ce14e63735ce85c28f6487516da16677484e0af8
[libav.git] / libavcodec / vp3.c
1 /*
2 * Copyright (C) 2003-2004 the ffmpeg project
3 *
4 * This file is part of Libav.
5 *
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 /**
22 * @file
23 * On2 VP3 Video Decoder
24 *
25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26 * For more information about the VP3 coding process, visit:
27 * http://wiki.multimedia.cx/index.php?title=On2_VP3
28 *
29 * Theora decoder by Alex Beregszaszi
30 */
31
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <string.h>
35
36 #include "libavutil/imgutils.h"
37 #include "avcodec.h"
38 #include "dsputil.h"
39 #include "get_bits.h"
40
41 #include "vp3data.h"
42 #include "xiph.h"
43 #include "thread.h"
44
45 #define FRAGMENT_PIXELS 8
46
47 static av_cold int vp3_decode_end(AVCodecContext *avctx);
48 static void vp3_decode_flush(AVCodecContext *avctx);
49
50 //FIXME split things out into their own arrays
51 typedef struct Vp3Fragment {
52 int16_t dc;
53 uint8_t coding_method;
54 uint8_t qpi;
55 } Vp3Fragment;
56
57 #define SB_NOT_CODED 0
58 #define SB_PARTIALLY_CODED 1
59 #define SB_FULLY_CODED 2
60
61 // This is the maximum length of a single long bit run that can be encoded
62 // for superblock coding or block qps. Theora special-cases this to read a
63 // bit instead of flipping the current bit to allow for runs longer than 4129.
64 #define MAXIMUM_LONG_BIT_RUN 4129
65
66 #define MODE_INTER_NO_MV 0
67 #define MODE_INTRA 1
68 #define MODE_INTER_PLUS_MV 2
69 #define MODE_INTER_LAST_MV 3
70 #define MODE_INTER_PRIOR_LAST 4
71 #define MODE_USING_GOLDEN 5
72 #define MODE_GOLDEN_MV 6
73 #define MODE_INTER_FOURMV 7
74 #define CODING_MODE_COUNT 8
75
76 /* special internal mode */
77 #define MODE_COPY 8
78
79 /* There are 6 preset schemes, plus a free-form scheme */
80 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
81 {
82 /* scheme 1: Last motion vector dominates */
83 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
84 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
85 MODE_INTRA, MODE_USING_GOLDEN,
86 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
87
88 /* scheme 2 */
89 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
90 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
91 MODE_INTRA, MODE_USING_GOLDEN,
92 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
93
94 /* scheme 3 */
95 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
96 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
97 MODE_INTRA, MODE_USING_GOLDEN,
98 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
99
100 /* scheme 4 */
101 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
102 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
103 MODE_INTRA, MODE_USING_GOLDEN,
104 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
105
106 /* scheme 5: No motion vector dominates */
107 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
108 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
109 MODE_INTRA, MODE_USING_GOLDEN,
110 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
111
112 /* scheme 6 */
113 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
114 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
115 MODE_INTER_PLUS_MV, MODE_INTRA,
116 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
117
118 };
119
120 static const uint8_t hilbert_offset[16][2] = {
121 {0,0}, {1,0}, {1,1}, {0,1},
122 {0,2}, {0,3}, {1,3}, {1,2},
123 {2,2}, {2,3}, {3,3}, {3,2},
124 {3,1}, {2,1}, {2,0}, {3,0}
125 };
126
127 #define MIN_DEQUANT_VAL 2
128
129 typedef struct Vp3DecodeContext {
130 AVCodecContext *avctx;
131 int theora, theora_tables;
132 int version;
133 int width, height;
134 int chroma_x_shift, chroma_y_shift;
135 AVFrame golden_frame;
136 AVFrame last_frame;
137 AVFrame current_frame;
138 int keyframe;
139 DSPContext dsp;
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 /*
259 * This function sets up all of the various blocks mappings:
260 * superblocks <-> fragments, macroblocks <-> fragments,
261 * superblocks <-> macroblocks
262 *
263 * @return 0 is successful; returns 1 if *anything* went wrong.
264 */
265 static int init_block_mapping(Vp3DecodeContext *s)
266 {
267 int sb_x, sb_y, plane;
268 int x, y, i, j = 0;
269
270 for (plane = 0; plane < 3; plane++) {
271 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
272 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
273 int frag_width = s->fragment_width[!!plane];
274 int frag_height = s->fragment_height[!!plane];
275
276 for (sb_y = 0; sb_y < sb_height; sb_y++)
277 for (sb_x = 0; sb_x < sb_width; sb_x++)
278 for (i = 0; i < 16; i++) {
279 x = 4*sb_x + hilbert_offset[i][0];
280 y = 4*sb_y + hilbert_offset[i][1];
281
282 if (x < frag_width && y < frag_height)
283 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
284 else
285 s->superblock_fragments[j++] = -1;
286 }
287 }
288
289 return 0; /* successful path out */
290 }
291
292 /*
293 * This function sets up the dequantization tables used for a particular
294 * frame.
295 */
296 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
297 {
298 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
299 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
300 int i, plane, inter, qri, bmi, bmj, qistart;
301
302 for(inter=0; inter<2; inter++){
303 for(plane=0; plane<3; plane++){
304 int sum=0;
305 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
306 sum+= s->qr_size[inter][plane][qri];
307 if(s->qps[qpi] <= sum)
308 break;
309 }
310 qistart= sum - s->qr_size[inter][plane][qri];
311 bmi= s->qr_base[inter][plane][qri ];
312 bmj= s->qr_base[inter][plane][qri+1];
313 for(i=0; i<64; i++){
314 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
315 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
316 + s->qr_size[inter][plane][qri])
317 / (2*s->qr_size[inter][plane][qri]);
318
319 int qmin= 8<<(inter + !i);
320 int qscale= i ? ac_scale_factor : dc_scale_factor;
321
322 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
323 }
324 // all DC coefficients use the same quant so as not to interfere with DC prediction
325 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
326 }
327 }
328 }
329
330 /*
331 * This function initializes the loop filter boundary limits if the frame's
332 * quality index is different from the previous frame's.
333 *
334 * The filter_limit_values may not be larger than 127.
335 */
336 static void init_loop_filter(Vp3DecodeContext *s)
337 {
338 int *bounding_values= s->bounding_values_array+127;
339 int filter_limit;
340 int x;
341 int value;
342
343 filter_limit = s->filter_limit_values[s->qps[0]];
344
345 /* set up the bounding values */
346 memset(s->bounding_values_array, 0, 256 * sizeof(int));
347 for (x = 0; x < filter_limit; x++) {
348 bounding_values[-x] = -x;
349 bounding_values[x] = x;
350 }
351 for (x = value = filter_limit; x < 128 && value; x++, value--) {
352 bounding_values[ x] = value;
353 bounding_values[-x] = -value;
354 }
355 if (value)
356 bounding_values[128] = value;
357 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
358 }
359
360 /*
361 * This function unpacks all of the superblock/macroblock/fragment coding
362 * information from the bitstream.
363 */
364 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
365 {
366 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
367 int bit = 0;
368 int current_superblock = 0;
369 int current_run = 0;
370 int num_partial_superblocks = 0;
371
372 int i, j;
373 int current_fragment;
374 int plane;
375
376 if (s->keyframe) {
377 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
378
379 } else {
380
381 /* unpack the list of partially-coded superblocks */
382 bit = get_bits1(gb) ^ 1;
383 current_run = 0;
384
385 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
386 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
387 bit = get_bits1(gb);
388 else
389 bit ^= 1;
390
391 current_run = get_vlc2(gb,
392 s->superblock_run_length_vlc.table, 6, 2) + 1;
393 if (current_run == 34)
394 current_run += get_bits(gb, 12);
395
396 if (current_superblock + current_run > s->superblock_count) {
397 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
398 return -1;
399 }
400
401 memset(s->superblock_coding + current_superblock, bit, current_run);
402
403 current_superblock += current_run;
404 if (bit)
405 num_partial_superblocks += current_run;
406 }
407
408 /* unpack the list of fully coded superblocks if any of the blocks were
409 * not marked as partially coded in the previous step */
410 if (num_partial_superblocks < s->superblock_count) {
411 int superblocks_decoded = 0;
412
413 current_superblock = 0;
414 bit = get_bits1(gb) ^ 1;
415 current_run = 0;
416
417 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
418 && get_bits_left(gb) > 0) {
419
420 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
421 bit = get_bits1(gb);
422 else
423 bit ^= 1;
424
425 current_run = get_vlc2(gb,
426 s->superblock_run_length_vlc.table, 6, 2) + 1;
427 if (current_run == 34)
428 current_run += get_bits(gb, 12);
429
430 for (j = 0; j < current_run; current_superblock++) {
431 if (current_superblock >= s->superblock_count) {
432 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
433 return -1;
434 }
435
436 /* skip any superblocks already marked as partially coded */
437 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
438 s->superblock_coding[current_superblock] = 2*bit;
439 j++;
440 }
441 }
442 superblocks_decoded += current_run;
443 }
444 }
445
446 /* if there were partial blocks, initialize bitstream for
447 * unpacking fragment codings */
448 if (num_partial_superblocks) {
449
450 current_run = 0;
451 bit = get_bits1(gb);
452 /* toggle the bit because as soon as the first run length is
453 * fetched the bit will be toggled again */
454 bit ^= 1;
455 }
456 }
457
458 /* figure out which fragments are coded; iterate through each
459 * superblock (all planes) */
460 s->total_num_coded_frags = 0;
461 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
462
463 for (plane = 0; plane < 3; plane++) {
464 int sb_start = superblock_starts[plane];
465 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
466 int num_coded_frags = 0;
467
468 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
469
470 /* iterate through all 16 fragments in a superblock */
471 for (j = 0; j < 16; j++) {
472
473 /* if the fragment is in bounds, check its coding status */
474 current_fragment = s->superblock_fragments[i * 16 + j];
475 if (current_fragment != -1) {
476 int coded = s->superblock_coding[i];
477
478 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
479
480 /* fragment may or may not be coded; this is the case
481 * that cares about the fragment coding runs */
482 if (current_run-- == 0) {
483 bit ^= 1;
484 current_run = get_vlc2(gb,
485 s->fragment_run_length_vlc.table, 5, 2);
486 }
487 coded = bit;
488 }
489
490 if (coded) {
491 /* default mode; actual mode will be decoded in
492 * the next phase */
493 s->all_fragments[current_fragment].coding_method =
494 MODE_INTER_NO_MV;
495 s->coded_fragment_list[plane][num_coded_frags++] =
496 current_fragment;
497 } else {
498 /* not coded; copy this fragment from the prior frame */
499 s->all_fragments[current_fragment].coding_method =
500 MODE_COPY;
501 }
502 }
503 }
504 }
505 s->total_num_coded_frags += num_coded_frags;
506 for (i = 0; i < 64; i++)
507 s->num_coded_frags[plane][i] = num_coded_frags;
508 if (plane < 2)
509 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
510 }
511 return 0;
512 }
513
514 /*
515 * This function unpacks all the coding mode data for individual macroblocks
516 * from the bitstream.
517 */
518 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
519 {
520 int i, j, k, sb_x, sb_y;
521 int scheme;
522 int current_macroblock;
523 int current_fragment;
524 int coding_mode;
525 int custom_mode_alphabet[CODING_MODE_COUNT];
526 const int *alphabet;
527 Vp3Fragment *frag;
528
529 if (s->keyframe) {
530 for (i = 0; i < s->fragment_count; i++)
531 s->all_fragments[i].coding_method = MODE_INTRA;
532
533 } else {
534
535 /* fetch the mode coding scheme for this frame */
536 scheme = get_bits(gb, 3);
537
538 /* is it a custom coding scheme? */
539 if (scheme == 0) {
540 for (i = 0; i < 8; i++)
541 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
542 for (i = 0; i < 8; i++)
543 custom_mode_alphabet[get_bits(gb, 3)] = i;
544 alphabet = custom_mode_alphabet;
545 } else
546 alphabet = ModeAlphabet[scheme-1];
547
548 /* iterate through all of the macroblocks that contain 1 or more
549 * coded fragments */
550 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
551 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
552 if (get_bits_left(gb) <= 0)
553 return -1;
554
555 for (j = 0; j < 4; j++) {
556 int mb_x = 2*sb_x + (j>>1);
557 int mb_y = 2*sb_y + (((j>>1)+j)&1);
558 current_macroblock = mb_y * s->macroblock_width + mb_x;
559
560 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
561 continue;
562
563 #define BLOCK_X (2*mb_x + (k&1))
564 #define BLOCK_Y (2*mb_y + (k>>1))
565 /* coding modes are only stored if the macroblock has at least one
566 * luma block coded, otherwise it must be INTER_NO_MV */
567 for (k = 0; k < 4; k++) {
568 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
569 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
570 break;
571 }
572 if (k == 4) {
573 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
574 continue;
575 }
576
577 /* mode 7 means get 3 bits for each coding mode */
578 if (scheme == 7)
579 coding_mode = get_bits(gb, 3);
580 else
581 coding_mode = alphabet
582 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
583
584 s->macroblock_coding[current_macroblock] = coding_mode;
585 for (k = 0; k < 4; k++) {
586 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
587 if (frag->coding_method != MODE_COPY)
588 frag->coding_method = coding_mode;
589 }
590
591 #define SET_CHROMA_MODES \
592 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
593 frag[s->fragment_start[1]].coding_method = coding_mode;\
594 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
595 frag[s->fragment_start[2]].coding_method = coding_mode;
596
597 if (s->chroma_y_shift) {
598 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
599 SET_CHROMA_MODES
600 } else if (s->chroma_x_shift) {
601 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
602 for (k = 0; k < 2; k++) {
603 SET_CHROMA_MODES
604 frag += s->fragment_width[1];
605 }
606 } else {
607 for (k = 0; k < 4; k++) {
608 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
609 SET_CHROMA_MODES
610 }
611 }
612 }
613 }
614 }
615 }
616
617 return 0;
618 }
619
620 /*
621 * This function unpacks all the motion vectors for the individual
622 * macroblocks from the bitstream.
623 */
624 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
625 {
626 int j, k, sb_x, sb_y;
627 int coding_mode;
628 int motion_x[4];
629 int motion_y[4];
630 int last_motion_x = 0;
631 int last_motion_y = 0;
632 int prior_last_motion_x = 0;
633 int prior_last_motion_y = 0;
634 int current_macroblock;
635 int current_fragment;
636 int frag;
637
638 if (s->keyframe)
639 return 0;
640
641 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
642 coding_mode = get_bits1(gb);
643
644 /* iterate through all of the macroblocks that contain 1 or more
645 * coded fragments */
646 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
647 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
648 if (get_bits_left(gb) <= 0)
649 return -1;
650
651 for (j = 0; j < 4; j++) {
652 int mb_x = 2*sb_x + (j>>1);
653 int mb_y = 2*sb_y + (((j>>1)+j)&1);
654 current_macroblock = mb_y * s->macroblock_width + mb_x;
655
656 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
657 (s->macroblock_coding[current_macroblock] == MODE_COPY))
658 continue;
659
660 switch (s->macroblock_coding[current_macroblock]) {
661
662 case MODE_INTER_PLUS_MV:
663 case MODE_GOLDEN_MV:
664 /* all 6 fragments use the same motion vector */
665 if (coding_mode == 0) {
666 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
667 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
668 } else {
669 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
670 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
671 }
672
673 /* vector maintenance, only on MODE_INTER_PLUS_MV */
674 if (s->macroblock_coding[current_macroblock] ==
675 MODE_INTER_PLUS_MV) {
676 prior_last_motion_x = last_motion_x;
677 prior_last_motion_y = last_motion_y;
678 last_motion_x = motion_x[0];
679 last_motion_y = motion_y[0];
680 }
681 break;
682
683 case MODE_INTER_FOURMV:
684 /* vector maintenance */
685 prior_last_motion_x = last_motion_x;
686 prior_last_motion_y = last_motion_y;
687
688 /* fetch 4 vectors from the bitstream, one for each
689 * Y fragment, then average for the C fragment vectors */
690 for (k = 0; k < 4; k++) {
691 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
692 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
693 if (coding_mode == 0) {
694 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
695 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
696 } else {
697 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
698 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
699 }
700 last_motion_x = motion_x[k];
701 last_motion_y = motion_y[k];
702 } else {
703 motion_x[k] = 0;
704 motion_y[k] = 0;
705 }
706 }
707 break;
708
709 case MODE_INTER_LAST_MV:
710 /* all 6 fragments use the last motion vector */
711 motion_x[0] = last_motion_x;
712 motion_y[0] = last_motion_y;
713
714 /* no vector maintenance (last vector remains the
715 * last vector) */
716 break;
717
718 case MODE_INTER_PRIOR_LAST:
719 /* all 6 fragments use the motion vector prior to the
720 * last motion vector */
721 motion_x[0] = prior_last_motion_x;
722 motion_y[0] = prior_last_motion_y;
723
724 /* vector maintenance */
725 prior_last_motion_x = last_motion_x;
726 prior_last_motion_y = last_motion_y;
727 last_motion_x = motion_x[0];
728 last_motion_y = motion_y[0];
729 break;
730
731 default:
732 /* covers intra, inter without MV, golden without MV */
733 motion_x[0] = 0;
734 motion_y[0] = 0;
735
736 /* no vector maintenance */
737 break;
738 }
739
740 /* assign the motion vectors to the correct fragments */
741 for (k = 0; k < 4; k++) {
742 current_fragment =
743 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
744 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
745 s->motion_val[0][current_fragment][0] = motion_x[k];
746 s->motion_val[0][current_fragment][1] = motion_y[k];
747 } else {
748 s->motion_val[0][current_fragment][0] = motion_x[0];
749 s->motion_val[0][current_fragment][1] = motion_y[0];
750 }
751 }
752
753 if (s->chroma_y_shift) {
754 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
755 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
756 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
757 }
758 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
759 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
760 frag = mb_y*s->fragment_width[1] + mb_x;
761 s->motion_val[1][frag][0] = motion_x[0];
762 s->motion_val[1][frag][1] = motion_y[0];
763 } else if (s->chroma_x_shift) {
764 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
765 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
766 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
767 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
768 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
769 } else {
770 motion_x[1] = motion_x[0];
771 motion_y[1] = motion_y[0];
772 }
773 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
774 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
775
776 frag = 2*mb_y*s->fragment_width[1] + mb_x;
777 for (k = 0; k < 2; k++) {
778 s->motion_val[1][frag][0] = motion_x[k];
779 s->motion_val[1][frag][1] = motion_y[k];
780 frag += s->fragment_width[1];
781 }
782 } else {
783 for (k = 0; k < 4; k++) {
784 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
785 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
786 s->motion_val[1][frag][0] = motion_x[k];
787 s->motion_val[1][frag][1] = motion_y[k];
788 } else {
789 s->motion_val[1][frag][0] = motion_x[0];
790 s->motion_val[1][frag][1] = motion_y[0];
791 }
792 }
793 }
794 }
795 }
796 }
797
798 return 0;
799 }
800
801 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
802 {
803 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
804 int num_blocks = s->total_num_coded_frags;
805
806 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
807 i = blocks_decoded = num_blocks_at_qpi = 0;
808
809 bit = get_bits1(gb) ^ 1;
810 run_length = 0;
811
812 do {
813 if (run_length == MAXIMUM_LONG_BIT_RUN)
814 bit = get_bits1(gb);
815 else
816 bit ^= 1;
817
818 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
819 if (run_length == 34)
820 run_length += get_bits(gb, 12);
821 blocks_decoded += run_length;
822
823 if (!bit)
824 num_blocks_at_qpi += run_length;
825
826 for (j = 0; j < run_length; i++) {
827 if (i >= s->total_num_coded_frags)
828 return -1;
829
830 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
831 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
832 j++;
833 }
834 }
835 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
836
837 num_blocks -= num_blocks_at_qpi;
838 }
839
840 return 0;
841 }
842
843 /*
844 * This function is called by unpack_dct_coeffs() to extract the VLCs from
845 * the bitstream. The VLCs encode tokens which are used to unpack DCT
846 * data. This function unpacks all the VLCs for either the Y plane or both
847 * C planes, and is called for DC coefficients or different AC coefficient
848 * levels (since different coefficient types require different VLC tables.
849 *
850 * This function returns a residual eob run. E.g, if a particular token gave
851 * instructions to EOB the next 5 fragments and there were only 2 fragments
852 * left in the current fragment range, 3 would be returned so that it could
853 * be passed into the next call to this same function.
854 */
855 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
856 VLC *table, int coeff_index,
857 int plane,
858 int eob_run)
859 {
860 int i, j = 0;
861 int token;
862 int zero_run = 0;
863 DCTELEM coeff = 0;
864 int bits_to_get;
865 int blocks_ended;
866 int coeff_i = 0;
867 int num_coeffs = s->num_coded_frags[plane][coeff_index];
868 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
869
870 /* local references to structure members to avoid repeated deferences */
871 int *coded_fragment_list = s->coded_fragment_list[plane];
872 Vp3Fragment *all_fragments = s->all_fragments;
873 VLC_TYPE (*vlc_table)[2] = table->table;
874
875 if (num_coeffs < 0)
876 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
877
878 if (eob_run > num_coeffs) {
879 coeff_i = blocks_ended = num_coeffs;
880 eob_run -= num_coeffs;
881 } else {
882 coeff_i = blocks_ended = eob_run;
883 eob_run = 0;
884 }
885
886 // insert fake EOB token to cover the split between planes or zzi
887 if (blocks_ended)
888 dct_tokens[j++] = blocks_ended << 2;
889
890 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
891 /* decode a VLC into a token */
892 token = get_vlc2(gb, vlc_table, 11, 3);
893 /* use the token to get a zero run, a coefficient, and an eob run */
894 if ((unsigned) token <= 6U) {
895 eob_run = eob_run_base[token];
896 if (eob_run_get_bits[token])
897 eob_run += get_bits(gb, eob_run_get_bits[token]);
898
899 // record only the number of blocks ended in this plane,
900 // any spill will be recorded in the next plane.
901 if (eob_run > num_coeffs - coeff_i) {
902 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
903 blocks_ended += num_coeffs - coeff_i;
904 eob_run -= num_coeffs - coeff_i;
905 coeff_i = num_coeffs;
906 } else {
907 dct_tokens[j++] = TOKEN_EOB(eob_run);
908 blocks_ended += eob_run;
909 coeff_i += eob_run;
910 eob_run = 0;
911 }
912 } else if (token >= 0) {
913 bits_to_get = coeff_get_bits[token];
914 if (bits_to_get)
915 bits_to_get = get_bits(gb, bits_to_get);
916 coeff = coeff_tables[token][bits_to_get];
917
918 zero_run = zero_run_base[token];
919 if (zero_run_get_bits[token])
920 zero_run += get_bits(gb, zero_run_get_bits[token]);
921
922 if (zero_run) {
923 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
924 } else {
925 // Save DC into the fragment structure. DC prediction is
926 // done in raster order, so the actual DC can't be in with
927 // other tokens. We still need the token in dct_tokens[]
928 // however, or else the structure collapses on itself.
929 if (!coeff_index)
930 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
931
932 dct_tokens[j++] = TOKEN_COEFF(coeff);
933 }
934
935 if (coeff_index + zero_run > 64) {
936 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
937 " %d coeffs left\n", zero_run, 64-coeff_index);
938 zero_run = 64 - coeff_index;
939 }
940
941 // zero runs code multiple coefficients,
942 // so don't try to decode coeffs for those higher levels
943 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
944 s->num_coded_frags[plane][i]--;
945 coeff_i++;
946 } else {
947 av_log(s->avctx, AV_LOG_ERROR,
948 "Invalid token %d\n", token);
949 return -1;
950 }
951 }
952
953 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
954 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
955
956 // decrement the number of blocks that have higher coeffecients for each
957 // EOB run at this level
958 if (blocks_ended)
959 for (i = coeff_index+1; i < 64; i++)
960 s->num_coded_frags[plane][i] -= blocks_ended;
961
962 // setup the next buffer
963 if (plane < 2)
964 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
965 else if (coeff_index < 63)
966 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
967
968 return eob_run;
969 }
970
971 static void reverse_dc_prediction(Vp3DecodeContext *s,
972 int first_fragment,
973 int fragment_width,
974 int fragment_height);
975 /*
976 * This function unpacks all of the DCT coefficient data from the
977 * bitstream.
978 */
979 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
980 {
981 int i;
982 int dc_y_table;
983 int dc_c_table;
984 int ac_y_table;
985 int ac_c_table;
986 int residual_eob_run = 0;
987 VLC *y_tables[64];
988 VLC *c_tables[64];
989
990 s->dct_tokens[0][0] = s->dct_tokens_base;
991
992 /* fetch the DC table indexes */
993 dc_y_table = get_bits(gb, 4);
994 dc_c_table = get_bits(gb, 4);
995
996 /* unpack the Y plane DC coefficients */
997 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
998 0, residual_eob_run);
999 if (residual_eob_run < 0)
1000 return residual_eob_run;
1001
1002 /* reverse prediction of the Y-plane DC coefficients */
1003 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1004
1005 /* unpack the C plane DC coefficients */
1006 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1007 1, residual_eob_run);
1008 if (residual_eob_run < 0)
1009 return residual_eob_run;
1010 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1011 2, residual_eob_run);
1012 if (residual_eob_run < 0)
1013 return residual_eob_run;
1014
1015 /* reverse prediction of the C-plane DC coefficients */
1016 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1017 {
1018 reverse_dc_prediction(s, s->fragment_start[1],
1019 s->fragment_width[1], s->fragment_height[1]);
1020 reverse_dc_prediction(s, s->fragment_start[2],
1021 s->fragment_width[1], s->fragment_height[1]);
1022 }
1023
1024 /* fetch the AC table indexes */
1025 ac_y_table = get_bits(gb, 4);
1026 ac_c_table = get_bits(gb, 4);
1027
1028 /* build tables of AC VLC tables */
1029 for (i = 1; i <= 5; i++) {
1030 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1031 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1032 }
1033 for (i = 6; i <= 14; i++) {
1034 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1035 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1036 }
1037 for (i = 15; i <= 27; i++) {
1038 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1039 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1040 }
1041 for (i = 28; i <= 63; i++) {
1042 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1043 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1044 }
1045
1046 /* decode all AC coefficents */
1047 for (i = 1; i <= 63; i++) {
1048 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1049 0, residual_eob_run);
1050 if (residual_eob_run < 0)
1051 return residual_eob_run;
1052
1053 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1054 1, residual_eob_run);
1055 if (residual_eob_run < 0)
1056 return residual_eob_run;
1057 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1058 2, residual_eob_run);
1059 if (residual_eob_run < 0)
1060 return residual_eob_run;
1061 }
1062
1063 return 0;
1064 }
1065
1066 /*
1067 * This function reverses the DC prediction for each coded fragment in
1068 * the frame. Much of this function is adapted directly from the original
1069 * VP3 source code.
1070 */
1071 #define COMPATIBLE_FRAME(x) \
1072 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1073 #define DC_COEFF(u) s->all_fragments[u].dc
1074
1075 static void reverse_dc_prediction(Vp3DecodeContext *s,
1076 int first_fragment,
1077 int fragment_width,
1078 int fragment_height)
1079 {
1080
1081 #define PUL 8
1082 #define PU 4
1083 #define PUR 2
1084 #define PL 1
1085
1086 int x, y;
1087 int i = first_fragment;
1088
1089 int predicted_dc;
1090
1091 /* DC values for the left, up-left, up, and up-right fragments */
1092 int vl, vul, vu, vur;
1093
1094 /* indexes for the left, up-left, up, and up-right fragments */
1095 int l, ul, u, ur;
1096
1097 /*
1098 * The 6 fields mean:
1099 * 0: up-left multiplier
1100 * 1: up multiplier
1101 * 2: up-right multiplier
1102 * 3: left multiplier
1103 */
1104 static const int predictor_transform[16][4] = {
1105 { 0, 0, 0, 0},
1106 { 0, 0, 0,128}, // PL
1107 { 0, 0,128, 0}, // PUR
1108 { 0, 0, 53, 75}, // PUR|PL
1109 { 0,128, 0, 0}, // PU
1110 { 0, 64, 0, 64}, // PU|PL
1111 { 0,128, 0, 0}, // PU|PUR
1112 { 0, 0, 53, 75}, // PU|PUR|PL
1113 {128, 0, 0, 0}, // PUL
1114 { 0, 0, 0,128}, // PUL|PL
1115 { 64, 0, 64, 0}, // PUL|PUR
1116 { 0, 0, 53, 75}, // PUL|PUR|PL
1117 { 0,128, 0, 0}, // PUL|PU
1118 {-104,116, 0,116}, // PUL|PU|PL
1119 { 24, 80, 24, 0}, // PUL|PU|PUR
1120 {-104,116, 0,116} // PUL|PU|PUR|PL
1121 };
1122
1123 /* This table shows which types of blocks can use other blocks for
1124 * prediction. For example, INTRA is the only mode in this table to
1125 * have a frame number of 0. That means INTRA blocks can only predict
1126 * from other INTRA blocks. There are 2 golden frame coding types;
1127 * blocks encoding in these modes can only predict from other blocks
1128 * that were encoded with these 1 of these 2 modes. */
1129 static const unsigned char compatible_frame[9] = {
1130 1, /* MODE_INTER_NO_MV */
1131 0, /* MODE_INTRA */
1132 1, /* MODE_INTER_PLUS_MV */
1133 1, /* MODE_INTER_LAST_MV */
1134 1, /* MODE_INTER_PRIOR_MV */
1135 2, /* MODE_USING_GOLDEN */
1136 2, /* MODE_GOLDEN_MV */
1137 1, /* MODE_INTER_FOUR_MV */
1138 3 /* MODE_COPY */
1139 };
1140 int current_frame_type;
1141
1142 /* there is a last DC predictor for each of the 3 frame types */
1143 short last_dc[3];
1144
1145 int transform = 0;
1146
1147 vul = vu = vur = vl = 0;
1148 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1149
1150 /* for each fragment row... */
1151 for (y = 0; y < fragment_height; y++) {
1152
1153 /* for each fragment in a row... */
1154 for (x = 0; x < fragment_width; x++, i++) {
1155
1156 /* reverse prediction if this block was coded */
1157 if (s->all_fragments[i].coding_method != MODE_COPY) {
1158
1159 current_frame_type =
1160 compatible_frame[s->all_fragments[i].coding_method];
1161
1162 transform= 0;
1163 if(x){
1164 l= i-1;
1165 vl = DC_COEFF(l);
1166 if(COMPATIBLE_FRAME(l))
1167 transform |= PL;
1168 }
1169 if(y){
1170 u= i-fragment_width;
1171 vu = DC_COEFF(u);
1172 if(COMPATIBLE_FRAME(u))
1173 transform |= PU;
1174 if(x){
1175 ul= i-fragment_width-1;
1176 vul = DC_COEFF(ul);
1177 if(COMPATIBLE_FRAME(ul))
1178 transform |= PUL;
1179 }
1180 if(x + 1 < fragment_width){
1181 ur= i-fragment_width+1;
1182 vur = DC_COEFF(ur);
1183 if(COMPATIBLE_FRAME(ur))
1184 transform |= PUR;
1185 }
1186 }
1187
1188 if (transform == 0) {
1189
1190 /* if there were no fragments to predict from, use last
1191 * DC saved */
1192 predicted_dc = last_dc[current_frame_type];
1193 } else {
1194
1195 /* apply the appropriate predictor transform */
1196 predicted_dc =
1197 (predictor_transform[transform][0] * vul) +
1198 (predictor_transform[transform][1] * vu) +
1199 (predictor_transform[transform][2] * vur) +
1200 (predictor_transform[transform][3] * vl);
1201
1202 predicted_dc /= 128;
1203
1204 /* check for outranging on the [ul u l] and
1205 * [ul u ur l] predictors */
1206 if ((transform == 15) || (transform == 13)) {
1207 if (FFABS(predicted_dc - vu) > 128)
1208 predicted_dc = vu;
1209 else if (FFABS(predicted_dc - vl) > 128)
1210 predicted_dc = vl;
1211 else if (FFABS(predicted_dc - vul) > 128)
1212 predicted_dc = vul;
1213 }
1214 }
1215
1216 /* at long last, apply the predictor */
1217 DC_COEFF(i) += predicted_dc;
1218 /* save the DC */
1219 last_dc[current_frame_type] = DC_COEFF(i);
1220 }
1221 }
1222 }
1223 }
1224
1225 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1226 {
1227 int x, y;
1228 int *bounding_values= s->bounding_values_array+127;
1229
1230 int width = s->fragment_width[!!plane];
1231 int height = s->fragment_height[!!plane];
1232 int fragment = s->fragment_start [plane] + ystart * width;
1233 int stride = s->current_frame.linesize[plane];
1234 uint8_t *plane_data = s->current_frame.data [plane];
1235 if (!s->flipped_image) stride = -stride;
1236 plane_data += s->data_offset[plane] + 8*ystart*stride;
1237
1238 for (y = ystart; y < yend; y++) {
1239
1240 for (x = 0; x < width; x++) {
1241 /* This code basically just deblocks on the edges of coded blocks.
1242 * However, it has to be much more complicated because of the
1243 * braindamaged deblock ordering used in VP3/Theora. Order matters
1244 * because some pixels get filtered twice. */
1245 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1246 {
1247 /* do not perform left edge filter for left columns frags */
1248 if (x > 0) {
1249 s->dsp.vp3_h_loop_filter(
1250 plane_data + 8*x,
1251 stride, bounding_values);
1252 }
1253
1254 /* do not perform top edge filter for top row fragments */
1255 if (y > 0) {
1256 s->dsp.vp3_v_loop_filter(
1257 plane_data + 8*x,
1258 stride, bounding_values);
1259 }
1260
1261 /* do not perform right edge filter for right column
1262 * fragments or if right fragment neighbor is also coded
1263 * in this frame (it will be filtered in next iteration) */
1264 if ((x < width - 1) &&
1265 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1266 s->dsp.vp3_h_loop_filter(
1267 plane_data + 8*x + 8,
1268 stride, bounding_values);
1269 }
1270
1271 /* do not perform bottom edge filter for bottom row
1272 * fragments or if bottom fragment neighbor is also coded
1273 * in this frame (it will be filtered in the next row) */
1274 if ((y < height - 1) &&
1275 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1276 s->dsp.vp3_v_loop_filter(
1277 plane_data + 8*x + 8*stride,
1278 stride, bounding_values);
1279 }
1280 }
1281
1282 fragment++;
1283 }
1284 plane_data += 8*stride;
1285 }
1286 }
1287
1288 /**
1289 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1290 * for the next block in coding order
1291 */
1292 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1293 int plane, int inter, DCTELEM block[64])
1294 {
1295 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1296 uint8_t *perm = s->scantable.permutated;
1297 int i = 0;
1298
1299 do {
1300 int token = *s->dct_tokens[plane][i];
1301 switch (token & 3) {
1302 case 0: // EOB
1303 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1304 s->dct_tokens[plane][i]++;
1305 else
1306 *s->dct_tokens[plane][i] = token & ~3;
1307 goto end;
1308 case 1: // zero run
1309 s->dct_tokens[plane][i]++;
1310 i += (token >> 2) & 0x7f;
1311 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1312 i++;
1313 break;
1314 case 2: // coeff
1315 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1316 s->dct_tokens[plane][i++]++;
1317 break;
1318 default: // shouldn't happen
1319 return i;
1320 }
1321 } while (i < 64);
1322 end:
1323 // the actual DC+prediction is in the fragment structure
1324 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1325 return i;
1326 }
1327
1328 /**
1329 * called when all pixels up to row y are complete
1330 */
1331 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1332 {
1333 int h, cy;
1334 int offset[4];
1335
1336 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1337 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1338
1339 // At the end of the frame, report INT_MAX instead of the height of the frame.
1340 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1341 // they don't have to clip their values.
1342 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1343 }
1344
1345 if(s->avctx->draw_horiz_band==NULL)
1346 return;
1347
1348 h= y - s->last_slice_end;
1349 s->last_slice_end= y;
1350 y -= h;
1351
1352 if (!s->flipped_image) {
1353 y = s->avctx->height - y - h;
1354 }
1355
1356 cy = y >> s->chroma_y_shift;
1357 offset[0] = s->current_frame.linesize[0]*y;
1358 offset[1] = s->current_frame.linesize[1]*cy;
1359 offset[2] = s->current_frame.linesize[2]*cy;
1360 offset[3] = 0;
1361
1362 emms_c();
1363 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1364 }
1365
1366 /**
1367 * Wait for the reference frame of the current fragment.
1368 * The progress value is in luma pixel rows.
1369 */
1370 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1371 {
1372 AVFrame *ref_frame;
1373 int ref_row;
1374 int border = motion_y&1;
1375
1376 if (fragment->coding_method == MODE_USING_GOLDEN ||
1377 fragment->coding_method == MODE_GOLDEN_MV)
1378 ref_frame = &s->golden_frame;
1379 else
1380 ref_frame = &s->last_frame;
1381
1382 ref_row = y + (motion_y>>1);
1383 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1384
1385 ff_thread_await_progress(ref_frame, ref_row, 0);
1386 }
1387
1388 /*
1389 * Perform the final rendering for a particular slice of data.
1390 * The slice number ranges from 0..(c_superblock_height - 1).
1391 */
1392 static void render_slice(Vp3DecodeContext *s, int slice)
1393 {
1394 int x, y, i, j, fragment;
1395 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1396 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1397 int motion_halfpel_index;
1398 uint8_t *motion_source;
1399 int plane, first_pixel;
1400
1401 if (slice >= s->c_superblock_height)
1402 return;
1403
1404 for (plane = 0; plane < 3; plane++) {
1405 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1406 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1407 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1408 int stride = s->current_frame.linesize[plane];
1409 int plane_width = s->width >> (plane && s->chroma_x_shift);
1410 int plane_height = s->height >> (plane && s->chroma_y_shift);
1411 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1412
1413 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1414 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1415 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1416
1417 int fragment_width = s->fragment_width[!!plane];
1418 int fragment_height = s->fragment_height[!!plane];
1419 int fragment_start = s->fragment_start[plane];
1420 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1421
1422 if (!s->flipped_image) stride = -stride;
1423 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1424 continue;
1425
1426 /* for each superblock row in the slice (both of them)... */
1427 for (; sb_y < slice_height; sb_y++) {
1428
1429 /* for each superblock in a row... */
1430 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1431
1432 /* for each block in a superblock... */
1433 for (j = 0; j < 16; j++) {
1434 x = 4*sb_x + hilbert_offset[j][0];
1435 y = 4*sb_y + hilbert_offset[j][1];
1436 fragment = y*fragment_width + x;
1437
1438 i = fragment_start + fragment;
1439
1440 // bounds check
1441 if (x >= fragment_width || y >= fragment_height)
1442 continue;
1443
1444 first_pixel = 8*y*stride + 8*x;
1445
1446 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1447 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1448
1449 /* transform if this block was coded */
1450 if (s->all_fragments[i].coding_method != MODE_COPY) {
1451 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1452 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1453 motion_source= golden_plane;
1454 else
1455 motion_source= last_plane;
1456
1457 motion_source += first_pixel;
1458 motion_halfpel_index = 0;
1459
1460 /* sort out the motion vector if this fragment is coded
1461 * using a motion vector method */
1462 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1463 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1464 int src_x, src_y;
1465 motion_x = motion_val[fragment][0];
1466 motion_y = motion_val[fragment][1];
1467
1468 src_x= (motion_x>>1) + 8*x;
1469 src_y= (motion_y>>1) + 8*y;
1470
1471 motion_halfpel_index = motion_x & 0x01;
1472 motion_source += (motion_x >> 1);
1473
1474 motion_halfpel_index |= (motion_y & 0x01) << 1;
1475 motion_source += ((motion_y >> 1) * stride);
1476
1477 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1478 uint8_t *temp= s->edge_emu_buffer;
1479 if(stride<0) temp -= 8*stride;
1480
1481 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1482 motion_source= temp;
1483 }
1484 }
1485
1486
1487 /* first, take care of copying a block from either the
1488 * previous or the golden frame */
1489 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1490 /* Note, it is possible to implement all MC cases with
1491 put_no_rnd_pixels_l2 which would look more like the
1492 VP3 source but this would be slower as
1493 put_no_rnd_pixels_tab is better optimzed */
1494 if(motion_halfpel_index != 3){
1495 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1496 output_plane + first_pixel,
1497 motion_source, stride, 8);
1498 }else{
1499 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1500 s->dsp.put_no_rnd_pixels_l2[1](
1501 output_plane + first_pixel,
1502 motion_source - d,
1503 motion_source + stride + 1 + d,
1504 stride, 8);
1505 }
1506 }
1507
1508 s->dsp.clear_block(block);
1509
1510 /* invert DCT and place (or add) in final output */
1511
1512 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1513 vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1514 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1515 block[0] += 128<<3;
1516 s->dsp.idct_put(
1517 output_plane + first_pixel,
1518 stride,
1519 block);
1520 } else {
1521 if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
1522 s->dsp.idct_add(
1523 output_plane + first_pixel,
1524 stride,
1525 block);
1526 } else {
1527 s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1528 }
1529 }
1530 } else {
1531
1532 /* copy directly from the previous frame */
1533 s->dsp.put_pixels_tab[1][0](
1534 output_plane + first_pixel,
1535 last_plane + first_pixel,
1536 stride, 8);
1537
1538 }
1539 }
1540 }
1541
1542 // Filter up to the last row in the superblock row
1543 if (!s->skip_loop_filter)
1544 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1545 }
1546 }
1547
1548 /* this looks like a good place for slice dispatch... */
1549 /* algorithm:
1550 * if (slice == s->macroblock_height - 1)
1551 * dispatch (both last slice & 2nd-to-last slice);
1552 * else if (slice > 0)
1553 * dispatch (slice - 1);
1554 */
1555
1556 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1557 }
1558
1559 /// Allocate tables for per-frame data in Vp3DecodeContext
1560 static av_cold int allocate_tables(AVCodecContext *avctx)
1561 {
1562 Vp3DecodeContext *s = avctx->priv_data;
1563 int y_fragment_count, c_fragment_count;
1564
1565 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1566 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1567
1568 s->superblock_coding = av_malloc(s->superblock_count);
1569 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1570 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1571 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1572 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1573 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1574
1575 /* work out the block mapping tables */
1576 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1577 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1578
1579 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1580 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1581 !s->motion_val[0] || !s->motion_val[1]) {
1582 vp3_decode_end(avctx);
1583 return -1;
1584 }
1585
1586 init_block_mapping(s);
1587
1588 return 0;
1589 }
1590
1591 /*
1592 * This is the ffmpeg/libavcodec API init function.
1593 */
1594 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1595 {
1596 Vp3DecodeContext *s = avctx->priv_data;
1597 int i, inter, plane;
1598 int c_width;
1599 int c_height;
1600 int y_fragment_count, c_fragment_count;
1601
1602 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1603 s->version = 0;
1604 else
1605 s->version = 1;
1606
1607 s->avctx = avctx;
1608 s->width = FFALIGN(avctx->width, 16);
1609 s->height = FFALIGN(avctx->height, 16);
1610 if (avctx->pix_fmt == PIX_FMT_NONE)
1611 avctx->pix_fmt = PIX_FMT_YUV420P;
1612 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1613 if(avctx->idct_algo==FF_IDCT_AUTO)
1614 avctx->idct_algo=FF_IDCT_VP3;
1615 dsputil_init(&s->dsp, avctx);
1616
1617 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1618
1619 /* initialize to an impossible value which will force a recalculation
1620 * in the first frame decode */
1621 for (i = 0; i < 3; i++)
1622 s->qps[i] = -1;
1623
1624 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1625
1626 s->y_superblock_width = (s->width + 31) / 32;
1627 s->y_superblock_height = (s->height + 31) / 32;
1628 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1629
1630 /* work out the dimensions for the C planes */
1631 c_width = s->width >> s->chroma_x_shift;
1632 c_height = s->height >> s->chroma_y_shift;
1633 s->c_superblock_width = (c_width + 31) / 32;
1634 s->c_superblock_height = (c_height + 31) / 32;
1635 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1636
1637 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1638 s->u_superblock_start = s->y_superblock_count;
1639 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1640
1641 s->macroblock_width = (s->width + 15) / 16;
1642 s->macroblock_height = (s->height + 15) / 16;
1643 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1644
1645 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1646 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1647 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1648 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1649
1650 /* fragment count covers all 8x8 blocks for all 3 planes */
1651 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1652 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1653 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1654 s->fragment_start[1] = y_fragment_count;
1655 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1656
1657 if (!s->theora_tables)
1658 {
1659 for (i = 0; i < 64; i++) {
1660 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1661 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1662 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1663 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1664 s->base_matrix[2][i] = vp31_inter_dequant[i];
1665 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1666 }
1667
1668 for(inter=0; inter<2; inter++){
1669 for(plane=0; plane<3; plane++){
1670 s->qr_count[inter][plane]= 1;
1671 s->qr_size [inter][plane][0]= 63;
1672 s->qr_base [inter][plane][0]=
1673 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1674 }
1675 }
1676
1677 /* init VLC tables */
1678 for (i = 0; i < 16; i++) {
1679
1680 /* DC histograms */
1681 init_vlc(&s->dc_vlc[i], 11, 32,
1682 &dc_bias[i][0][1], 4, 2,
1683 &dc_bias[i][0][0], 4, 2, 0);
1684
1685 /* group 1 AC histograms */
1686 init_vlc(&s->ac_vlc_1[i], 11, 32,
1687 &ac_bias_0[i][0][1], 4, 2,
1688 &ac_bias_0[i][0][0], 4, 2, 0);
1689
1690 /* group 2 AC histograms */
1691 init_vlc(&s->ac_vlc_2[i], 11, 32,
1692 &ac_bias_1[i][0][1], 4, 2,
1693 &ac_bias_1[i][0][0], 4, 2, 0);
1694
1695 /* group 3 AC histograms */
1696 init_vlc(&s->ac_vlc_3[i], 11, 32,
1697 &ac_bias_2[i][0][1], 4, 2,
1698 &ac_bias_2[i][0][0], 4, 2, 0);
1699
1700 /* group 4 AC histograms */
1701 init_vlc(&s->ac_vlc_4[i], 11, 32,
1702 &ac_bias_3[i][0][1], 4, 2,
1703 &ac_bias_3[i][0][0], 4, 2, 0);
1704 }
1705 } else {
1706
1707 for (i = 0; i < 16; i++) {
1708 /* DC histograms */
1709 if (init_vlc(&s->dc_vlc[i], 11, 32,
1710 &s->huffman_table[i][0][1], 8, 4,
1711 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1712 goto vlc_fail;
1713
1714 /* group 1 AC histograms */
1715 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1716 &s->huffman_table[i+16][0][1], 8, 4,
1717 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1718 goto vlc_fail;
1719
1720 /* group 2 AC histograms */
1721 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1722 &s->huffman_table[i+16*2][0][1], 8, 4,
1723 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1724 goto vlc_fail;
1725
1726 /* group 3 AC histograms */
1727 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1728 &s->huffman_table[i+16*3][0][1], 8, 4,
1729 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1730 goto vlc_fail;
1731
1732 /* group 4 AC histograms */
1733 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1734 &s->huffman_table[i+16*4][0][1], 8, 4,
1735 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1736 goto vlc_fail;
1737 }
1738 }
1739
1740 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1741 &superblock_run_length_vlc_table[0][1], 4, 2,
1742 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1743
1744 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1745 &fragment_run_length_vlc_table[0][1], 4, 2,
1746 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1747
1748 init_vlc(&s->mode_code_vlc, 3, 8,
1749 &mode_code_vlc_table[0][1], 2, 1,
1750 &mode_code_vlc_table[0][0], 2, 1, 0);
1751
1752 init_vlc(&s->motion_vector_vlc, 6, 63,
1753 &motion_vector_vlc_table[0][1], 2, 1,
1754 &motion_vector_vlc_table[0][0], 2, 1, 0);
1755
1756 for (i = 0; i < 3; i++) {
1757 s->current_frame.data[i] = NULL;
1758 s->last_frame.data[i] = NULL;
1759 s->golden_frame.data[i] = NULL;
1760 }
1761
1762 return allocate_tables(avctx);
1763
1764 vlc_fail:
1765 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1766 return -1;
1767 }
1768
1769 /// Release and shuffle frames after decode finishes
1770 static void update_frames(AVCodecContext *avctx)
1771 {
1772 Vp3DecodeContext *s = avctx->priv_data;
1773
1774 /* release the last frame, if it is allocated and if it is not the
1775 * golden frame */
1776 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1777 ff_thread_release_buffer(avctx, &s->last_frame);
1778
1779 /* shuffle frames (last = current) */
1780 s->last_frame= s->current_frame;
1781
1782 if (s->keyframe) {
1783 if (s->golden_frame.data[0])
1784 ff_thread_release_buffer(avctx, &s->golden_frame);
1785 s->golden_frame = s->current_frame;
1786 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1787 }
1788
1789 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1790 }
1791
1792 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1793 {
1794 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1795 int qps_changed = 0, i, err;
1796
1797 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1798
1799 if (!s1->current_frame.data[0]
1800 ||s->width != s1->width
1801 ||s->height!= s1->height) {
1802 if (s != s1)
1803 copy_fields(s, s1, golden_frame, current_frame);
1804 return -1;
1805 }
1806
1807 if (s != s1) {
1808 // init tables if the first frame hasn't been decoded
1809 if (!s->current_frame.data[0]) {
1810 int y_fragment_count, c_fragment_count;
1811 s->avctx = dst;
1812 err = allocate_tables(dst);
1813 if (err)
1814 return err;
1815 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1816 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1817 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1818 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1819 }
1820
1821 // copy previous frame data
1822 copy_fields(s, s1, golden_frame, dsp);
1823
1824 // copy qscale data if necessary
1825 for (i = 0; i < 3; i++) {
1826 if (s->qps[i] != s1->qps[1]) {
1827 qps_changed = 1;
1828 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1829 }
1830 }
1831
1832 if (s->qps[0] != s1->qps[0])
1833 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1834
1835 if (qps_changed)
1836 copy_fields(s, s1, qps, superblock_count);
1837 #undef copy_fields
1838 }
1839
1840 update_frames(dst);
1841
1842 return 0;
1843 }
1844
1845 /*
1846 * This is the ffmpeg/libavcodec API frame decode function.
1847 */
1848 static int vp3_decode_frame(AVCodecContext *avctx,
1849 void *data, int *data_size,
1850 AVPacket *avpkt)
1851 {
1852 const uint8_t *buf = avpkt->data;
1853 int buf_size = avpkt->size;
1854 Vp3DecodeContext *s = avctx->priv_data;
1855 GetBitContext gb;
1856 int i;
1857
1858 init_get_bits(&gb, buf, buf_size * 8);
1859
1860 if (s->theora && get_bits1(&gb))
1861 {
1862 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1863 return -1;
1864 }
1865
1866 s->keyframe = !get_bits1(&gb);
1867 if (!s->theora)
1868 skip_bits(&gb, 1);
1869 for (i = 0; i < 3; i++)
1870 s->last_qps[i] = s->qps[i];
1871
1872 s->nqps=0;
1873 do{
1874 s->qps[s->nqps++]= get_bits(&gb, 6);
1875 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1876 for (i = s->nqps; i < 3; i++)
1877 s->qps[i] = -1;
1878
1879 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1880 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1881 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1882
1883 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1884 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1885
1886 if (s->qps[0] != s->last_qps[0])
1887 init_loop_filter(s);
1888
1889 for (i = 0; i < s->nqps; i++)
1890 // reinit all dequantizers if the first one changed, because
1891 // the DC of the first quantizer must be used for all matrices
1892 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1893 init_dequantizer(s, i);
1894
1895 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1896 return buf_size;
1897
1898 s->current_frame.reference = 3;
1899 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1900 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1901 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1902 goto error;
1903 }
1904
1905 if (!s->edge_emu_buffer)
1906 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1907
1908 if (s->keyframe) {
1909 if (!s->theora)
1910 {
1911 skip_bits(&gb, 4); /* width code */
1912 skip_bits(&gb, 4); /* height code */
1913 if (s->version)
1914 {
1915 s->version = get_bits(&gb, 5);
1916 if (avctx->frame_number == 0)
1917 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1918 }
1919 }
1920 if (s->version || s->theora)
1921 {
1922 if (get_bits1(&gb))
1923 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1924 skip_bits(&gb, 2); /* reserved? */
1925 }
1926 } else {
1927 if (!s->golden_frame.data[0]) {
1928 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1929
1930 s->golden_frame.reference = 3;
1931 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1932 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1933 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1934 goto error;
1935 }
1936 s->last_frame = s->golden_frame;
1937 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1938 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
1939 }
1940 }
1941
1942 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1943 ff_thread_finish_setup(avctx);
1944
1945 if (unpack_superblocks(s, &gb)){
1946 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1947 goto error;
1948 }
1949 if (unpack_modes(s, &gb)){
1950 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1951 goto error;
1952 }
1953 if (unpack_vectors(s, &gb)){
1954 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1955 goto error;
1956 }
1957 if (unpack_block_qpis(s, &gb)){
1958 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1959 goto error;
1960 }
1961 if (unpack_dct_coeffs(s, &gb)){
1962 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1963 goto error;
1964 }
1965
1966 for (i = 0; i < 3; i++) {
1967 int height = s->height >> (i && s->chroma_y_shift);
1968 if (s->flipped_image)
1969 s->data_offset[i] = 0;
1970 else
1971 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1972 }
1973
1974 s->last_slice_end = 0;
1975 for (i = 0; i < s->c_superblock_height; i++)
1976 render_slice(s, i);
1977
1978 // filter the last row
1979 for (i = 0; i < 3; i++) {
1980 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1981 apply_loop_filter(s, i, row, row+1);
1982 }
1983 vp3_draw_horiz_band(s, s->avctx->height);
1984
1985 *data_size=sizeof(AVFrame);
1986 *(AVFrame*)data= s->current_frame;
1987
1988 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1989 update_frames(avctx);
1990
1991 return buf_size;
1992
1993 error:
1994 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
1995
1996 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
1997 avctx->release_buffer(avctx, &s->current_frame);
1998
1999 return -1;
2000 }
2001
2002 /*
2003 * This is the ffmpeg/libavcodec API module cleanup function.
2004 */
2005 static av_cold int vp3_decode_end(AVCodecContext *avctx)
2006 {
2007 Vp3DecodeContext *s = avctx->priv_data;
2008 int i;
2009
2010 av_free(s->superblock_coding);
2011 av_free(s->all_fragments);
2012 av_free(s->coded_fragment_list[0]);
2013 av_free(s->dct_tokens_base);
2014 av_free(s->superblock_fragments);
2015 av_free(s->macroblock_coding);
2016 av_free(s->motion_val[0]);
2017 av_free(s->motion_val[1]);
2018 av_free(s->edge_emu_buffer);
2019
2020 if (avctx->is_copy) return 0;
2021
2022 for (i = 0; i < 16; i++) {
2023 free_vlc(&s->dc_vlc[i]);
2024 free_vlc(&s->ac_vlc_1[i]);
2025 free_vlc(&s->ac_vlc_2[i]);
2026 free_vlc(&s->ac_vlc_3[i]);
2027 free_vlc(&s->ac_vlc_4[i]);
2028 }
2029
2030 free_vlc(&s->superblock_run_length_vlc);
2031 free_vlc(&s->fragment_run_length_vlc);
2032 free_vlc(&s->mode_code_vlc);
2033 free_vlc(&s->motion_vector_vlc);
2034
2035 /* release all frames */
2036 vp3_decode_flush(avctx);
2037
2038 return 0;
2039 }
2040
2041 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2042 {
2043 Vp3DecodeContext *s = avctx->priv_data;
2044
2045 if (get_bits1(gb)) {
2046 int token;
2047 if (s->entries >= 32) { /* overflow */
2048 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2049 return -1;
2050 }
2051 token = get_bits(gb, 5);
2052 //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);
2053 s->huffman_table[s->hti][token][0] = s->hbits;
2054 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2055 s->entries++;
2056 }
2057 else {
2058 if (s->huff_code_size >= 32) {/* overflow */
2059 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2060 return -1;
2061 }
2062 s->huff_code_size++;
2063 s->hbits <<= 1;
2064 if (read_huffman_tree(avctx, gb))
2065 return -1;
2066 s->hbits |= 1;
2067 if (read_huffman_tree(avctx, gb))
2068 return -1;
2069 s->hbits >>= 1;
2070 s->huff_code_size--;
2071 }
2072 return 0;
2073 }
2074
2075 #if CONFIG_THEORA_DECODER
2076 static const enum PixelFormat theora_pix_fmts[4] = {
2077 PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
2078 };
2079
2080 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2081 {
2082 Vp3DecodeContext *s = avctx->priv_data;
2083 int visible_width, visible_height, colorspace;
2084 int offset_x = 0, offset_y = 0;
2085 AVRational fps, aspect;
2086
2087 s->theora = get_bits_long(gb, 24);
2088 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2089
2090 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2091 /* but previous versions have the image flipped relative to vp3 */
2092 if (s->theora < 0x030200)
2093 {
2094 s->flipped_image = 1;
2095 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2096 }
2097
2098 visible_width = s->width = get_bits(gb, 16) << 4;
2099 visible_height = s->height = get_bits(gb, 16) << 4;
2100
2101 if(av_image_check_size(s->width, s->height, 0, avctx)){
2102 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2103 s->width= s->height= 0;
2104 return -1;
2105 }
2106
2107 if (s->theora >= 0x030200) {
2108 visible_width = get_bits_long(gb, 24);
2109 visible_height = get_bits_long(gb, 24);
2110
2111 offset_x = get_bits(gb, 8); /* offset x */
2112 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2113 }
2114
2115 fps.num = get_bits_long(gb, 32);
2116 fps.den = get_bits_long(gb, 32);
2117 if (fps.num && fps.den) {
2118 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2119 fps.den, fps.num, 1<<30);
2120 }
2121
2122 aspect.num = get_bits_long(gb, 24);
2123 aspect.den = get_bits_long(gb, 24);
2124 if (aspect.num && aspect.den) {
2125 av_reduce(&avctx->sample_aspect_ratio.num,
2126 &avctx->sample_aspect_ratio.den,
2127 aspect.num, aspect.den, 1<<30);
2128 }
2129
2130 if (s->theora < 0x030200)
2131 skip_bits(gb, 5); /* keyframe frequency force */
2132 colorspace = get_bits(gb, 8);
2133 skip_bits(gb, 24); /* bitrate */
2134
2135 skip_bits(gb, 6); /* quality hint */
2136
2137 if (s->theora >= 0x030200)
2138 {
2139 skip_bits(gb, 5); /* keyframe frequency force */
2140 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2141 skip_bits(gb, 3); /* reserved */
2142 }
2143
2144 // align_get_bits(gb);
2145
2146 if ( visible_width <= s->width && visible_width > s->width-16
2147 && visible_height <= s->height && visible_height > s->height-16
2148 && !offset_x && (offset_y == s->height - visible_height))
2149 avcodec_set_dimensions(avctx, visible_width, visible_height);
2150 else
2151 avcodec_set_dimensions(avctx, s->width, s->height);
2152
2153 if (colorspace == 1) {
2154 avctx->color_primaries = AVCOL_PRI_BT470M;
2155 } else if (colorspace == 2) {
2156 avctx->color_primaries = AVCOL_PRI_BT470BG;
2157 }
2158 if (colorspace == 1 || colorspace == 2) {
2159 avctx->colorspace = AVCOL_SPC_BT470BG;
2160 avctx->color_trc = AVCOL_TRC_BT709;
2161 }
2162
2163 return 0;
2164 }
2165
2166 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2167 {
2168 Vp3DecodeContext *s = avctx->priv_data;
2169 int i, n, matrices, inter, plane;
2170
2171 if (s->theora >= 0x030200) {
2172 n = get_bits(gb, 3);
2173 /* loop filter limit values table */
2174 if (n)
2175 for (i = 0; i < 64; i++)
2176 s->filter_limit_values[i] = get_bits(gb, n);
2177 }
2178
2179 if (s->theora >= 0x030200)
2180 n = get_bits(gb, 4) + 1;
2181 else
2182 n = 16;
2183 /* quality threshold table */
2184 for (i = 0; i < 64; i++)
2185 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2186
2187 if (s->theora >= 0x030200)
2188 n = get_bits(gb, 4) + 1;
2189 else
2190 n = 16;
2191 /* dc scale factor table */
2192 for (i = 0; i < 64; i++)
2193 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2194
2195 if (s->theora >= 0x030200)
2196 matrices = get_bits(gb, 9) + 1;
2197 else
2198 matrices = 3;
2199
2200 if(matrices > 384){
2201 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2202 return -1;
2203 }
2204
2205 for(n=0; n<matrices; n++){
2206 for (i = 0; i < 64; i++)
2207 s->base_matrix[n][i]= get_bits(gb, 8);
2208 }
2209
2210 for (inter = 0; inter <= 1; inter++) {
2211 for (plane = 0; plane <= 2; plane++) {
2212 int newqr= 1;
2213 if (inter || plane > 0)
2214 newqr = get_bits1(gb);
2215 if (!newqr) {
2216 int qtj, plj;
2217 if(inter && get_bits1(gb)){
2218 qtj = 0;
2219 plj = plane;
2220 }else{
2221 qtj= (3*inter + plane - 1) / 3;
2222 plj= (plane + 2) % 3;
2223 }
2224 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2225 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2226 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2227 } else {
2228 int qri= 0;
2229 int qi = 0;
2230
2231 for(;;){
2232 i= get_bits(gb, av_log2(matrices-1)+1);
2233 if(i>= matrices){
2234 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2235 return -1;
2236 }
2237 s->qr_base[inter][plane][qri]= i;
2238 if(qi >= 63)
2239 break;
2240 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2241 s->qr_size[inter][plane][qri++]= i;
2242 qi += i;
2243 }
2244
2245 if (qi > 63) {
2246 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2247 return -1;
2248 }
2249 s->qr_count[inter][plane]= qri;
2250 }
2251 }
2252 }
2253
2254 /* Huffman tables */
2255 for (s->hti = 0; s->hti < 80; s->hti++) {
2256 s->entries = 0;
2257 s->huff_code_size = 1;
2258 if (!get_bits1(gb)) {
2259 s->hbits = 0;
2260 if(read_huffman_tree(avctx, gb))
2261 return -1;
2262 s->hbits = 1;
2263 if(read_huffman_tree(avctx, gb))
2264 return -1;
2265 }
2266 }
2267
2268 s->theora_tables = 1;
2269
2270 return 0;
2271 }
2272
2273 static av_cold int theora_decode_init(AVCodecContext *avctx)
2274 {
2275 Vp3DecodeContext *s = avctx->priv_data;
2276 GetBitContext gb;
2277 int ptype;
2278 uint8_t *header_start[3];
2279 int header_len[3];
2280 int i;
2281
2282 s->theora = 1;
2283
2284 if (!avctx->extradata_size)
2285 {
2286 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2287 return -1;
2288 }
2289
2290 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2291 42, header_start, header_len) < 0) {
2292 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2293 return -1;
2294 }
2295
2296 for(i=0;i<3;i++) {
2297 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2298
2299 ptype = get_bits(&gb, 8);
2300
2301 if (!(ptype & 0x80))
2302 {
2303 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2304 // return -1;
2305 }
2306
2307 // FIXME: Check for this as well.
2308 skip_bits_long(&gb, 6*8); /* "theora" */
2309
2310 switch(ptype)
2311 {
2312 case 0x80:
2313 theora_decode_header(avctx, &gb);
2314 break;
2315 case 0x81:
2316 // FIXME: is this needed? it breaks sometimes
2317 // theora_decode_comments(avctx, gb);
2318 break;
2319 case 0x82:
2320 if (theora_decode_tables(avctx, &gb))
2321 return -1;
2322 break;
2323 default:
2324 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2325 break;
2326 }
2327 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2328 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2329 if (s->theora < 0x030200)
2330 break;
2331 }
2332
2333 return vp3_decode_init(avctx);
2334 }
2335
2336 static void vp3_decode_flush(AVCodecContext *avctx)
2337 {
2338 Vp3DecodeContext *s = avctx->priv_data;
2339
2340 if (s->golden_frame.data[0]) {
2341 if (s->golden_frame.data[0] == s->last_frame.data[0])
2342 memset(&s->last_frame, 0, sizeof(AVFrame));
2343 if (s->current_frame.data[0] == s->golden_frame.data[0])
2344 memset(&s->current_frame, 0, sizeof(AVFrame));
2345 ff_thread_release_buffer(avctx, &s->golden_frame);
2346 }
2347 if (s->last_frame.data[0]) {
2348 if (s->current_frame.data[0] == s->last_frame.data[0])
2349 memset(&s->current_frame, 0, sizeof(AVFrame));
2350 ff_thread_release_buffer(avctx, &s->last_frame);
2351 }
2352 if (s->current_frame.data[0])
2353 ff_thread_release_buffer(avctx, &s->current_frame);
2354 }
2355
2356 static int vp3_init_thread_copy(AVCodecContext *avctx)
2357 {
2358 Vp3DecodeContext *s = avctx->priv_data;
2359
2360 s->superblock_coding = NULL;
2361 s->all_fragments = NULL;
2362 s->coded_fragment_list[0] = NULL;
2363 s->dct_tokens_base = NULL;
2364 s->superblock_fragments = NULL;
2365 s->macroblock_coding = NULL;
2366 s->motion_val[0] = NULL;
2367 s->motion_val[1] = NULL;
2368 s->edge_emu_buffer = NULL;
2369
2370 return 0;
2371 }
2372
2373 AVCodec ff_theora_decoder = {
2374 .name = "theora",
2375 .type = AVMEDIA_TYPE_VIDEO,
2376 .id = CODEC_ID_THEORA,
2377 .priv_data_size = sizeof(Vp3DecodeContext),
2378 .init = theora_decode_init,
2379 .close = vp3_decode_end,
2380 .decode = vp3_decode_frame,
2381 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2382 .flush = vp3_decode_flush,
2383 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2384 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2385 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2386 };
2387 #endif
2388
2389 AVCodec ff_vp3_decoder = {
2390 .name = "vp3",
2391 .type = AVMEDIA_TYPE_VIDEO,
2392 .id = CODEC_ID_VP3,
2393 .priv_data_size = sizeof(Vp3DecodeContext),
2394 .init = vp3_decode_init,
2395 .close = vp3_decode_end,
2396 .decode = vp3_decode_frame,
2397 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_FRAME_THREADS,
2398 .flush = vp3_decode_flush,
2399 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2400 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2401 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2402 };