394 int filter_limit;
395 int x;
396 int value;
397
398 filter_limit = s->filter_limit_values[s->qps[0]];
399
400 /* set up the bounding values */
401 memset(s->bounding_values_array, 0, 256 * sizeof(int));
402 for (x = 0; x < filter_limit; x++) {
403 bounding_values[-x] = -x;
404 bounding_values[x] = x;
405 }
406 for (x = value = filter_limit; x < 128 && value; x++, value--) {
407 bounding_values[ x] = value;
408 bounding_values[-x] = -value;
409 }
410 if (value)
411 bounding_values[128] = value;
412 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
413 }
414
415 /*
416 * This function unpacks all of the superblock/macroblock/fragment coding
417 * information from the bitstream.
418 */
419 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
420 {
421 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
422 int bit = 0;
423 int current_superblock = 0;
424 int current_run = 0;
425 int num_partial_superblocks = 0;
426
427 int i, j;
428 int current_fragment;
429 int plane;
430
431 if (s->keyframe) {
432 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
433
434 } else {
435
436 /* unpack the list of partially-coded superblocks */
437 bit = get_bits1(gb) ^ 1;
438 current_run = 0;
439
440 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
441 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
442 bit = get_bits1(gb);
443 else
444 bit ^= 1;
445
446 current_run = get_vlc2(gb,
447 s->superblock_run_length_vlc.table, 6, 2) + 1;
448 if (current_run == 34)
449 current_run += get_bits(gb, 12);
450
451 if (current_superblock + current_run > s->superblock_count) {
452 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
453 return -1;
454 }
455
456 memset(s->superblock_coding + current_superblock, bit, current_run);
457
458 current_superblock += current_run;
459 if (bit)
460 num_partial_superblocks += current_run;
461 }
462
463 /* unpack the list of fully coded superblocks if any of the blocks were
464 * not marked as partially coded in the previous step */
465 if (num_partial_superblocks < s->superblock_count) {
466 int superblocks_decoded = 0;
467
468 current_superblock = 0;
469 bit = get_bits1(gb) ^ 1;
470 current_run = 0;
471
472 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
473 && get_bits_left(gb) > 0) {
474
475 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
476 bit = get_bits1(gb);
477 else
478 bit ^= 1;
479
480 current_run = get_vlc2(gb,
481 s->superblock_run_length_vlc.table, 6, 2) + 1;
482 if (current_run == 34)
483 current_run += get_bits(gb, 12);
484
485 for (j = 0; j < current_run; current_superblock++) {
486 if (current_superblock >= s->superblock_count) {
487 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
488 return -1;
489 }
490
491 /* skip any superblocks already marked as partially coded */
492 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
493 s->superblock_coding[current_superblock] = 2*bit;
494 j++;
495 }
496 }
497 superblocks_decoded += current_run;
498 }
499 }
500
501 /* if there were partial blocks, initialize bitstream for
502 * unpacking fragment codings */
503 if (num_partial_superblocks) {
504
505 current_run = 0;
506 bit = get_bits1(gb);
507 /* toggle the bit because as soon as the first run length is
508 * fetched the bit will be toggled again */
509 bit ^= 1;
510 }
511 }
512
513 /* figure out which fragments are coded; iterate through each
514 * superblock (all planes) */
515 s->total_num_coded_frags = 0;
516 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
517
518 for (plane = 0; plane < 3; plane++) {
519 int sb_start = superblock_starts[plane];
520 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
521 int num_coded_frags = 0;
522
523 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
524
525 /* iterate through all 16 fragments in a superblock */
526 for (j = 0; j < 16; j++) {
527
528 /* if the fragment is in bounds, check its coding status */
529 current_fragment = s->superblock_fragments[i * 16 + j];
530 if (current_fragment != -1) {
531 int coded = s->superblock_coding[i];
532
533 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
534
535 /* fragment may or may not be coded; this is the case
536 * that cares about the fragment coding runs */
537 if (current_run-- == 0) {
538 bit ^= 1;
539 current_run = get_vlc2(gb,
540 s->fragment_run_length_vlc.table, 5, 2);
541 }
542 coded = bit;
543 }
544
545 if (coded) {
546 /* default mode; actual mode will be decoded in
547 * the next phase */
548 s->all_fragments[current_fragment].coding_method =
549 MODE_INTER_NO_MV;
550 s->coded_fragment_list[plane][num_coded_frags++] =
551 current_fragment;
552 } else {
553 /* not coded; copy this fragment from the prior frame */
554 s->all_fragments[current_fragment].coding_method =
555 MODE_COPY;
556 }
557 }
558 }
559 }
560 s->total_num_coded_frags += num_coded_frags;
561 for (i = 0; i < 64; i++)
562 s->num_coded_frags[plane][i] = num_coded_frags;
563 if (plane < 2)
564 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
565 }
566 return 0;
567 }
568
569 /*
570 * This function unpacks all the coding mode data for individual macroblocks
571 * from the bitstream.
572 */
573 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
574 {
575 int i, j, k, sb_x, sb_y;
576 int scheme;
577 int current_macroblock;
578 int current_fragment;
579 int coding_mode;
580 int custom_mode_alphabet[CODING_MODE_COUNT];
581 const int *alphabet;
582 Vp3Fragment *frag;
583
584 if (s->keyframe) {
585 for (i = 0; i < s->fragment_count; i++)
586 s->all_fragments[i].coding_method = MODE_INTRA;
587
588 } else {
589
590 /* fetch the mode coding scheme for this frame */
591 scheme = get_bits(gb, 3);
592
593 /* is it a custom coding scheme? */
594 if (scheme == 0) {
595 for (i = 0; i < 8; i++)
596 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
597 for (i = 0; i < 8; i++)
598 custom_mode_alphabet[get_bits(gb, 3)] = i;
599 alphabet = custom_mode_alphabet;
600 } else
601 alphabet = ModeAlphabet[scheme-1];
602
603 /* iterate through all of the macroblocks that contain 1 or more
604 * coded fragments */
605 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
606 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
607 if (get_bits_left(gb) <= 0)
608 return -1;
609
610 for (j = 0; j < 4; j++) {
611 int mb_x = 2*sb_x + (j>>1);
612 int mb_y = 2*sb_y + (((j>>1)+j)&1);
613 current_macroblock = mb_y * s->macroblock_width + mb_x;
614
615 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
616 continue;
617
618 #define BLOCK_X (2*mb_x + (k&1))
619 #define BLOCK_Y (2*mb_y + (k>>1))
620 /* coding modes are only stored if the macroblock has at least one
621 * luma block coded, otherwise it must be INTER_NO_MV */
622 for (k = 0; k < 4; k++) {
623 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
624 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
625 break;
626 }
627 if (k == 4) {
628 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
629 continue;
630 }
631
632 /* mode 7 means get 3 bits for each coding mode */
633 if (scheme == 7)
634 coding_mode = get_bits(gb, 3);
635 else
636 coding_mode = alphabet
637 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
638
639 s->macroblock_coding[current_macroblock] = coding_mode;
640 for (k = 0; k < 4; k++) {
641 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
642 if (frag->coding_method != MODE_COPY)
643 frag->coding_method = coding_mode;
644 }
645
646 #define SET_CHROMA_MODES \
647 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
648 frag[s->fragment_start[1]].coding_method = coding_mode;\
649 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
650 frag[s->fragment_start[2]].coding_method = coding_mode;
651
652 if (s->chroma_y_shift) {
653 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
654 SET_CHROMA_MODES
655 } else if (s->chroma_x_shift) {
656 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
657 for (k = 0; k < 2; k++) {
658 SET_CHROMA_MODES
659 frag += s->fragment_width[1];
660 }
661 } else {
662 for (k = 0; k < 4; k++) {
663 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
664 SET_CHROMA_MODES
665 }
666 }
667 }
668 }
669 }
670 }
671
672 return 0;
673 }
674
675 /*
676 * This function unpacks all the motion vectors for the individual
677 * macroblocks from the bitstream.
678 */
679 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
680 {
681 int j, k, sb_x, sb_y;
682 int coding_mode;
683 int motion_x[4];
684 int motion_y[4];
685 int last_motion_x = 0;
686 int last_motion_y = 0;
687 int prior_last_motion_x = 0;
688 int prior_last_motion_y = 0;
689 int current_macroblock;
690 int current_fragment;
691 int frag;
692
693 if (s->keyframe)
694 return 0;
695
696 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
697 coding_mode = get_bits1(gb);
698
699 /* iterate through all of the macroblocks that contain 1 or more
700 * coded fragments */
701 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
702 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
703 if (get_bits_left(gb) <= 0)
704 return -1;
705
706 for (j = 0; j < 4; j++) {
707 int mb_x = 2*sb_x + (j>>1);
708 int mb_y = 2*sb_y + (((j>>1)+j)&1);
709 current_macroblock = mb_y * s->macroblock_width + mb_x;
710
711 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
712 (s->macroblock_coding[current_macroblock] == MODE_COPY))
713 continue;
714
715 switch (s->macroblock_coding[current_macroblock]) {
716
717 case MODE_INTER_PLUS_MV:
718 case MODE_GOLDEN_MV:
719 /* all 6 fragments use the same motion vector */
720 if (coding_mode == 0) {
721 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
722 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
723 } else {
724 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
725 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
726 }
727
728 /* vector maintenance, only on MODE_INTER_PLUS_MV */
729 if (s->macroblock_coding[current_macroblock] ==
730 MODE_INTER_PLUS_MV) {
731 prior_last_motion_x = last_motion_x;
732 prior_last_motion_y = last_motion_y;
733 last_motion_x = motion_x[0];
734 last_motion_y = motion_y[0];
735 }
736 break;
737
738 case MODE_INTER_FOURMV:
739 /* vector maintenance */
740 prior_last_motion_x = last_motion_x;
741 prior_last_motion_y = last_motion_y;
742
743 /* fetch 4 vectors from the bitstream, one for each
744 * Y fragment, then average for the C fragment vectors */
745 for (k = 0; k < 4; k++) {
746 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
747 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
748 if (coding_mode == 0) {
749 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
750 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
751 } else {
752 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
753 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
754 }
755 last_motion_x = motion_x[k];
756 last_motion_y = motion_y[k];
757 } else {
758 motion_x[k] = 0;
759 motion_y[k] = 0;
760 }
761 }
762 break;
763
764 case MODE_INTER_LAST_MV:
765 /* all 6 fragments use the last motion vector */
766 motion_x[0] = last_motion_x;
767 motion_y[0] = last_motion_y;
768
769 /* no vector maintenance (last vector remains the
770 * last vector) */
771 break;
772
773 case MODE_INTER_PRIOR_LAST:
774 /* all 6 fragments use the motion vector prior to the
775 * last motion vector */
776 motion_x[0] = prior_last_motion_x;
777 motion_y[0] = prior_last_motion_y;
778
779 /* vector maintenance */
780 prior_last_motion_x = last_motion_x;
781 prior_last_motion_y = last_motion_y;
782 last_motion_x = motion_x[0];
783 last_motion_y = motion_y[0];
784 break;
785
786 default:
787 /* covers intra, inter without MV, golden without MV */
788 motion_x[0] = 0;
789 motion_y[0] = 0;
790
791 /* no vector maintenance */
792 break;
793 }
794
795 /* assign the motion vectors to the correct fragments */
796 for (k = 0; k < 4; k++) {
797 current_fragment =
798 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
799 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
800 s->motion_val[0][current_fragment][0] = motion_x[k];
801 s->motion_val[0][current_fragment][1] = motion_y[k];
802 } else {
803 s->motion_val[0][current_fragment][0] = motion_x[0];
804 s->motion_val[0][current_fragment][1] = motion_y[0];
805 }
806 }
807
808 if (s->chroma_y_shift) {
809 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
810 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
811 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
812 }
813 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
814 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
815 frag = mb_y*s->fragment_width[1] + mb_x;
816 s->motion_val[1][frag][0] = motion_x[0];
817 s->motion_val[1][frag][1] = motion_y[0];
818 } else if (s->chroma_x_shift) {
819 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
820 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
821 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
822 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
823 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
824 } else {
825 motion_x[1] = motion_x[0];
826 motion_y[1] = motion_y[0];
827 }
828 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
829 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
830
831 frag = 2*mb_y*s->fragment_width[1] + mb_x;
832 for (k = 0; k < 2; k++) {
833 s->motion_val[1][frag][0] = motion_x[k];
834 s->motion_val[1][frag][1] = motion_y[k];
835 frag += s->fragment_width[1];
836 }
837 } else {
838 for (k = 0; k < 4; k++) {
839 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
840 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
841 s->motion_val[1][frag][0] = motion_x[k];
842 s->motion_val[1][frag][1] = motion_y[k];
843 } else {
844 s->motion_val[1][frag][0] = motion_x[0];
845 s->motion_val[1][frag][1] = motion_y[0];
846 }
847 }
848 }
849 }
850 }
851 }
852
853 return 0;
854 }
855
856 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
857 {
858 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
859 int num_blocks = s->total_num_coded_frags;
860
861 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
862 i = blocks_decoded = num_blocks_at_qpi = 0;
863
864 bit = get_bits1(gb) ^ 1;
865 run_length = 0;
866
867 do {
868 if (run_length == MAXIMUM_LONG_BIT_RUN)
869 bit = get_bits1(gb);
870 else
871 bit ^= 1;
872
873 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
874 if (run_length == 34)
875 run_length += get_bits(gb, 12);
876 blocks_decoded += run_length;
877
878 if (!bit)
879 num_blocks_at_qpi += run_length;
880
881 for (j = 0; j < run_length; i++) {
882 if (i >= s->total_num_coded_frags)
883 return -1;
884
885 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
886 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
887 j++;
888 }
889 }
890 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
891
892 num_blocks -= num_blocks_at_qpi;
893 }
894
895 return 0;
896 }
897
898 /*
899 * This function is called by unpack_dct_coeffs() to extract the VLCs from
900 * the bitstream. The VLCs encode tokens which are used to unpack DCT
901 * data. This function unpacks all the VLCs for either the Y plane or both
902 * C planes, and is called for DC coefficients or different AC coefficient
903 * levels (since different coefficient types require different VLC tables.
904 *
905 * This function returns a residual eob run. E.g, if a particular token gave
906 * instructions to EOB the next 5 fragments and there were only 2 fragments
907 * left in the current fragment range, 3 would be returned so that it could
908 * be passed into the next call to this same function.
909 */
910 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
911 VLC *table, int coeff_index,
912 int plane,
913 int eob_run)
914 {
915 int i, j = 0;
916 int token;
917 int zero_run = 0;
918 DCTELEM coeff = 0;
919 int bits_to_get;
920 int blocks_ended;
921 int coeff_i = 0;
922 int num_coeffs = s->num_coded_frags[plane][coeff_index];
923 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
924
925 /* local references to structure members to avoid repeated deferences */
926 int *coded_fragment_list = s->coded_fragment_list[plane];
927 Vp3Fragment *all_fragments = s->all_fragments;
928 VLC_TYPE (*vlc_table)[2] = table->table;
929
930 if (num_coeffs < 0)
931 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
932
933 if (eob_run > num_coeffs) {
934 coeff_i = blocks_ended = num_coeffs;
935 eob_run -= num_coeffs;
936 } else {
937 coeff_i = blocks_ended = eob_run;
938 eob_run = 0;
939 }
940
941 // insert fake EOB token to cover the split between planes or zzi
942 if (blocks_ended)
943 dct_tokens[j++] = blocks_ended << 2;
944
945 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
946 /* decode a VLC into a token */
947 token = get_vlc2(gb, vlc_table, 11, 3);
948 /* use the token to get a zero run, a coefficient, and an eob run */
949 if ((unsigned) token <= 6U) {
950 eob_run = eob_run_base[token];
951 if (eob_run_get_bits[token])
952 eob_run += get_bits(gb, eob_run_get_bits[token]);
953
954 // record only the number of blocks ended in this plane,
955 // any spill will be recorded in the next plane.
956 if (eob_run > num_coeffs - coeff_i) {
957 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
958 blocks_ended += num_coeffs - coeff_i;
959 eob_run -= num_coeffs - coeff_i;
960 coeff_i = num_coeffs;
961 } else {
962 dct_tokens[j++] = TOKEN_EOB(eob_run);
963 blocks_ended += eob_run;
964 coeff_i += eob_run;
965 eob_run = 0;
966 }
967 } else if (token >= 0) {
968 bits_to_get = coeff_get_bits[token];
969 if (bits_to_get)
970 bits_to_get = get_bits(gb, bits_to_get);
971 coeff = coeff_tables[token][bits_to_get];
972
973 zero_run = zero_run_base[token];
974 if (zero_run_get_bits[token])
975 zero_run += get_bits(gb, zero_run_get_bits[token]);
976
977 if (zero_run) {
978 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
979 } else {
980 // Save DC into the fragment structure. DC prediction is
981 // done in raster order, so the actual DC can't be in with
982 // other tokens. We still need the token in dct_tokens[]
983 // however, or else the structure collapses on itself.
984 if (!coeff_index)
985 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
986
987 dct_tokens[j++] = TOKEN_COEFF(coeff);
988 }
989
990 if (coeff_index + zero_run > 64) {
991 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
992 " %d coeffs left\n", zero_run, 64-coeff_index);
993 zero_run = 64 - coeff_index;
994 }
995
996 // zero runs code multiple coefficients,
997 // so don't try to decode coeffs for those higher levels
998 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
999 s->num_coded_frags[plane][i]--;
1000 coeff_i++;
1001 } else {
1002 av_log(s->avctx, AV_LOG_ERROR,
1003 "Invalid token %d\n", token);
1004 return -1;
1005 }
1006 }
1007
1008 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1009 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1010
1011 // decrement the number of blocks that have higher coeffecients for each
1012 // EOB run at this level
1013 if (blocks_ended)
1014 for (i = coeff_index+1; i < 64; i++)
1015 s->num_coded_frags[plane][i] -= blocks_ended;
1016
1017 // setup the next buffer
1018 if (plane < 2)
1019 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
1020 else if (coeff_index < 63)
1021 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
1022
1023 return eob_run;
1024 }
1025
1026 static void reverse_dc_prediction(Vp3DecodeContext *s,
1027 int first_fragment,
1028 int fragment_width,
1029 int fragment_height);
1030 /*
1031 * This function unpacks all of the DCT coefficient data from the
1032 * bitstream.
1033 */
1034 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1035 {
1036 int i;
1037 int dc_y_table;
1038 int dc_c_table;
1039 int ac_y_table;
1040 int ac_c_table;
1041 int residual_eob_run = 0;
1042 VLC *y_tables[64];
1043 VLC *c_tables[64];
1044
1045 s->dct_tokens[0][0] = s->dct_tokens_base;
1046
1047 /* fetch the DC table indexes */
1048 dc_y_table = get_bits(gb, 4);
1049 dc_c_table = get_bits(gb, 4);
1050
1051 /* unpack the Y plane DC coefficients */
1052 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1053 0, residual_eob_run);
1054 if (residual_eob_run < 0)
1055 return residual_eob_run;
1056
1057 /* reverse prediction of the Y-plane DC coefficients */
1058 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1059
1060 /* unpack the C plane DC coefficients */
1061 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1062 1, residual_eob_run);
1063 if (residual_eob_run < 0)
1064 return residual_eob_run;
1065 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1066 2, residual_eob_run);
1067 if (residual_eob_run < 0)
1068 return residual_eob_run;
1069
1070 /* reverse prediction of the C-plane DC coefficients */
1071 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1072 {
1073 reverse_dc_prediction(s, s->fragment_start[1],
1074 s->fragment_width[1], s->fragment_height[1]);
1075 reverse_dc_prediction(s, s->fragment_start[2],
1076 s->fragment_width[1], s->fragment_height[1]);
1077 }
1078
1079 /* fetch the AC table indexes */
1080 ac_y_table = get_bits(gb, 4);
1081 ac_c_table = get_bits(gb, 4);
1082
1083 /* build tables of AC VLC tables */
1084 for (i = 1; i <= 5; i++) {
1085 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1086 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1087 }
1088 for (i = 6; i <= 14; i++) {
1089 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1090 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1091 }
1092 for (i = 15; i <= 27; i++) {
1093 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1094 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1095 }
1096 for (i = 28; i <= 63; i++) {
1097 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1098 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1099 }
1100
1101 /* decode all AC coefficents */
1102 for (i = 1; i <= 63; i++) {
1103 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1104 0, residual_eob_run);
1105 if (residual_eob_run < 0)
1106 return residual_eob_run;
1107
1108 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1109 1, residual_eob_run);
1110 if (residual_eob_run < 0)
1111 return residual_eob_run;
1112 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1113 2, residual_eob_run);
1114 if (residual_eob_run < 0)
1115 return residual_eob_run;
1116 }
1117
1118 return 0;
1119 }
1120
1121 /*
1122 * This function reverses the DC prediction for each coded fragment in
1123 * the frame. Much of this function is adapted directly from the original
1124 * VP3 source code.
1125 */
1126 #define COMPATIBLE_FRAME(x) \
1127 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1128 #define DC_COEFF(u) s->all_fragments[u].dc
1129
1130 static void reverse_dc_prediction(Vp3DecodeContext *s,
1131 int first_fragment,
1132 int fragment_width,
1133 int fragment_height)
1134 {
1135
1136 #define PUL 8
1137 #define PU 4
1138 #define PUR 2
1139 #define PL 1
1140
1141 int x, y;
1142 int i = first_fragment;
1143
1144 int predicted_dc;
1145
1146 /* DC values for the left, up-left, up, and up-right fragments */
1147 int vl, vul, vu, vur;
1148
1149 /* indexes for the left, up-left, up, and up-right fragments */
1150 int l, ul, u, ur;
1151
1152 /*
1153 * The 6 fields mean:
1154 * 0: up-left multiplier
1155 * 1: up multiplier
1156 * 2: up-right multiplier
1157 * 3: left multiplier
1158 */
1159 static const int predictor_transform[16][4] = {
1160 { 0, 0, 0, 0},
1161 { 0, 0, 0,128}, // PL
1162 { 0, 0,128, 0}, // PUR
1163 { 0, 0, 53, 75}, // PUR|PL
1164 { 0,128, 0, 0}, // PU
1165 { 0, 64, 0, 64}, // PU|PL
1166 { 0,128, 0, 0}, // PU|PUR
1167 { 0, 0, 53, 75}, // PU|PUR|PL
1168 {128, 0, 0, 0}, // PUL
1169 { 0, 0, 0,128}, // PUL|PL
1170 { 64, 0, 64, 0}, // PUL|PUR
1171 { 0, 0, 53, 75}, // PUL|PUR|PL
1172 { 0,128, 0, 0}, // PUL|PU
1173 {-104,116, 0,116}, // PUL|PU|PL
1174 { 24, 80, 24, 0}, // PUL|PU|PUR
1175 {-104,116, 0,116} // PUL|PU|PUR|PL
1176 };
1177
1178 /* This table shows which types of blocks can use other blocks for
1179 * prediction. For example, INTRA is the only mode in this table to
1180 * have a frame number of 0. That means INTRA blocks can only predict
1181 * from other INTRA blocks. There are 2 golden frame coding types;
1182 * blocks encoding in these modes can only predict from other blocks
1183 * that were encoded with these 1 of these 2 modes. */
1184 static const unsigned char compatible_frame[9] = {
1185 1, /* MODE_INTER_NO_MV */
1186 0, /* MODE_INTRA */
1187 1, /* MODE_INTER_PLUS_MV */
1188 1, /* MODE_INTER_LAST_MV */
1189 1, /* MODE_INTER_PRIOR_MV */
1190 2, /* MODE_USING_GOLDEN */
1191 2, /* MODE_GOLDEN_MV */
1192 1, /* MODE_INTER_FOUR_MV */
1193 3 /* MODE_COPY */
1194 };
1195 int current_frame_type;
1196
1197 /* there is a last DC predictor for each of the 3 frame types */
1198 short last_dc[3];
1199
1200 int transform = 0;
1201
1202 vul = vu = vur = vl = 0;
1203 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1204
1205 /* for each fragment row... */
1206 for (y = 0; y < fragment_height; y++) {
1207
1208 /* for each fragment in a row... */
1209 for (x = 0; x < fragment_width; x++, i++) {
1210
1211 /* reverse prediction if this block was coded */
1212 if (s->all_fragments[i].coding_method != MODE_COPY) {
1213
1214 current_frame_type =
1215 compatible_frame[s->all_fragments[i].coding_method];
1216
1217 transform= 0;
1218 if(x){
1219 l= i-1;
1220 vl = DC_COEFF(l);
1221 if(COMPATIBLE_FRAME(l))
1222 transform |= PL;
1223 }
1224 if(y){
1225 u= i-fragment_width;
1226 vu = DC_COEFF(u);
1227 if(COMPATIBLE_FRAME(u))
1228 transform |= PU;
1229 if(x){
1230 ul= i-fragment_width-1;
1231 vul = DC_COEFF(ul);
1232 if(COMPATIBLE_FRAME(ul))
1233 transform |= PUL;
1234 }
1235 if(x + 1 < fragment_width){
1236 ur= i-fragment_width+1;
1237 vur = DC_COEFF(ur);
1238 if(COMPATIBLE_FRAME(ur))
1239 transform |= PUR;
1240 }
1241 }
1242
1243 if (transform == 0) {
1244
1245 /* if there were no fragments to predict from, use last
1246 * DC saved */
1247 predicted_dc = last_dc[current_frame_type];
1248 } else {
1249
1250 /* apply the appropriate predictor transform */
1251 predicted_dc =
1252 (predictor_transform[transform][0] * vul) +
1253 (predictor_transform[transform][1] * vu) +
1254 (predictor_transform[transform][2] * vur) +
1255 (predictor_transform[transform][3] * vl);
1256
1257 predicted_dc /= 128;
1258
1259 /* check for outranging on the [ul u l] and
1260 * [ul u ur l] predictors */
1261 if ((transform == 15) || (transform == 13)) {
1262 if (FFABS(predicted_dc - vu) > 128)
1263 predicted_dc = vu;
1264 else if (FFABS(predicted_dc - vl) > 128)
1265 predicted_dc = vl;
1266 else if (FFABS(predicted_dc - vul) > 128)
1267 predicted_dc = vul;
1268 }
1269 }
1270
1271 /* at long last, apply the predictor */
1272 DC_COEFF(i) += predicted_dc;
1273 /* save the DC */
1274 last_dc[current_frame_type] = DC_COEFF(i);
1275 }
1276 }
1277 }
1278 }
1279
1280 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1281 {
1282 int x, y;
1283 int *bounding_values= s->bounding_values_array+127;
1284
1285 int width = s->fragment_width[!!plane];
1286 int height = s->fragment_height[!!plane];
1287 int fragment = s->fragment_start [plane] + ystart * width;
1288 int stride = s->current_frame.linesize[plane];
1289 uint8_t *plane_data = s->current_frame.data [plane];
1290 if (!s->flipped_image) stride = -stride;
1291 plane_data += s->data_offset[plane] + 8*ystart*stride;
1292
1293 for (y = ystart; y < yend; y++) {
1294
1295 for (x = 0; x < width; x++) {
1296 /* This code basically just deblocks on the edges of coded blocks.
1297 * However, it has to be much more complicated because of the
1298 * braindamaged deblock ordering used in VP3/Theora. Order matters
1299 * because some pixels get filtered twice. */
1300 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1301 {
1302 /* do not perform left edge filter for left columns frags */
1303 if (x > 0) {
1304 s->dsp.vp3_h_loop_filter(
1305 plane_data + 8*x,
1306 stride, bounding_values);
1307 }
1308
1309 /* do not perform top edge filter for top row fragments */
1310 if (y > 0) {
1311 s->dsp.vp3_v_loop_filter(
1312 plane_data + 8*x,
1313 stride, bounding_values);
1314 }
1315
1316 /* do not perform right edge filter for right column
1317 * fragments or if right fragment neighbor is also coded
1318 * in this frame (it will be filtered in next iteration) */
1319 if ((x < width - 1) &&
1320 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1321 s->dsp.vp3_h_loop_filter(
1322 plane_data + 8*x + 8,
1323 stride, bounding_values);
1324 }
1325
1326 /* do not perform bottom edge filter for bottom row
1327 * fragments or if bottom fragment neighbor is also coded
1328 * in this frame (it will be filtered in the next row) */
1329 if ((y < height - 1) &&
1330 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1331 s->dsp.vp3_v_loop_filter(
1332 plane_data + 8*x + 8*stride,
1333 stride, bounding_values);
1334 }
1335 }
1336
1337 fragment++;
1338 }
1339 plane_data += 8*stride;
1340 }
1341 }
1342
1343 /**
1344 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1345 * for the next block in coding order
1346 */
1347 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1348 int plane, int inter, DCTELEM block[64])
1349 {
1350 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1351 uint8_t *perm = s->scantable.permutated;
1352 int i = 0;
1353
1354 do {
1355 int token = *s->dct_tokens[plane][i];
1356 switch (token & 3) {
1357 case 0: // EOB
1358 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1359 s->dct_tokens[plane][i]++;
1360 else
1361 *s->dct_tokens[plane][i] = token & ~3;
1362 goto end;
1363 case 1: // zero run
1364 s->dct_tokens[plane][i]++;
1365 i += (token >> 2) & 0x7f;
1366 if (i > 63) {
1367 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1368 return i;
1369 }
1370 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1371 i++;
1372 break;
1373 case 2: // coeff
1374 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1375 s->dct_tokens[plane][i++]++;
1376 break;
1377 default: // shouldn't happen
1378 return i;
1379 }
1380 } while (i < 64);
1381 end:
1382 // the actual DC+prediction is in the fragment structure
1383 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1384 return i;
1385 }
1386
1387 /**
1388 * called when all pixels up to row y are complete
1389 */
1390 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1391 {
1392 int h, cy, i;
1393 int offset[AV_NUM_DATA_POINTERS];
1394
1395 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1396 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1397
1398 // At the end of the frame, report INT_MAX instead of the height of the frame.
1399 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1400 // they don't have to clip their values.
1401 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1402 }
1403
1404 if(s->avctx->draw_horiz_band==NULL)
1405 return;
1406
1407 h= y - s->last_slice_end;
1408 s->last_slice_end= y;
1409 y -= h;
1410
1411 if (!s->flipped_image) {
1412 y = s->avctx->height - y - h;
1413 }
1414
1415 cy = y >> s->chroma_y_shift;
1416 offset[0] = s->current_frame.linesize[0]*y;
1417 offset[1] = s->current_frame.linesize[1]*cy;
1418 offset[2] = s->current_frame.linesize[2]*cy;
1419 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1420 offset[i] = 0;
1421
1422 emms_c();
1423 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1424 }
1425
1426 /**
1427 * Wait for the reference frame of the current fragment.
1428 * The progress value is in luma pixel rows.
1429 */
1430 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1431 {
1432 AVFrame *ref_frame;
1433 int ref_row;
1434 int border = motion_y&1;
1435
1436 if (fragment->coding_method == MODE_USING_GOLDEN ||
1437 fragment->coding_method == MODE_GOLDEN_MV)
1438 ref_frame = &s->golden_frame;
1439 else
1440 ref_frame = &s->last_frame;
1441
1442 ref_row = y + (motion_y>>1);
1443 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1444
1445 ff_thread_await_progress(ref_frame, ref_row, 0);
1446 }
1447
1448 /*
1449 * Perform the final rendering for a particular slice of data.
1450 * The slice number ranges from 0..(c_superblock_height - 1).
1451 */
1452 static void render_slice(Vp3DecodeContext *s, int slice)
1453 {
1454 int x, y, i, j, fragment;
1455 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1456 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1457 int motion_halfpel_index;
1458 uint8_t *motion_source;
1459 int plane, first_pixel;
1460
1461 if (slice >= s->c_superblock_height)
1462 return;
1463
1464 for (plane = 0; plane < 3; plane++) {
1465 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1466 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1467 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1468 int stride = s->current_frame.linesize[plane];
1469 int plane_width = s->width >> (plane && s->chroma_x_shift);
1470 int plane_height = s->height >> (plane && s->chroma_y_shift);
1471 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1472
1473 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1474 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1475 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1476
1477 int fragment_width = s->fragment_width[!!plane];
1478 int fragment_height = s->fragment_height[!!plane];
1479 int fragment_start = s->fragment_start[plane];
1480 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1481
1482 if (!s->flipped_image) stride = -stride;
1483 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1484 continue;
1485
1486 /* for each superblock row in the slice (both of them)... */
1487 for (; sb_y < slice_height; sb_y++) {
1488
1489 /* for each superblock in a row... */
1490 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1491
1492 /* for each block in a superblock... */
1493 for (j = 0; j < 16; j++) {
1494 x = 4*sb_x + hilbert_offset[j][0];
1495 y = 4*sb_y + hilbert_offset[j][1];
1496 fragment = y*fragment_width + x;
1497
1498 i = fragment_start + fragment;
1499
1500 // bounds check
1501 if (x >= fragment_width || y >= fragment_height)
1502 continue;
1503
1504 first_pixel = 8*y*stride + 8*x;
1505
1506 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1507 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1508
1509 /* transform if this block was coded */
1510 if (s->all_fragments[i].coding_method != MODE_COPY) {
1511 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1512 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1513 motion_source= golden_plane;
1514 else
1515 motion_source= last_plane;
1516
1517 motion_source += first_pixel;
1518 motion_halfpel_index = 0;
1519
1520 /* sort out the motion vector if this fragment is coded
1521 * using a motion vector method */
1522 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1523 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1524 int src_x, src_y;
1525 motion_x = motion_val[fragment][0];
1526 motion_y = motion_val[fragment][1];
1527
1528 src_x= (motion_x>>1) + 8*x;
1529 src_y= (motion_y>>1) + 8*y;
1530
1531 motion_halfpel_index = motion_x & 0x01;
1532 motion_source += (motion_x >> 1);
1533
1534 motion_halfpel_index |= (motion_y & 0x01) << 1;
1535 motion_source += ((motion_y >> 1) * stride);
1536
1537 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1538 uint8_t *temp= s->edge_emu_buffer;
1539 if(stride<0) temp -= 8*stride;
1540
1541 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1542 motion_source= temp;
1543 }
1544 }
1545
1546
1547 /* first, take care of copying a block from either the
1548 * previous or the golden frame */
1549 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1550 /* Note, it is possible to implement all MC cases with
1551 put_no_rnd_pixels_l2 which would look more like the
1552 VP3 source but this would be slower as
1553 put_no_rnd_pixels_tab is better optimzed */
1554 if(motion_halfpel_index != 3){
1555 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1556 output_plane + first_pixel,
1557 motion_source, stride, 8);
1558 }else{
1559 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1560 s->dsp.put_no_rnd_pixels_l2[1](
1561 output_plane + first_pixel,
1562 motion_source - d,
1563 motion_source + stride + 1 + d,
1564 stride, 8);
1565 }
1566 }
1567
1568 s->dsp.clear_block(block);
1569
1570 /* invert DCT and place (or add) in final output */
1571
1572 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1573 int index;
1574 index = vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1575 if (index > 63)
1576 continue;
1577 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1578 block[0] += 128<<3;
1579 s->dsp.idct_put(
1580 output_plane + first_pixel,
1581 stride,
1582 block);
1583 } else {
1584 int index = vp3_dequant(s, s->all_fragments + i, plane, 1, block);
1585 if (index > 63)
1586 continue;
1587 if (index > 0) {
1588 s->dsp.idct_add(
1589 output_plane + first_pixel,
1590 stride,
1591 block);
1592 } else {
1593 s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1594 }
1595 }
1596 } else {
1597
1598 /* copy directly from the previous frame */
1599 s->dsp.put_pixels_tab[1][0](
1600 output_plane + first_pixel,
1601 last_plane + first_pixel,
1602 stride, 8);
1603
1604 }
1605 }
1606 }
1607
1608 // Filter up to the last row in the superblock row
1609 if (!s->skip_loop_filter)
1610 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1611 }
1612 }
1613
1614 /* this looks like a good place for slice dispatch... */
1615 /* algorithm:
1616 * if (slice == s->macroblock_height - 1)
1617 * dispatch (both last slice & 2nd-to-last slice);
1618 * else if (slice > 0)
1619 * dispatch (slice - 1);
1620 */
1621
1622 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1623 }
1624
1625 /// Allocate tables for per-frame data in Vp3DecodeContext
1626 static av_cold int allocate_tables(AVCodecContext *avctx)
1627 {
1628 Vp3DecodeContext *s = avctx->priv_data;
1629 int y_fragment_count, c_fragment_count;
1630
1631 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1632 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1633
1634 s->superblock_coding = av_malloc(s->superblock_count);
1635 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1636 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1637 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1638 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1639 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1640
1641 /* work out the block mapping tables */
1642 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1643 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1644
1645 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1646 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1647 !s->motion_val[0] || !s->motion_val[1]) {
1648 vp3_decode_end(avctx);
1649 return -1;
1650 }
1651
1652 init_block_mapping(s);
1653
1654 return 0;
1655 }
1656
1657 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1658 {
1659 Vp3DecodeContext *s = avctx->priv_data;
1660 int i, inter, plane;
1661 int c_width;
1662 int c_height;
1663 int y_fragment_count, c_fragment_count;
1664
1665 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1666 s->version = 0;
1667 else
1668 s->version = 1;
1669
1670 s->avctx = avctx;
1671 s->width = FFALIGN(avctx->width, 16);
1672 s->height = FFALIGN(avctx->height, 16);
1673 if (avctx->pix_fmt == PIX_FMT_NONE)
1674 avctx->pix_fmt = PIX_FMT_YUV420P;
1675 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1676 if(avctx->idct_algo==FF_IDCT_AUTO)
1677 avctx->idct_algo=FF_IDCT_VP3;
1678 dsputil_init(&s->dsp, avctx);
1679
1680 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1681
1682 /* initialize to an impossible value which will force a recalculation
1683 * in the first frame decode */
1684 for (i = 0; i < 3; i++)
1685 s->qps[i] = -1;
1686
1687 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1688
1689 s->y_superblock_width = (s->width + 31) / 32;
1690 s->y_superblock_height = (s->height + 31) / 32;
1691 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1692
1693 /* work out the dimensions for the C planes */
1694 c_width = s->width >> s->chroma_x_shift;
1695 c_height = s->height >> s->chroma_y_shift;
1696 s->c_superblock_width = (c_width + 31) / 32;
1697 s->c_superblock_height = (c_height + 31) / 32;
1698 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1699
1700 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1701 s->u_superblock_start = s->y_superblock_count;
1702 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1703
1704 s->macroblock_width = (s->width + 15) / 16;
1705 s->macroblock_height = (s->height + 15) / 16;
1706 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1707
1708 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1709 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1710 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1711 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1712
1713 /* fragment count covers all 8x8 blocks for all 3 planes */
1714 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1715 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1716 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1717 s->fragment_start[1] = y_fragment_count;
1718 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1719
1720 if (!s->theora_tables)
1721 {
1722 for (i = 0; i < 64; i++) {
1723 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1724 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1725 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1726 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1727 s->base_matrix[2][i] = vp31_inter_dequant[i];
1728 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1729 }
1730
1731 for(inter=0; inter<2; inter++){
1732 for(plane=0; plane<3; plane++){
1733 s->qr_count[inter][plane]= 1;
1734 s->qr_size [inter][plane][0]= 63;
1735 s->qr_base [inter][plane][0]=
1736 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1737 }
1738 }
1739
1740 /* init VLC tables */
1741 for (i = 0; i < 16; i++) {
1742
1743 /* DC histograms */
1744 init_vlc(&s->dc_vlc[i], 11, 32,
1745 &dc_bias[i][0][1], 4, 2,
1746 &dc_bias[i][0][0], 4, 2, 0);
1747
1748 /* group 1 AC histograms */
1749 init_vlc(&s->ac_vlc_1[i], 11, 32,
1750 &ac_bias_0[i][0][1], 4, 2,
1751 &ac_bias_0[i][0][0], 4, 2, 0);
1752
1753 /* group 2 AC histograms */
1754 init_vlc(&s->ac_vlc_2[i], 11, 32,
1755 &ac_bias_1[i][0][1], 4, 2,
1756 &ac_bias_1[i][0][0], 4, 2, 0);
1757
1758 /* group 3 AC histograms */
1759 init_vlc(&s->ac_vlc_3[i], 11, 32,
1760 &ac_bias_2[i][0][1], 4, 2,
1761 &ac_bias_2[i][0][0], 4, 2, 0);
1762
1763 /* group 4 AC histograms */
1764 init_vlc(&s->ac_vlc_4[i], 11, 32,
1765 &ac_bias_3[i][0][1], 4, 2,
1766 &ac_bias_3[i][0][0], 4, 2, 0);
1767 }
1768 } else {
1769
1770 for (i = 0; i < 16; i++) {
1771 /* DC histograms */
1772 if (init_vlc(&s->dc_vlc[i], 11, 32,
1773 &s->huffman_table[i][0][1], 8, 4,
1774 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1775 goto vlc_fail;
1776
1777 /* group 1 AC histograms */
1778 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1779 &s->huffman_table[i+16][0][1], 8, 4,
1780 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1781 goto vlc_fail;
1782
1783 /* group 2 AC histograms */
1784 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1785 &s->huffman_table[i+16*2][0][1], 8, 4,
1786 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1787 goto vlc_fail;
1788
1789 /* group 3 AC histograms */
1790 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1791 &s->huffman_table[i+16*3][0][1], 8, 4,
1792 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1793 goto vlc_fail;
1794
1795 /* group 4 AC histograms */
1796 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1797 &s->huffman_table[i+16*4][0][1], 8, 4,
1798 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1799 goto vlc_fail;
1800 }
1801 }
1802
1803 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1804 &superblock_run_length_vlc_table[0][1], 4, 2,
1805 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1806
1807 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1808 &fragment_run_length_vlc_table[0][1], 4, 2,
1809 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1810
1811 init_vlc(&s->mode_code_vlc, 3, 8,
1812 &mode_code_vlc_table[0][1], 2, 1,
1813 &mode_code_vlc_table[0][0], 2, 1, 0);
1814
1815 init_vlc(&s->motion_vector_vlc, 6, 63,
1816 &motion_vector_vlc_table[0][1], 2, 1,
1817 &motion_vector_vlc_table[0][0], 2, 1, 0);
1818
1819 for (i = 0; i < 3; i++) {
1820 s->current_frame.data[i] = NULL;
1821 s->last_frame.data[i] = NULL;
1822 s->golden_frame.data[i] = NULL;
1823 }
1824
1825 return allocate_tables(avctx);
1826
1827 vlc_fail:
1828 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1829 return -1;
1830 }
1831
1832 /// Release and shuffle frames after decode finishes
1833 static void update_frames(AVCodecContext *avctx)
1834 {
1835 Vp3DecodeContext *s = avctx->priv_data;
1836
1837 /* release the last frame, if it is allocated and if it is not the
1838 * golden frame */
1839 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1840 ff_thread_release_buffer(avctx, &s->last_frame);
1841
1842 /* shuffle frames (last = current) */
1843 s->last_frame= s->current_frame;
1844
1845 if (s->keyframe) {
1846 if (s->golden_frame.data[0])
1847 ff_thread_release_buffer(avctx, &s->golden_frame);
1848 s->golden_frame = s->current_frame;
1849 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1850 }
1851
1852 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1853 }
1854
1855 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1856 {
1857 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1858 int qps_changed = 0, i, err;
1859
1860 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1861
1862 if (!s1->current_frame.data[0]
1863 ||s->width != s1->width
1864 ||s->height!= s1->height) {
1865 if (s != s1)
1866 copy_fields(s, s1, golden_frame, current_frame);
1867 return -1;
1868 }
1869
1870 if (s != s1) {
1871 // init tables if the first frame hasn't been decoded
1872 if (!s->current_frame.data[0]) {
1873 int y_fragment_count, c_fragment_count;
1874 s->avctx = dst;
1875 err = allocate_tables(dst);
1876 if (err)
1877 return err;
1878 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1879 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1880 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1881 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1882 }
1883
1884 // copy previous frame data
1885 copy_fields(s, s1, golden_frame, dsp);
1886
1887 // copy qscale data if necessary
1888 for (i = 0; i < 3; i++) {
1889 if (s->qps[i] != s1->qps[1]) {
1890 qps_changed = 1;
1891 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1892 }
1893 }
1894
1895 if (s->qps[0] != s1->qps[0])
1896 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1897
1898 if (qps_changed)
1899 copy_fields(s, s1, qps, superblock_count);
1900 #undef copy_fields
1901 }
1902
1903 update_frames(dst);
1904
1905 return 0;
1906 }
1907
1908 static int vp3_decode_frame(AVCodecContext *avctx,
1909 void *data, int *data_size,
1910 AVPacket *avpkt)
1911 {
1912 const uint8_t *buf = avpkt->data;
1913 int buf_size = avpkt->size;
1914 Vp3DecodeContext *s = avctx->priv_data;
1915 GetBitContext gb;
1916 int i;
1917
1918 init_get_bits(&gb, buf, buf_size * 8);
1919
1920 if (s->theora && get_bits1(&gb))
1921 {
1922 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1923 return -1;
1924 }
1925
1926 s->keyframe = !get_bits1(&gb);
1927 if (!s->theora)
1928 skip_bits(&gb, 1);
1929 for (i = 0; i < 3; i++)
1930 s->last_qps[i] = s->qps[i];
1931
1932 s->nqps=0;
1933 do{
1934 s->qps[s->nqps++]= get_bits(&gb, 6);
1935 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1936 for (i = s->nqps; i < 3; i++)
1937 s->qps[i] = -1;
1938
1939 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1940 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1941 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1942
1943 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1944 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1945
1946 if (s->qps[0] != s->last_qps[0])
1947 init_loop_filter(s);
1948
1949 for (i = 0; i < s->nqps; i++)
1950 // reinit all dequantizers if the first one changed, because
1951 // the DC of the first quantizer must be used for all matrices
1952 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1953 init_dequantizer(s, i);
1954
1955 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1956 return buf_size;
1957
1958 s->current_frame.reference = 3;
1959 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1960 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1961 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1962 goto error;
1963 }
1964
1965 if (!s->edge_emu_buffer)
1966 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1967
1968 if (s->keyframe) {
1969 if (!s->theora)
1970 {
1971 skip_bits(&gb, 4); /* width code */
1972 skip_bits(&gb, 4); /* height code */
1973 if (s->version)
1974 {
1975 s->version = get_bits(&gb, 5);
1976 if (avctx->frame_number == 0)
1977 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1978 }
1979 }
1980 if (s->version || s->theora)
1981 {
1982 if (get_bits1(&gb))
1983 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1984 skip_bits(&gb, 2); /* reserved? */
1985 }
1986 } else {
1987 if (!s->golden_frame.data[0]) {
1988 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1989
1990 s->golden_frame.reference = 3;
1991 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1992 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1993 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1994 goto error;
1995 }
1996 s->last_frame = s->golden_frame;
1997 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1998 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
1999 }
2000 }
2001
2002 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2003 ff_thread_finish_setup(avctx);
2004
2005 if (unpack_superblocks(s, &gb)){
2006 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2007 goto error;
2008 }
2009 if (unpack_modes(s, &gb)){
2010 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2011 goto error;
2012 }
2013 if (unpack_vectors(s, &gb)){
2014 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2015 goto error;
2016 }
2017 if (unpack_block_qpis(s, &gb)){
2018 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2019 goto error;
2020 }
2021 if (unpack_dct_coeffs(s, &gb)){
2022 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2023 goto error;
2024 }
2025
2026 for (i = 0; i < 3; i++) {
2027 int height = s->height >> (i && s->chroma_y_shift);
2028 if (s->flipped_image)
2029 s->data_offset[i] = 0;
2030 else
2031 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
2032 }
2033
2034 s->last_slice_end = 0;
2035 for (i = 0; i < s->c_superblock_height; i++)
2036 render_slice(s, i);
2037
2038 // filter the last row
2039 for (i = 0; i < 3; i++) {
2040 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
2041 apply_loop_filter(s, i, row, row+1);
2042 }
2043 vp3_draw_horiz_band(s, s->avctx->height);
2044
2045 *data_size=sizeof(AVFrame);
2046 *(AVFrame*)data= s->current_frame;
2047
2048 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2049 update_frames(avctx);
2050
2051 return buf_size;
2052
2053 error:
2054 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2055
2056 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2057 avctx->release_buffer(avctx, &s->current_frame);
2058
2059 return -1;
2060 }
2061
2062 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2063 {
2064 Vp3DecodeContext *s = avctx->priv_data;
2065
2066 if (get_bits1(gb)) {
2067 int token;
2068 if (s->entries >= 32) { /* overflow */
2069 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2070 return -1;
2071 }
2072 token = get_bits(gb, 5);
2073 //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);
2074 s->huffman_table[s->hti][token][0] = s->hbits;
2075 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2076 s->entries++;
2077 }
2078 else {
2079 if (s->huff_code_size >= 32) {/* overflow */
2080 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2081 return -1;
2082 }
2083 s->huff_code_size++;
2084 s->hbits <<= 1;
2085 if (read_huffman_tree(avctx, gb))
2086 return -1;
2087 s->hbits |= 1;
2088 if (read_huffman_tree(avctx, gb))
2089 return -1;
2090 s->hbits >>= 1;
2091 s->huff_code_size--;
2092 }
2093 return 0;
2094 }
2095
2096 static int vp3_init_thread_copy(AVCodecContext *avctx)
2097 {
2098 Vp3DecodeContext *s = avctx->priv_data;
2099
2100 s->superblock_coding = NULL;
2101 s->all_fragments = NULL;
2102 s->coded_fragment_list[0] = NULL;
2103 s->dct_tokens_base = NULL;
2104 s->superblock_fragments = NULL;
2105 s->macroblock_coding = NULL;
2106 s->motion_val[0] = NULL;
2107 s->motion_val[1] = NULL;
2108 s->edge_emu_buffer = NULL;
2109
2110 return 0;
2111 }
2112
2113 #if CONFIG_THEORA_DECODER
2114 static const enum PixelFormat theora_pix_fmts[4] = {
2115 PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
2116 };
2117
2118 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2119 {
2120 Vp3DecodeContext *s = avctx->priv_data;
2121 int visible_width, visible_height, colorspace;
2122 int offset_x = 0, offset_y = 0;
2123 AVRational fps, aspect;
2124
2125 s->theora = get_bits_long(gb, 24);
2126 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2127
2128 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2129 /* but previous versions have the image flipped relative to vp3 */
2130 if (s->theora < 0x030200)
2131 {
2132 s->flipped_image = 1;
2133 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2134 }
2135
2136 visible_width = s->width = get_bits(gb, 16) << 4;
2137 visible_height = s->height = get_bits(gb, 16) << 4;
2138
2139 if(av_image_check_size(s->width, s->height, 0, avctx)){
2140 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2141 s->width= s->height= 0;
2142 return -1;
2143 }
2144
2145 if (s->theora >= 0x030200) {
2146 visible_width = get_bits_long(gb, 24);
2147 visible_height = get_bits_long(gb, 24);
2148
2149 offset_x = get_bits(gb, 8); /* offset x */
2150 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2151 }
2152
2153 fps.num = get_bits_long(gb, 32);
2154 fps.den = get_bits_long(gb, 32);
2155 if (fps.num && fps.den) {
2156 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2157 fps.den, fps.num, 1<<30);
2158 }
2159
2160 aspect.num = get_bits_long(gb, 24);
2161 aspect.den = get_bits_long(gb, 24);
2162 if (aspect.num && aspect.den) {
2163 av_reduce(&avctx->sample_aspect_ratio.num,
2164 &avctx->sample_aspect_ratio.den,
2165 aspect.num, aspect.den, 1<<30);
2166 }
2167
2168 if (s->theora < 0x030200)
2169 skip_bits(gb, 5); /* keyframe frequency force */
2170 colorspace = get_bits(gb, 8);
2171 skip_bits(gb, 24); /* bitrate */
2172
2173 skip_bits(gb, 6); /* quality hint */
2174
2175 if (s->theora >= 0x030200)
2176 {
2177 skip_bits(gb, 5); /* keyframe frequency force */
2178 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2179 skip_bits(gb, 3); /* reserved */
2180 }
2181
2182 // align_get_bits(gb);
2183
2184 if ( visible_width <= s->width && visible_width > s->width-16
2185 && visible_height <= s->height && visible_height > s->height-16
2186 && !offset_x && (offset_y == s->height - visible_height))
2187 avcodec_set_dimensions(avctx, visible_width, visible_height);
2188 else
2189 avcodec_set_dimensions(avctx, s->width, s->height);
2190
2191 if (colorspace == 1) {
2192 avctx->color_primaries = AVCOL_PRI_BT470M;
2193 } else if (colorspace == 2) {
2194 avctx->color_primaries = AVCOL_PRI_BT470BG;
2195 }
2196 if (colorspace == 1 || colorspace == 2) {
2197 avctx->colorspace = AVCOL_SPC_BT470BG;
2198 avctx->color_trc = AVCOL_TRC_BT709;
2199 }
2200
2201 return 0;
2202 }
2203
2204 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2205 {
2206 Vp3DecodeContext *s = avctx->priv_data;
2207 int i, n, matrices, inter, plane;
2208
2209 if (s->theora >= 0x030200) {
2210 n = get_bits(gb, 3);
2211 /* loop filter limit values table */
2212 if (n)
2213 for (i = 0; i < 64; i++)
2214 s->filter_limit_values[i] = get_bits(gb, n);
2215 }
2216
2217 if (s->theora >= 0x030200)
2218 n = get_bits(gb, 4) + 1;
2219 else
2220 n = 16;
2221 /* quality threshold table */
2222 for (i = 0; i < 64; i++)
2223 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2224
2225 if (s->theora >= 0x030200)
2226 n = get_bits(gb, 4) + 1;
2227 else
2228 n = 16;
2229 /* dc scale factor table */
2230 for (i = 0; i < 64; i++)
2231 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2232
2233 if (s->theora >= 0x030200)
2234 matrices = get_bits(gb, 9) + 1;
2235 else
2236 matrices = 3;
2237
2238 if(matrices > 384){
2239 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2240 return -1;
2241 }
2242
2243 for(n=0; n<matrices; n++){
2244 for (i = 0; i < 64; i++)
2245 s->base_matrix[n][i]= get_bits(gb, 8);
2246 }
2247
2248 for (inter = 0; inter <= 1; inter++) {
2249 for (plane = 0; plane <= 2; plane++) {
2250 int newqr= 1;
2251 if (inter || plane > 0)
2252 newqr = get_bits1(gb);
2253 if (!newqr) {
2254 int qtj, plj;
2255 if(inter && get_bits1(gb)){
2256 qtj = 0;
2257 plj = plane;
2258 }else{
2259 qtj= (3*inter + plane - 1) / 3;
2260 plj= (plane + 2) % 3;
2261 }
2262 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2263 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2264 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2265 } else {
2266 int qri= 0;
2267 int qi = 0;
2268
2269 for(;;){
2270 i= get_bits(gb, av_log2(matrices-1)+1);
2271 if(i>= matrices){
2272 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2273 return -1;
2274 }
2275 s->qr_base[inter][plane][qri]= i;
2276 if(qi >= 63)
2277 break;
2278 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2279 s->qr_size[inter][plane][qri++]= i;
2280 qi += i;
2281 }
2282
2283 if (qi > 63) {
2284 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2285 return -1;
2286 }
2287 s->qr_count[inter][plane]= qri;
2288 }
2289 }
2290 }
2291
2292 /* Huffman tables */
2293 for (s->hti = 0; s->hti < 80; s->hti++) {
2294 s->entries = 0;
2295 s->huff_code_size = 1;
2296 if (!get_bits1(gb)) {
2297 s->hbits = 0;
2298 if(read_huffman_tree(avctx, gb))
2299 return -1;
2300 s->hbits = 1;
2301 if(read_huffman_tree(avctx, gb))
2302 return -1;
2303 }
2304 }
2305
2306 s->theora_tables = 1;
2307
2308 return 0;
2309 }
2310
2311 static av_cold int theora_decode_init(AVCodecContext *avctx)
2312 {
2313 Vp3DecodeContext *s = avctx->priv_data;
2314 GetBitContext gb;
2315 int ptype;
2316 uint8_t *header_start[3];
2317 int header_len[3];
2318 int i;
2319
2320 s->theora = 1;
2321
2322 if (!avctx->extradata_size)
2323 {
2324 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2325 return -1;
2326 }
2327
2328 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2329 42, header_start, header_len) < 0) {
2330 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2331 return -1;
2332 }
2333
2334 for(i=0;i<3;i++) {
2335 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2336
2337 ptype = get_bits(&gb, 8);
2338
2339 if (!(ptype & 0x80))
2340 {
2341 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2342 // return -1;
2343 }
2344
2345 // FIXME: Check for this as well.
2346 skip_bits_long(&gb, 6*8); /* "theora" */
2347
2348 switch(ptype)
2349 {
2350 case 0x80:
2351 theora_decode_header(avctx, &gb);
2352 break;
2353 case 0x81:
2354 // FIXME: is this needed? it breaks sometimes
2355 // theora_decode_comments(avctx, gb);
2356 break;
2357 case 0x82:
2358 if (theora_decode_tables(avctx, &gb))
2359 return -1;
2360 break;
2361 default:
2362 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2363 break;
2364 }
2365 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2366 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2367 if (s->theora < 0x030200)
2368 break;
2369 }
2370
2371 return vp3_decode_init(avctx);
2372 }
2373
2374 AVCodec ff_theora_decoder = {
2375 .name = "theora",
2376 .type = AVMEDIA_TYPE_VIDEO,
2377 .id = CODEC_ID_THEORA,
2378 .priv_data_size = sizeof(Vp3DecodeContext),
2379 .init = theora_decode_init,
2380 .close = vp3_decode_end,
2381 .decode = vp3_decode_frame,
2382 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2383 .flush = vp3_decode_flush,
2384 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2385 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2386 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2387 };
2388 #endif
2389
2390 AVCodec ff_vp3_decoder = {
2391 .name = "vp3",
2392 .type = AVMEDIA_TYPE_VIDEO,
2393 .id = CODEC_ID_VP3,
2394 .priv_data_size = sizeof(Vp3DecodeContext),
2395 .init = vp3_decode_init,
2396 .close = vp3_decode_end,
2397 .decode = vp3_decode_frame,
2398 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2399 .flush = vp3_decode_flush,
2400 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2401 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2402 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2403 };