?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
h264: simplify calls to ff_er_add_slice().
[FFMpeg-mirror/mplayer-patches.git] / libavcodec / vp3.c
blobf46ac32a9aa092cafc0dbe767b0cb074885e59b2
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 #include "videodsp.h"
42 #include "vp3data.h"
43 #include "vp3dsp.h"
44 #include "xiph.h"
45 #include "thread.h"
46
47 #define FRAGMENT_PIXELS 8
48
49 //FIXME split things out into their own arrays
50 typedef struct Vp3Fragment {
51 int16_t dc;
52 uint8_t coding_method;
53 uint8_t qpi;
54 } Vp3Fragment;
55
56 #define SB_NOT_CODED 0
57 #define SB_PARTIALLY_CODED 1
58 #define SB_FULLY_CODED 2
59
60 // This is the maximum length of a single long bit run that can be encoded
61 // for superblock coding or block qps. Theora special-cases this to read a
62 // bit instead of flipping the current bit to allow for runs longer than 4129.
63 #define MAXIMUM_LONG_BIT_RUN 4129
64
65 #define MODE_INTER_NO_MV 0
66 #define MODE_INTRA 1
67 #define MODE_INTER_PLUS_MV 2
68 #define MODE_INTER_LAST_MV 3
69 #define MODE_INTER_PRIOR_LAST 4
70 #define MODE_USING_GOLDEN 5
71 #define MODE_GOLDEN_MV 6
72 #define MODE_INTER_FOURMV 7
73 #define CODING_MODE_COUNT 8
74
75 /* special internal mode */
76 #define MODE_COPY 8
77
78 /* There are 6 preset schemes, plus a free-form scheme */
79 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
80 {
81 /* scheme 1: Last motion vector dominates */
82 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
83 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
84 MODE_INTRA, MODE_USING_GOLDEN,
85 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
86
87 /* scheme 2 */
88 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
89 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
90 MODE_INTRA, MODE_USING_GOLDEN,
91 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
92
93 /* scheme 3 */
94 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
95 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
96 MODE_INTRA, MODE_USING_GOLDEN,
97 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
98
99 /* scheme 4 */
100 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
101 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
102 MODE_INTRA, MODE_USING_GOLDEN,
103 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
104
105 /* scheme 5: No motion vector dominates */
106 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
107 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
108 MODE_INTRA, MODE_USING_GOLDEN,
109 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
110
111 /* scheme 6 */
112 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
113 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
114 MODE_INTER_PLUS_MV, MODE_INTRA,
115 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
116
117 };
118
119 static const uint8_t hilbert_offset[16][2] = {
120 {0,0}, {1,0}, {1,1}, {0,1},
121 {0,2}, {0,3}, {1,3}, {1,2},
122 {2,2}, {2,3}, {3,3}, {3,2},
123 {3,1}, {2,1}, {2,0}, {3,0}
124 };
125
126 #define MIN_DEQUANT_VAL 2
127
128 typedef struct Vp3DecodeContext {
129 AVCodecContext *avctx;
130 int theora, theora_tables;
131 int version;
132 int width, height;
133 int chroma_x_shift, chroma_y_shift;
134 AVFrame golden_frame;
135 AVFrame last_frame;
136 AVFrame current_frame;
137 int keyframe;
138 DSPContext dsp;
139 VideoDSPContext vdsp;
140 VP3DSPContext vp3dsp;
141 DECLARE_ALIGNED(16, int16_t, block)[64];
142 int flipped_image;
143 int last_slice_end;
144 int skip_loop_filter;
145
146 int qps[3];
147 int nqps;
148 int last_qps[3];
149
150 int superblock_count;
151 int y_superblock_width;
152 int y_superblock_height;
153 int y_superblock_count;
154 int c_superblock_width;
155 int c_superblock_height;
156 int c_superblock_count;
157 int u_superblock_start;
158 int v_superblock_start;
159 unsigned char *superblock_coding;
160
161 int macroblock_count;
162 int macroblock_width;
163 int macroblock_height;
164
165 int fragment_count;
166 int fragment_width[2];
167 int fragment_height[2];
168
169 Vp3Fragment *all_fragments;
170 int fragment_start[3];
171 int data_offset[3];
172
173 int8_t (*motion_val[2])[2];
174
175 ScanTable scantable;
176
177 /* tables */
178 uint16_t coded_dc_scale_factor[64];
179 uint32_t coded_ac_scale_factor[64];
180 uint8_t base_matrix[384][64];
181 uint8_t qr_count[2][3];
182 uint8_t qr_size [2][3][64];
183 uint16_t qr_base[2][3][64];
184
185 /**
186 * This is a list of all tokens in bitstream order. Reordering takes place
187 * by pulling from each level during IDCT. As a consequence, IDCT must be
188 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
189 * otherwise. The 32 different tokens with up to 12 bits of extradata are
190 * collapsed into 3 types, packed as follows:
191 * (from the low to high bits)
192 *
193 * 2 bits: type (0,1,2)
194 * 0: EOB run, 14 bits for run length (12 needed)
195 * 1: zero run, 7 bits for run length
196 * 7 bits for the next coefficient (3 needed)
197 * 2: coefficient, 14 bits (11 needed)
198 *
199 * Coefficients are signed, so are packed in the highest bits for automatic
200 * sign extension.
201 */
202 int16_t *dct_tokens[3][64];
203 int16_t *dct_tokens_base;
204 #define TOKEN_EOB(eob_run) ((eob_run) << 2)
205 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
206 #define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
207
208 /**
209 * number of blocks that contain DCT coefficients at the given level or higher
210 */
211 int num_coded_frags[3][64];
212 int total_num_coded_frags;
213
214 /* this is a list of indexes into the all_fragments array indicating
215 * which of the fragments are coded */
216 int *coded_fragment_list[3];
217
218 VLC dc_vlc[16];
219 VLC ac_vlc_1[16];
220 VLC ac_vlc_2[16];
221 VLC ac_vlc_3[16];
222 VLC ac_vlc_4[16];
223
224 VLC superblock_run_length_vlc;
225 VLC fragment_run_length_vlc;
226 VLC mode_code_vlc;
227 VLC motion_vector_vlc;
228
229 /* these arrays need to be on 16-byte boundaries since SSE2 operations
230 * index into them */
231 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
232
233 /* This table contains superblock_count * 16 entries. Each set of 16
234 * numbers corresponds to the fragment indexes 0..15 of the superblock.
235 * An entry will be -1 to indicate that no entry corresponds to that
236 * index. */
237 int *superblock_fragments;
238
239 /* This is an array that indicates how a particular macroblock
240 * is coded. */
241 unsigned char *macroblock_coding;
242
243 uint8_t *edge_emu_buffer;
244
245 /* Huffman decode */
246 int hti;
247 unsigned int hbits;
248 int entries;
249 int huff_code_size;
250 uint32_t huffman_table[80][32][2];
251
252 uint8_t filter_limit_values[64];
253 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
254 } Vp3DecodeContext;
255
256 /************************************************************************
257 * VP3 specific functions
258 ************************************************************************/
259
260 static void vp3_decode_flush(AVCodecContext *avctx)
261 {
262 Vp3DecodeContext *s = avctx->priv_data;
263
264 if (s->golden_frame.data[0]) {
265 if (s->golden_frame.data[0] == s->last_frame.data[0])
266 memset(&s->last_frame, 0, sizeof(AVFrame));
267 if (s->current_frame.data[0] == s->golden_frame.data[0])
268 memset(&s->current_frame, 0, sizeof(AVFrame));
269 ff_thread_release_buffer(avctx, &s->golden_frame);
270 }
271 if (s->last_frame.data[0]) {
272 if (s->current_frame.data[0] == s->last_frame.data[0])
273 memset(&s->current_frame, 0, sizeof(AVFrame));
274 ff_thread_release_buffer(avctx, &s->last_frame);
275 }
276 if (s->current_frame.data[0])
277 ff_thread_release_buffer(avctx, &s->current_frame);
278 }
279
280 static av_cold int vp3_decode_end(AVCodecContext *avctx)
281 {
282 Vp3DecodeContext *s = avctx->priv_data;
283 int i;
284
285 av_freep(&s->superblock_coding);
286 av_freep(&s->all_fragments);
287 av_freep(&s->coded_fragment_list[0]);
288 av_freep(&s->dct_tokens_base);
289 av_freep(&s->superblock_fragments);
290 av_freep(&s->macroblock_coding);
291 av_freep(&s->motion_val[0]);
292 av_freep(&s->motion_val[1]);
293 av_freep(&s->edge_emu_buffer);
294
295 if (avctx->internal->is_copy)
296 return 0;
297
298 for (i = 0; i < 16; i++) {
299 ff_free_vlc(&s->dc_vlc[i]);
300 ff_free_vlc(&s->ac_vlc_1[i]);
301 ff_free_vlc(&s->ac_vlc_2[i]);
302 ff_free_vlc(&s->ac_vlc_3[i]);
303 ff_free_vlc(&s->ac_vlc_4[i]);
304 }
305
306 ff_free_vlc(&s->superblock_run_length_vlc);
307 ff_free_vlc(&s->fragment_run_length_vlc);
308 ff_free_vlc(&s->mode_code_vlc);
309 ff_free_vlc(&s->motion_vector_vlc);
310
311 /* release all frames */
312 vp3_decode_flush(avctx);
313
314 return 0;
315 }
316
317 /*
318 * This function sets up all of the various blocks mappings:
319 * superblocks <-> fragments, macroblocks <-> fragments,
320 * superblocks <-> macroblocks
321 *
322 * @return 0 is successful; returns 1 if *anything* went wrong.
323 */
324 static int init_block_mapping(Vp3DecodeContext *s)
325 {
326 int sb_x, sb_y, plane;
327 int x, y, i, j = 0;
328
329 for (plane = 0; plane < 3; plane++) {
330 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
331 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
332 int frag_width = s->fragment_width[!!plane];
333 int frag_height = s->fragment_height[!!plane];
334
335 for (sb_y = 0; sb_y < sb_height; sb_y++)
336 for (sb_x = 0; sb_x < sb_width; sb_x++)
337 for (i = 0; i < 16; i++) {
338 x = 4*sb_x + hilbert_offset[i][0];
339 y = 4*sb_y + hilbert_offset[i][1];
340
341 if (x < frag_width && y < frag_height)
342 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
343 else
344 s->superblock_fragments[j++] = -1;
345 }
346 }
347
348 return 0; /* successful path out */
349 }
350
351 /*
352 * This function sets up the dequantization tables used for a particular
353 * frame.
354 */
355 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
356 {
357 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
358 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
359 int i, plane, inter, qri, bmi, bmj, qistart;
360
361 for(inter=0; inter<2; inter++){
362 for(plane=0; plane<3; plane++){
363 int sum=0;
364 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
365 sum+= s->qr_size[inter][plane][qri];
366 if(s->qps[qpi] <= sum)
367 break;
368 }
369 qistart= sum - s->qr_size[inter][plane][qri];
370 bmi= s->qr_base[inter][plane][qri ];
371 bmj= s->qr_base[inter][plane][qri+1];
372 for(i=0; i<64; i++){
373 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
374 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
375 + s->qr_size[inter][plane][qri])
376 / (2*s->qr_size[inter][plane][qri]);
377
378 int qmin= 8<<(inter + !i);
379 int qscale= i ? ac_scale_factor : dc_scale_factor;
380
381 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
382 }
383 // all DC coefficients use the same quant so as not to interfere with DC prediction
384 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
385 }
386 }
387 }
388
389 /*
390 * This function initializes the loop filter boundary limits if the frame's
391 * quality index is different from the previous frame's.
392 *
393 * The filter_limit_values may not be larger than 127.
394 */
395 static void init_loop_filter(Vp3DecodeContext *s)
396 {
397 int *bounding_values= s->bounding_values_array+127;
398 int filter_limit;
399 int x;
400 int value;
401
402 filter_limit = s->filter_limit_values[s->qps[0]];
403 assert(filter_limit < 128);
404
405 /* set up the bounding values */
406 memset(s->bounding_values_array, 0, 256 * sizeof(int));
407 for (x = 0; x < filter_limit; x++) {
408 bounding_values[-x] = -x;
409 bounding_values[x] = x;
410 }
411 for (x = value = filter_limit; x < 128 && value; x++, value--) {
412 bounding_values[ x] = value;
413 bounding_values[-x] = -value;
414 }
415 if (value)
416 bounding_values[128] = value;
417 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
418 }
419
420 /*
421 * This function unpacks all of the superblock/macroblock/fragment coding
422 * information from the bitstream.
423 */
424 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
425 {
426 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
427 int bit = 0;
428 int current_superblock = 0;
429 int current_run = 0;
430 int num_partial_superblocks = 0;
431
432 int i, j;
433 int current_fragment;
434 int plane;
435
436 if (s->keyframe) {
437 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
438
439 } else {
440
441 /* unpack the list of partially-coded superblocks */
442 bit = get_bits1(gb) ^ 1;
443 current_run = 0;
444
445 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
446 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
447 bit = get_bits1(gb);
448 else
449 bit ^= 1;
450
451 current_run = get_vlc2(gb,
452 s->superblock_run_length_vlc.table, 6, 2) + 1;
453 if (current_run == 34)
454 current_run += get_bits(gb, 12);
455
456 if (current_superblock + current_run > s->superblock_count) {
457 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
458 return -1;
459 }
460
461 memset(s->superblock_coding + current_superblock, bit, current_run);
462
463 current_superblock += current_run;
464 if (bit)
465 num_partial_superblocks += current_run;
466 }
467
468 /* unpack the list of fully coded superblocks if any of the blocks were
469 * not marked as partially coded in the previous step */
470 if (num_partial_superblocks < s->superblock_count) {
471 int superblocks_decoded = 0;
472
473 current_superblock = 0;
474 bit = get_bits1(gb) ^ 1;
475 current_run = 0;
476
477 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
478 && get_bits_left(gb) > 0) {
479
480 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
481 bit = get_bits1(gb);
482 else
483 bit ^= 1;
484
485 current_run = get_vlc2(gb,
486 s->superblock_run_length_vlc.table, 6, 2) + 1;
487 if (current_run == 34)
488 current_run += get_bits(gb, 12);
489
490 for (j = 0; j < current_run; current_superblock++) {
491 if (current_superblock >= s->superblock_count) {
492 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
493 return -1;
494 }
495
496 /* skip any superblocks already marked as partially coded */
497 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
498 s->superblock_coding[current_superblock] = 2*bit;
499 j++;
500 }
501 }
502 superblocks_decoded += current_run;
503 }
504 }
505
506 /* if there were partial blocks, initialize bitstream for
507 * unpacking fragment codings */
508 if (num_partial_superblocks) {
509
510 current_run = 0;
511 bit = get_bits1(gb);
512 /* toggle the bit because as soon as the first run length is
513 * fetched the bit will be toggled again */
514 bit ^= 1;
515 }
516 }
517
518 /* figure out which fragments are coded; iterate through each
519 * superblock (all planes) */
520 s->total_num_coded_frags = 0;
521 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
522
523 for (plane = 0; plane < 3; plane++) {
524 int sb_start = superblock_starts[plane];
525 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
526 int num_coded_frags = 0;
527
528 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
529
530 /* iterate through all 16 fragments in a superblock */
531 for (j = 0; j < 16; j++) {
532
533 /* if the fragment is in bounds, check its coding status */
534 current_fragment = s->superblock_fragments[i * 16 + j];
535 if (current_fragment != -1) {
536 int coded = s->superblock_coding[i];
537
538 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
539
540 /* fragment may or may not be coded; this is the case
541 * that cares about the fragment coding runs */
542 if (current_run-- == 0) {
543 bit ^= 1;
544 current_run = get_vlc2(gb,
545 s->fragment_run_length_vlc.table, 5, 2);
546 }
547 coded = bit;
548 }
549
550 if (coded) {
551 /* default mode; actual mode will be decoded in
552 * the next phase */
553 s->all_fragments[current_fragment].coding_method =
554 MODE_INTER_NO_MV;
555 s->coded_fragment_list[plane][num_coded_frags++] =
556 current_fragment;
557 } else {
558 /* not coded; copy this fragment from the prior frame */
559 s->all_fragments[current_fragment].coding_method =
560 MODE_COPY;
561 }
562 }
563 }
564 }
565 s->total_num_coded_frags += num_coded_frags;
566 for (i = 0; i < 64; i++)
567 s->num_coded_frags[plane][i] = num_coded_frags;
568 if (plane < 2)
569 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
570 }
571 return 0;
572 }
573
574 /*
575 * This function unpacks all the coding mode data for individual macroblocks
576 * from the bitstream.
577 */
578 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
579 {
580 int i, j, k, sb_x, sb_y;
581 int scheme;
582 int current_macroblock;
583 int current_fragment;
584 int coding_mode;
585 int custom_mode_alphabet[CODING_MODE_COUNT];
586 const int *alphabet;
587 Vp3Fragment *frag;
588
589 if (s->keyframe) {
590 for (i = 0; i < s->fragment_count; i++)
591 s->all_fragments[i].coding_method = MODE_INTRA;
592
593 } else {
594
595 /* fetch the mode coding scheme for this frame */
596 scheme = get_bits(gb, 3);
597
598 /* is it a custom coding scheme? */
599 if (scheme == 0) {
600 for (i = 0; i < 8; i++)
601 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
602 for (i = 0; i < 8; i++)
603 custom_mode_alphabet[get_bits(gb, 3)] = i;
604 alphabet = custom_mode_alphabet;
605 } else
606 alphabet = ModeAlphabet[scheme-1];
607
608 /* iterate through all of the macroblocks that contain 1 or more
609 * coded fragments */
610 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
611 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
612 if (get_bits_left(gb) <= 0)
613 return -1;
614
615 for (j = 0; j < 4; j++) {
616 int mb_x = 2*sb_x + (j>>1);
617 int mb_y = 2*sb_y + (((j>>1)+j)&1);
618 current_macroblock = mb_y * s->macroblock_width + mb_x;
619
620 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
621 continue;
622
623 #define BLOCK_X (2*mb_x + (k&1))
624 #define BLOCK_Y (2*mb_y + (k>>1))
625 /* coding modes are only stored if the macroblock has at least one
626 * luma block coded, otherwise it must be INTER_NO_MV */
627 for (k = 0; k < 4; k++) {
628 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
629 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
630 break;
631 }
632 if (k == 4) {
633 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
634 continue;
635 }
636
637 /* mode 7 means get 3 bits for each coding mode */
638 if (scheme == 7)
639 coding_mode = get_bits(gb, 3);
640 else
641 coding_mode = alphabet
642 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
643
644 s->macroblock_coding[current_macroblock] = coding_mode;
645 for (k = 0; k < 4; k++) {
646 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
647 if (frag->coding_method != MODE_COPY)
648 frag->coding_method = coding_mode;
649 }
650
651 #define SET_CHROMA_MODES \
652 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
653 frag[s->fragment_start[1]].coding_method = coding_mode;\
654 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
655 frag[s->fragment_start[2]].coding_method = coding_mode;
656
657 if (s->chroma_y_shift) {
658 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
659 SET_CHROMA_MODES
660 } else if (s->chroma_x_shift) {
661 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
662 for (k = 0; k < 2; k++) {
663 SET_CHROMA_MODES
664 frag += s->fragment_width[1];
665 }
666 } else {
667 for (k = 0; k < 4; k++) {
668 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
669 SET_CHROMA_MODES
670 }
671 }
672 }
673 }
674 }
675 }
676
677 return 0;
678 }
679
680 /*
681 * This function unpacks all the motion vectors for the individual
682 * macroblocks from the bitstream.
683 */
684 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
685 {
686 int j, k, sb_x, sb_y;
687 int coding_mode;
688 int motion_x[4];
689 int motion_y[4];
690 int last_motion_x = 0;
691 int last_motion_y = 0;
692 int prior_last_motion_x = 0;
693 int prior_last_motion_y = 0;
694 int current_macroblock;
695 int current_fragment;
696 int frag;
697
698 if (s->keyframe)
699 return 0;
700
701 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
702 coding_mode = get_bits1(gb);
703
704 /* iterate through all of the macroblocks that contain 1 or more
705 * coded fragments */
706 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
707 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
708 if (get_bits_left(gb) <= 0)
709 return -1;
710
711 for (j = 0; j < 4; j++) {
712 int mb_x = 2*sb_x + (j>>1);
713 int mb_y = 2*sb_y + (((j>>1)+j)&1);
714 current_macroblock = mb_y * s->macroblock_width + mb_x;
715
716 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
717 (s->macroblock_coding[current_macroblock] == MODE_COPY))
718 continue;
719
720 switch (s->macroblock_coding[current_macroblock]) {
721
722 case MODE_INTER_PLUS_MV:
723 case MODE_GOLDEN_MV:
724 /* all 6 fragments use the same motion vector */
725 if (coding_mode == 0) {
726 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
727 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
728 } else {
729 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
730 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
731 }
732
733 /* vector maintenance, only on MODE_INTER_PLUS_MV */
734 if (s->macroblock_coding[current_macroblock] ==
735 MODE_INTER_PLUS_MV) {
736 prior_last_motion_x = last_motion_x;
737 prior_last_motion_y = last_motion_y;
738 last_motion_x = motion_x[0];
739 last_motion_y = motion_y[0];
740 }
741 break;
742
743 case MODE_INTER_FOURMV:
744 /* vector maintenance */
745 prior_last_motion_x = last_motion_x;
746 prior_last_motion_y = last_motion_y;
747
748 /* fetch 4 vectors from the bitstream, one for each
749 * Y fragment, then average for the C fragment vectors */
750 for (k = 0; k < 4; k++) {
751 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
752 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
753 if (coding_mode == 0) {
754 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
755 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
756 } else {
757 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
758 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
759 }
760 last_motion_x = motion_x[k];
761 last_motion_y = motion_y[k];
762 } else {
763 motion_x[k] = 0;
764 motion_y[k] = 0;
765 }
766 }
767 break;
768
769 case MODE_INTER_LAST_MV:
770 /* all 6 fragments use the last motion vector */
771 motion_x[0] = last_motion_x;
772 motion_y[0] = last_motion_y;
773
774 /* no vector maintenance (last vector remains the
775 * last vector) */
776 break;
777
778 case MODE_INTER_PRIOR_LAST:
779 /* all 6 fragments use the motion vector prior to the
780 * last motion vector */
781 motion_x[0] = prior_last_motion_x;
782 motion_y[0] = prior_last_motion_y;
783
784 /* vector maintenance */
785 prior_last_motion_x = last_motion_x;
786 prior_last_motion_y = last_motion_y;
787 last_motion_x = motion_x[0];
788 last_motion_y = motion_y[0];
789 break;
790
791 default:
792 /* covers intra, inter without MV, golden without MV */
793 motion_x[0] = 0;
794 motion_y[0] = 0;
795
796 /* no vector maintenance */
797 break;
798 }
799
800 /* assign the motion vectors to the correct fragments */
801 for (k = 0; k < 4; k++) {
802 current_fragment =
803 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
804 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
805 s->motion_val[0][current_fragment][0] = motion_x[k];
806 s->motion_val[0][current_fragment][1] = motion_y[k];
807 } else {
808 s->motion_val[0][current_fragment][0] = motion_x[0];
809 s->motion_val[0][current_fragment][1] = motion_y[0];
810 }
811 }
812
813 if (s->chroma_y_shift) {
814 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
815 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
816 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
817 }
818 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
819 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
820 frag = mb_y*s->fragment_width[1] + mb_x;
821 s->motion_val[1][frag][0] = motion_x[0];
822 s->motion_val[1][frag][1] = motion_y[0];
823 } else if (s->chroma_x_shift) {
824 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
825 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
826 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
827 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
828 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
829 } else {
830 motion_x[1] = motion_x[0];
831 motion_y[1] = motion_y[0];
832 }
833 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
834 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
835
836 frag = 2*mb_y*s->fragment_width[1] + mb_x;
837 for (k = 0; k < 2; k++) {
838 s->motion_val[1][frag][0] = motion_x[k];
839 s->motion_val[1][frag][1] = motion_y[k];
840 frag += s->fragment_width[1];
841 }
842 } else {
843 for (k = 0; k < 4; k++) {
844 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
845 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
846 s->motion_val[1][frag][0] = motion_x[k];
847 s->motion_val[1][frag][1] = motion_y[k];
848 } else {
849 s->motion_val[1][frag][0] = motion_x[0];
850 s->motion_val[1][frag][1] = motion_y[0];
851 }
852 }
853 }
854 }
855 }
856 }
857
858 return 0;
859 }
860
861 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
862 {
863 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
864 int num_blocks = s->total_num_coded_frags;
865
866 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
867 i = blocks_decoded = num_blocks_at_qpi = 0;
868
869 bit = get_bits1(gb) ^ 1;
870 run_length = 0;
871
872 do {
873 if (run_length == MAXIMUM_LONG_BIT_RUN)
874 bit = get_bits1(gb);
875 else
876 bit ^= 1;
877
878 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
879 if (run_length == 34)
880 run_length += get_bits(gb, 12);
881 blocks_decoded += run_length;
882
883 if (!bit)
884 num_blocks_at_qpi += run_length;
885
886 for (j = 0; j < run_length; i++) {
887 if (i >= s->total_num_coded_frags)
888 return -1;
889
890 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
891 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
892 j++;
893 }
894 }
895 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
896
897 num_blocks -= num_blocks_at_qpi;
898 }
899
900 return 0;
901 }
902
903 /*
904 * This function is called by unpack_dct_coeffs() to extract the VLCs from
905 * the bitstream. The VLCs encode tokens which are used to unpack DCT
906 * data. This function unpacks all the VLCs for either the Y plane or both
907 * C planes, and is called for DC coefficients or different AC coefficient
908 * levels (since different coefficient types require different VLC tables.
909 *
910 * This function returns a residual eob run. E.g, if a particular token gave
911 * instructions to EOB the next 5 fragments and there were only 2 fragments
912 * left in the current fragment range, 3 would be returned so that it could
913 * be passed into the next call to this same function.
914 */
915 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
916 VLC *table, int coeff_index,
917 int plane,
918 int eob_run)
919 {
920 int i, j = 0;
921 int token;
922 int zero_run = 0;
923 int16_t coeff = 0;
924 int bits_to_get;
925 int blocks_ended;
926 int coeff_i = 0;
927 int num_coeffs = s->num_coded_frags[plane][coeff_index];
928 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
929
930 /* local references to structure members to avoid repeated deferences */
931 int *coded_fragment_list = s->coded_fragment_list[plane];
932 Vp3Fragment *all_fragments = s->all_fragments;
933 VLC_TYPE (*vlc_table)[2] = table->table;
934
935 if (num_coeffs < 0)
936 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
937
938 if (eob_run > num_coeffs) {
939 coeff_i = blocks_ended = num_coeffs;
940 eob_run -= num_coeffs;
941 } else {
942 coeff_i = blocks_ended = eob_run;
943 eob_run = 0;
944 }
945
946 // insert fake EOB token to cover the split between planes or zzi
947 if (blocks_ended)
948 dct_tokens[j++] = blocks_ended << 2;
949
950 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
951 /* decode a VLC into a token */
952 token = get_vlc2(gb, vlc_table, 11, 3);
953 /* use the token to get a zero run, a coefficient, and an eob run */
954 if ((unsigned) token <= 6U) {
955 eob_run = eob_run_base[token];
956 if (eob_run_get_bits[token])
957 eob_run += get_bits(gb, eob_run_get_bits[token]);
958
959 // record only the number of blocks ended in this plane,
960 // any spill will be recorded in the next plane.
961 if (eob_run > num_coeffs - coeff_i) {
962 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
963 blocks_ended += num_coeffs - coeff_i;
964 eob_run -= num_coeffs - coeff_i;
965 coeff_i = num_coeffs;
966 } else {
967 dct_tokens[j++] = TOKEN_EOB(eob_run);
968 blocks_ended += eob_run;
969 coeff_i += eob_run;
970 eob_run = 0;
971 }
972 } else if (token >= 0) {
973 bits_to_get = coeff_get_bits[token];
974 if (bits_to_get)
975 bits_to_get = get_bits(gb, bits_to_get);
976 coeff = coeff_tables[token][bits_to_get];
977
978 zero_run = zero_run_base[token];
979 if (zero_run_get_bits[token])
980 zero_run += get_bits(gb, zero_run_get_bits[token]);
981
982 if (zero_run) {
983 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
984 } else {
985 // Save DC into the fragment structure. DC prediction is
986 // done in raster order, so the actual DC can't be in with
987 // other tokens. We still need the token in dct_tokens[]
988 // however, or else the structure collapses on itself.
989 if (!coeff_index)
990 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
991
992 dct_tokens[j++] = TOKEN_COEFF(coeff);
993 }
994
995 if (coeff_index + zero_run > 64) {
996 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
997 " %d coeffs left\n", zero_run, 64-coeff_index);
998 zero_run = 64 - coeff_index;
999 }
1000
1001 // zero runs code multiple coefficients,
1002 // so don't try to decode coeffs for those higher levels
1003 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
1004 s->num_coded_frags[plane][i]--;
1005 coeff_i++;
1006 } else {
1007 av_log(s->avctx, AV_LOG_ERROR,
1008 "Invalid token %d\n", token);
1009 return -1;
1010 }
1011 }
1012
1013 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1014 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1015
1016 // decrement the number of blocks that have higher coeffecients for each
1017 // EOB run at this level
1018 if (blocks_ended)
1019 for (i = coeff_index+1; i < 64; i++)
1020 s->num_coded_frags[plane][i] -= blocks_ended;
1021
1022 // setup the next buffer
1023 if (plane < 2)
1024 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
1025 else if (coeff_index < 63)
1026 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
1027
1028 return eob_run;
1029 }
1030
1031 static void reverse_dc_prediction(Vp3DecodeContext *s,
1032 int first_fragment,
1033 int fragment_width,
1034 int fragment_height);
1035 /*
1036 * This function unpacks all of the DCT coefficient data from the
1037 * bitstream.
1038 */
1039 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1040 {
1041 int i;
1042 int dc_y_table;
1043 int dc_c_table;
1044 int ac_y_table;
1045 int ac_c_table;
1046 int residual_eob_run = 0;
1047 VLC *y_tables[64];
1048 VLC *c_tables[64];
1049
1050 s->dct_tokens[0][0] = s->dct_tokens_base;
1051
1052 /* fetch the DC table indexes */
1053 dc_y_table = get_bits(gb, 4);
1054 dc_c_table = get_bits(gb, 4);
1055
1056 /* unpack the Y plane DC coefficients */
1057 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1058 0, residual_eob_run);
1059 if (residual_eob_run < 0)
1060 return residual_eob_run;
1061
1062 /* reverse prediction of the Y-plane DC coefficients */
1063 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
1064
1065 /* unpack the C plane DC coefficients */
1066 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1067 1, residual_eob_run);
1068 if (residual_eob_run < 0)
1069 return residual_eob_run;
1070 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1071 2, residual_eob_run);
1072 if (residual_eob_run < 0)
1073 return residual_eob_run;
1074
1075 /* reverse prediction of the C-plane DC coefficients */
1076 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1077 {
1078 reverse_dc_prediction(s, s->fragment_start[1],
1079 s->fragment_width[1], s->fragment_height[1]);
1080 reverse_dc_prediction(s, s->fragment_start[2],
1081 s->fragment_width[1], s->fragment_height[1]);
1082 }
1083
1084 /* fetch the AC table indexes */
1085 ac_y_table = get_bits(gb, 4);
1086 ac_c_table = get_bits(gb, 4);
1087
1088 /* build tables of AC VLC tables */
1089 for (i = 1; i <= 5; i++) {
1090 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1091 c_tables[i] = &s->ac_vlc_1[ac_c_table];
1092 }
1093 for (i = 6; i <= 14; i++) {
1094 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1095 c_tables[i] = &s->ac_vlc_2[ac_c_table];
1096 }
1097 for (i = 15; i <= 27; i++) {
1098 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1099 c_tables[i] = &s->ac_vlc_3[ac_c_table];
1100 }
1101 for (i = 28; i <= 63; i++) {
1102 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1103 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1104 }
1105
1106 /* decode all AC coefficents */
1107 for (i = 1; i <= 63; i++) {
1108 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1109 0, residual_eob_run);
1110 if (residual_eob_run < 0)
1111 return residual_eob_run;
1112
1113 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1114 1, residual_eob_run);
1115 if (residual_eob_run < 0)
1116 return residual_eob_run;
1117 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1118 2, residual_eob_run);
1119 if (residual_eob_run < 0)
1120 return residual_eob_run;
1121 }
1122
1123 return 0;
1124 }
1125
1126 /*
1127 * This function reverses the DC prediction for each coded fragment in
1128 * the frame. Much of this function is adapted directly from the original
1129 * VP3 source code.
1130 */
1131 #define COMPATIBLE_FRAME(x) \
1132 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1133 #define DC_COEFF(u) s->all_fragments[u].dc
1134
1135 static void reverse_dc_prediction(Vp3DecodeContext *s,
1136 int first_fragment,
1137 int fragment_width,
1138 int fragment_height)
1139 {
1140
1141 #define PUL 8
1142 #define PU 4
1143 #define PUR 2
1144 #define PL 1
1145
1146 int x, y;
1147 int i = first_fragment;
1148
1149 int predicted_dc;
1150
1151 /* DC values for the left, up-left, up, and up-right fragments */
1152 int vl, vul, vu, vur;
1153
1154 /* indexes for the left, up-left, up, and up-right fragments */
1155 int l, ul, u, ur;
1156
1157 /*
1158 * The 6 fields mean:
1159 * 0: up-left multiplier
1160 * 1: up multiplier
1161 * 2: up-right multiplier
1162 * 3: left multiplier
1163 */
1164 static const int predictor_transform[16][4] = {
1165 { 0, 0, 0, 0},
1166 { 0, 0, 0,128}, // PL
1167 { 0, 0,128, 0}, // PUR
1168 { 0, 0, 53, 75}, // PUR|PL
1169 { 0,128, 0, 0}, // PU
1170 { 0, 64, 0, 64}, // PU|PL
1171 { 0,128, 0, 0}, // PU|PUR
1172 { 0, 0, 53, 75}, // PU|PUR|PL
1173 {128, 0, 0, 0}, // PUL
1174 { 0, 0, 0,128}, // PUL|PL
1175 { 64, 0, 64, 0}, // PUL|PUR
1176 { 0, 0, 53, 75}, // PUL|PUR|PL
1177 { 0,128, 0, 0}, // PUL|PU
1178 {-104,116, 0,116}, // PUL|PU|PL
1179 { 24, 80, 24, 0}, // PUL|PU|PUR
1180 {-104,116, 0,116} // PUL|PU|PUR|PL
1181 };
1182
1183 /* This table shows which types of blocks can use other blocks for
1184 * prediction. For example, INTRA is the only mode in this table to
1185 * have a frame number of 0. That means INTRA blocks can only predict
1186 * from other INTRA blocks. There are 2 golden frame coding types;
1187 * blocks encoding in these modes can only predict from other blocks
1188 * that were encoded with these 1 of these 2 modes. */
1189 static const unsigned char compatible_frame[9] = {
1190 1, /* MODE_INTER_NO_MV */
1191 0, /* MODE_INTRA */
1192 1, /* MODE_INTER_PLUS_MV */
1193 1, /* MODE_INTER_LAST_MV */
1194 1, /* MODE_INTER_PRIOR_MV */
1195 2, /* MODE_USING_GOLDEN */
1196 2, /* MODE_GOLDEN_MV */
1197 1, /* MODE_INTER_FOUR_MV */
1198 3 /* MODE_COPY */
1199 };
1200 int current_frame_type;
1201
1202 /* there is a last DC predictor for each of the 3 frame types */
1203 short last_dc[3];
1204
1205 int transform = 0;
1206
1207 vul = vu = vur = vl = 0;
1208 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1209
1210 /* for each fragment row... */
1211 for (y = 0; y < fragment_height; y++) {
1212
1213 /* for each fragment in a row... */
1214 for (x = 0; x < fragment_width; x++, i++) {
1215
1216 /* reverse prediction if this block was coded */
1217 if (s->all_fragments[i].coding_method != MODE_COPY) {
1218
1219 current_frame_type =
1220 compatible_frame[s->all_fragments[i].coding_method];
1221
1222 transform= 0;
1223 if(x){
1224 l= i-1;
1225 vl = DC_COEFF(l);
1226 if(COMPATIBLE_FRAME(l))
1227 transform |= PL;
1228 }
1229 if(y){
1230 u= i-fragment_width;
1231 vu = DC_COEFF(u);
1232 if(COMPATIBLE_FRAME(u))
1233 transform |= PU;
1234 if(x){
1235 ul= i-fragment_width-1;
1236 vul = DC_COEFF(ul);
1237 if(COMPATIBLE_FRAME(ul))
1238 transform |= PUL;
1239 }
1240 if(x + 1 < fragment_width){
1241 ur= i-fragment_width+1;
1242 vur = DC_COEFF(ur);
1243 if(COMPATIBLE_FRAME(ur))
1244 transform |= PUR;
1245 }
1246 }
1247
1248 if (transform == 0) {
1249
1250 /* if there were no fragments to predict from, use last
1251 * DC saved */
1252 predicted_dc = last_dc[current_frame_type];
1253 } else {
1254
1255 /* apply the appropriate predictor transform */
1256 predicted_dc =
1257 (predictor_transform[transform][0] * vul) +
1258 (predictor_transform[transform][1] * vu) +
1259 (predictor_transform[transform][2] * vur) +
1260 (predictor_transform[transform][3] * vl);
1261
1262 predicted_dc /= 128;
1263
1264 /* check for outranging on the [ul u l] and
1265 * [ul u ur l] predictors */
1266 if ((transform == 15) || (transform == 13)) {
1267 if (FFABS(predicted_dc - vu) > 128)
1268 predicted_dc = vu;
1269 else if (FFABS(predicted_dc - vl) > 128)
1270 predicted_dc = vl;
1271 else if (FFABS(predicted_dc - vul) > 128)
1272 predicted_dc = vul;
1273 }
1274 }
1275
1276 /* at long last, apply the predictor */
1277 DC_COEFF(i) += predicted_dc;
1278 /* save the DC */
1279 last_dc[current_frame_type] = DC_COEFF(i);
1280 }
1281 }
1282 }
1283 }
1284
1285 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1286 {
1287 int x, y;
1288 int *bounding_values= s->bounding_values_array+127;
1289
1290 int width = s->fragment_width[!!plane];
1291 int height = s->fragment_height[!!plane];
1292 int fragment = s->fragment_start [plane] + ystart * width;
1293 int stride = s->current_frame.linesize[plane];
1294 uint8_t *plane_data = s->current_frame.data [plane];
1295 if (!s->flipped_image) stride = -stride;
1296 plane_data += s->data_offset[plane] + 8*ystart*stride;
1297
1298 for (y = ystart; y < yend; y++) {
1299
1300 for (x = 0; x < width; x++) {
1301 /* This code basically just deblocks on the edges of coded blocks.
1302 * However, it has to be much more complicated because of the
1303 * braindamaged deblock ordering used in VP3/Theora. Order matters
1304 * because some pixels get filtered twice. */
1305 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1306 {
1307 /* do not perform left edge filter for left columns frags */
1308 if (x > 0) {
1309 s->vp3dsp.h_loop_filter(
1310 plane_data + 8*x,
1311 stride, bounding_values);
1312 }
1313
1314 /* do not perform top edge filter for top row fragments */
1315 if (y > 0) {
1316 s->vp3dsp.v_loop_filter(
1317 plane_data + 8*x,
1318 stride, bounding_values);
1319 }
1320
1321 /* do not perform right edge filter for right column
1322 * fragments or if right fragment neighbor is also coded
1323 * in this frame (it will be filtered in next iteration) */
1324 if ((x < width - 1) &&
1325 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1326 s->vp3dsp.h_loop_filter(
1327 plane_data + 8*x + 8,
1328 stride, bounding_values);
1329 }
1330
1331 /* do not perform bottom edge filter for bottom row
1332 * fragments or if bottom fragment neighbor is also coded
1333 * in this frame (it will be filtered in the next row) */
1334 if ((y < height - 1) &&
1335 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1336 s->vp3dsp.v_loop_filter(
1337 plane_data + 8*x + 8*stride,
1338 stride, bounding_values);
1339 }
1340 }
1341
1342 fragment++;
1343 }
1344 plane_data += 8*stride;
1345 }
1346 }
1347
1348 /**
1349 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1350 * for the next block in coding order
1351 */
1352 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1353 int plane, int inter, int16_t block[64])
1354 {
1355 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1356 uint8_t *perm = s->scantable.permutated;
1357 int i = 0;
1358
1359 do {
1360 int token = *s->dct_tokens[plane][i];
1361 switch (token & 3) {
1362 case 0: // EOB
1363 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1364 s->dct_tokens[plane][i]++;
1365 else
1366 *s->dct_tokens[plane][i] = token & ~3;
1367 goto end;
1368 case 1: // zero run
1369 s->dct_tokens[plane][i]++;
1370 i += (token >> 2) & 0x7f;
1371 if (i > 63) {
1372 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1373 return i;
1374 }
1375 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1376 i++;
1377 break;
1378 case 2: // coeff
1379 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1380 s->dct_tokens[plane][i++]++;
1381 break;
1382 default: // shouldn't happen
1383 return i;
1384 }
1385 } while (i < 64);
1386 // return value is expected to be a valid level
1387 i--;
1388 end:
1389 // the actual DC+prediction is in the fragment structure
1390 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1391 return i;
1392 }
1393
1394 /**
1395 * called when all pixels up to row y are complete
1396 */
1397 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1398 {
1399 int h, cy, i;
1400 int offset[AV_NUM_DATA_POINTERS];
1401
1402 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
1403 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1404
1405 // At the end of the frame, report INT_MAX instead of the height of the frame.
1406 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1407 // they don't have to clip their values.
1408 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1409 }
1410
1411 if(s->avctx->draw_horiz_band==NULL)
1412 return;
1413
1414 h= y - s->last_slice_end;
1415 s->last_slice_end= y;
1416 y -= h;
1417
1418 if (!s->flipped_image) {
1419 y = s->avctx->height - y - h;
1420 }
1421
1422 cy = y >> s->chroma_y_shift;
1423 offset[0] = s->current_frame.linesize[0]*y;
1424 offset[1] = s->current_frame.linesize[1]*cy;
1425 offset[2] = s->current_frame.linesize[2]*cy;
1426 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1427 offset[i] = 0;
1428
1429 emms_c();
1430 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1431 }
1432
1433 /**
1434 * Wait for the reference frame of the current fragment.
1435 * The progress value is in luma pixel rows.
1436 */
1437 static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1438 {
1439 AVFrame *ref_frame;
1440 int ref_row;
1441 int border = motion_y&1;
1442
1443 if (fragment->coding_method == MODE_USING_GOLDEN ||
1444 fragment->coding_method == MODE_GOLDEN_MV)
1445 ref_frame = &s->golden_frame;
1446 else
1447 ref_frame = &s->last_frame;
1448
1449 ref_row = y + (motion_y>>1);
1450 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1451
1452 ff_thread_await_progress(ref_frame, ref_row, 0);
1453 }
1454
1455 /*
1456 * Perform the final rendering for a particular slice of data.
1457 * The slice number ranges from 0..(c_superblock_height - 1).
1458 */
1459 static void render_slice(Vp3DecodeContext *s, int slice)
1460 {
1461 int x, y, i, j, fragment;
1462 int16_t *block = s->block;
1463 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1464 int motion_halfpel_index;
1465 uint8_t *motion_source;
1466 int plane, first_pixel;
1467
1468 if (slice >= s->c_superblock_height)
1469 return;
1470
1471 for (plane = 0; plane < 3; plane++) {
1472 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1473 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1474 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1475 int stride = s->current_frame.linesize[plane];
1476 int plane_width = s->width >> (plane && s->chroma_x_shift);
1477 int plane_height = s->height >> (plane && s->chroma_y_shift);
1478 int8_t (*motion_val)[2] = s->motion_val[!!plane];
1479
1480 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1481 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
1482 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1483
1484 int fragment_width = s->fragment_width[!!plane];
1485 int fragment_height = s->fragment_height[!!plane];
1486 int fragment_start = s->fragment_start[plane];
1487 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1488
1489 if (!s->flipped_image) stride = -stride;
1490 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1491 continue;
1492
1493 /* for each superblock row in the slice (both of them)... */
1494 for (; sb_y < slice_height; sb_y++) {
1495
1496 /* for each superblock in a row... */
1497 for (sb_x = 0; sb_x < slice_width; sb_x++) {
1498
1499 /* for each block in a superblock... */
1500 for (j = 0; j < 16; j++) {
1501 x = 4*sb_x + hilbert_offset[j][0];
1502 y = 4*sb_y + hilbert_offset[j][1];
1503 fragment = y*fragment_width + x;
1504
1505 i = fragment_start + fragment;
1506
1507 // bounds check
1508 if (x >= fragment_width || y >= fragment_height)
1509 continue;
1510
1511 first_pixel = 8*y*stride + 8*x;
1512
1513 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1514 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1515
1516 /* transform if this block was coded */
1517 if (s->all_fragments[i].coding_method != MODE_COPY) {
1518 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1519 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1520 motion_source= golden_plane;
1521 else
1522 motion_source= last_plane;
1523
1524 motion_source += first_pixel;
1525 motion_halfpel_index = 0;
1526
1527 /* sort out the motion vector if this fragment is coded
1528 * using a motion vector method */
1529 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1530 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1531 int src_x, src_y;
1532 motion_x = motion_val[fragment][0];
1533 motion_y = motion_val[fragment][1];
1534
1535 src_x= (motion_x>>1) + 8*x;
1536 src_y= (motion_y>>1) + 8*y;
1537
1538 motion_halfpel_index = motion_x & 0x01;
1539 motion_source += (motion_x >> 1);
1540
1541 motion_halfpel_index |= (motion_y & 0x01) << 1;
1542 motion_source += ((motion_y >> 1) * stride);
1543
1544 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1545 uint8_t *temp= s->edge_emu_buffer;
1546 if(stride<0) temp -= 8*stride;
1547
1548 s->vdsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1549 motion_source= temp;
1550 }
1551 }
1552
1553
1554 /* first, take care of copying a block from either the
1555 * previous or the golden frame */
1556 if (s->all_fragments[i].coding_method != MODE_INTRA) {
1557 /* Note, it is possible to implement all MC cases with
1558 put_no_rnd_pixels_l2 which would look more like the
1559 VP3 source but this would be slower as
1560 put_no_rnd_pixels_tab is better optimzed */
1561 if(motion_halfpel_index != 3){
1562 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1563 output_plane + first_pixel,
1564 motion_source, stride, 8);
1565 }else{
1566 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1567 s->vp3dsp.put_no_rnd_pixels_l2(
1568 output_plane + first_pixel,
1569 motion_source - d,
1570 motion_source + stride + 1 + d,
1571 stride, 8);
1572 }
1573 }
1574
1575 /* invert DCT and place (or add) in final output */
1576
1577 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1578 int index;
1579 index = vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1580 if (index > 63)
1581 continue;
1582 s->vp3dsp.idct_put(
1583 output_plane + first_pixel,
1584 stride,
1585 block);
1586 } else {
1587 int index = vp3_dequant(s, s->all_fragments + i, plane, 1, block);
1588 if (index > 63)
1589 continue;
1590 if (index > 0) {
1591 s->vp3dsp.idct_add(
1592 output_plane + first_pixel,
1593 stride,
1594 block);
1595 } else {
1596 s->vp3dsp.idct_dc_add(output_plane + first_pixel, stride, block);
1597 }
1598 }
1599 } else {
1600
1601 /* copy directly from the previous frame */
1602 s->dsp.put_pixels_tab[1][0](
1603 output_plane + first_pixel,
1604 last_plane + first_pixel,
1605 stride, 8);
1606
1607 }
1608 }
1609 }
1610
1611 // Filter up to the last row in the superblock row
1612 if (!s->skip_loop_filter)
1613 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1614 }
1615 }
1616
1617 /* this looks like a good place for slice dispatch... */
1618 /* algorithm:
1619 * if (slice == s->macroblock_height - 1)
1620 * dispatch (both last slice & 2nd-to-last slice);
1621 * else if (slice > 0)
1622 * dispatch (slice - 1);
1623 */
1624
1625 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1626 }
1627
1628 /// Allocate tables for per-frame data in Vp3DecodeContext
1629 static av_cold int allocate_tables(AVCodecContext *avctx)
1630 {
1631 Vp3DecodeContext *s = avctx->priv_data;
1632 int y_fragment_count, c_fragment_count;
1633
1634 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1635 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1636
1637 s->superblock_coding = av_malloc(s->superblock_count);
1638 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1639 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1640 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1641 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1642 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1643
1644 /* work out the block mapping tables */
1645 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1646 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1647
1648 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1649 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1650 !s->motion_val[0] || !s->motion_val[1]) {
1651 vp3_decode_end(avctx);
1652 return -1;
1653 }
1654
1655 init_block_mapping(s);
1656
1657 return 0;
1658 }
1659
1660 static av_cold int vp3_decode_init(AVCodecContext *avctx)
1661 {
1662 Vp3DecodeContext *s = avctx->priv_data;
1663 int i, inter, plane;
1664 int c_width;
1665 int c_height;
1666 int y_fragment_count, c_fragment_count;
1667
1668 if (avctx->codec_tag == MKTAG('V','P','3','0'))
1669 s->version = 0;
1670 else
1671 s->version = 1;
1672
1673 s->avctx = avctx;
1674 s->width = FFALIGN(avctx->width, 16);
1675 s->height = FFALIGN(avctx->height, 16);
1676 if (avctx->pix_fmt == AV_PIX_FMT_NONE)
1677 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1678 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1679 ff_dsputil_init(&s->dsp, avctx);
1680 ff_videodsp_init(&s->vdsp, 8);
1681 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
1682
1683 ff_init_scantable_permutation(s->dsp.idct_permutation, s->vp3dsp.idct_perm);
1684 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1685
1686 /* initialize to an impossible value which will force a recalculation
1687 * in the first frame decode */
1688 for (i = 0; i < 3; i++)
1689 s->qps[i] = -1;
1690
1691 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift,
1692 &s->chroma_y_shift);
1693
1694 s->y_superblock_width = (s->width + 31) / 32;
1695 s->y_superblock_height = (s->height + 31) / 32;
1696 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1697
1698 /* work out the dimensions for the C planes */
1699 c_width = s->width >> s->chroma_x_shift;
1700 c_height = s->height >> s->chroma_y_shift;
1701 s->c_superblock_width = (c_width + 31) / 32;
1702 s->c_superblock_height = (c_height + 31) / 32;
1703 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1704
1705 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1706 s->u_superblock_start = s->y_superblock_count;
1707 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1708
1709 s->macroblock_width = (s->width + 15) / 16;
1710 s->macroblock_height = (s->height + 15) / 16;
1711 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1712
1713 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1714 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1715 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1716 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1717
1718 /* fragment count covers all 8x8 blocks for all 3 planes */
1719 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1720 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1721 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1722 s->fragment_start[1] = y_fragment_count;
1723 s->fragment_start[2] = y_fragment_count + c_fragment_count;
1724
1725 if (!s->theora_tables)
1726 {
1727 for (i = 0; i < 64; i++) {
1728 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1729 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1730 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1731 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1732 s->base_matrix[2][i] = vp31_inter_dequant[i];
1733 s->filter_limit_values[i] = vp31_filter_limit_values[i];
1734 }
1735
1736 for(inter=0; inter<2; inter++){
1737 for(plane=0; plane<3; plane++){
1738 s->qr_count[inter][plane]= 1;
1739 s->qr_size [inter][plane][0]= 63;
1740 s->qr_base [inter][plane][0]=
1741 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1742 }
1743 }
1744
1745 /* init VLC tables */
1746 for (i = 0; i < 16; i++) {
1747
1748 /* DC histograms */
1749 init_vlc(&s->dc_vlc[i], 11, 32,
1750 &dc_bias[i][0][1], 4, 2,
1751 &dc_bias[i][0][0], 4, 2, 0);
1752
1753 /* group 1 AC histograms */
1754 init_vlc(&s->ac_vlc_1[i], 11, 32,
1755 &ac_bias_0[i][0][1], 4, 2,
1756 &ac_bias_0[i][0][0], 4, 2, 0);
1757
1758 /* group 2 AC histograms */
1759 init_vlc(&s->ac_vlc_2[i], 11, 32,
1760 &ac_bias_1[i][0][1], 4, 2,
1761 &ac_bias_1[i][0][0], 4, 2, 0);
1762
1763 /* group 3 AC histograms */
1764 init_vlc(&s->ac_vlc_3[i], 11, 32,
1765 &ac_bias_2[i][0][1], 4, 2,
1766 &ac_bias_2[i][0][0], 4, 2, 0);
1767
1768 /* group 4 AC histograms */
1769 init_vlc(&s->ac_vlc_4[i], 11, 32,
1770 &ac_bias_3[i][0][1], 4, 2,
1771 &ac_bias_3[i][0][0], 4, 2, 0);
1772 }
1773 } else {
1774
1775 for (i = 0; i < 16; i++) {
1776 /* DC histograms */
1777 if (init_vlc(&s->dc_vlc[i], 11, 32,
1778 &s->huffman_table[i][0][1], 8, 4,
1779 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1780 goto vlc_fail;
1781
1782 /* group 1 AC histograms */
1783 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1784 &s->huffman_table[i+16][0][1], 8, 4,
1785 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1786 goto vlc_fail;
1787
1788 /* group 2 AC histograms */
1789 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1790 &s->huffman_table[i+16*2][0][1], 8, 4,
1791 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1792 goto vlc_fail;
1793
1794 /* group 3 AC histograms */
1795 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1796 &s->huffman_table[i+16*3][0][1], 8, 4,
1797 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1798 goto vlc_fail;
1799
1800 /* group 4 AC histograms */
1801 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1802 &s->huffman_table[i+16*4][0][1], 8, 4,
1803 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1804 goto vlc_fail;
1805 }
1806 }
1807
1808 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1809 &superblock_run_length_vlc_table[0][1], 4, 2,
1810 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1811
1812 init_vlc(&s->fragment_run_length_vlc, 5, 30,
1813 &fragment_run_length_vlc_table[0][1], 4, 2,
1814 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1815
1816 init_vlc(&s->mode_code_vlc, 3, 8,
1817 &mode_code_vlc_table[0][1], 2, 1,
1818 &mode_code_vlc_table[0][0], 2, 1, 0);
1819
1820 init_vlc(&s->motion_vector_vlc, 6, 63,
1821 &motion_vector_vlc_table[0][1], 2, 1,
1822 &motion_vector_vlc_table[0][0], 2, 1, 0);
1823
1824 for (i = 0; i < 3; i++) {
1825 s->current_frame.data[i] = NULL;
1826 s->last_frame.data[i] = NULL;
1827 s->golden_frame.data[i] = NULL;
1828 }
1829
1830 return allocate_tables(avctx);
1831
1832 vlc_fail:
1833 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1834 return -1;
1835 }
1836
1837 /// Release and shuffle frames after decode finishes
1838 static void update_frames(AVCodecContext *avctx)
1839 {
1840 Vp3DecodeContext *s = avctx->priv_data;
1841
1842 /* release the last frame, if it is allocated and if it is not the
1843 * golden frame */
1844 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1845 ff_thread_release_buffer(avctx, &s->last_frame);
1846
1847 /* shuffle frames (last = current) */
1848 s->last_frame= s->current_frame;
1849
1850 if (s->keyframe) {
1851 if (s->golden_frame.data[0])
1852 ff_thread_release_buffer(avctx, &s->golden_frame);
1853 s->golden_frame = s->current_frame;
1854 s->last_frame.type = FF_BUFFER_TYPE_COPY;
1855 }
1856
1857 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1858 }
1859
1860 static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1861 {
1862 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1863 int qps_changed = 0, i, err;
1864
1865 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1866
1867 if (!s1->current_frame.data[0]
1868 ||s->width != s1->width
1869 ||s->height!= s1->height) {
1870 if (s != s1)
1871 copy_fields(s, s1, golden_frame, current_frame);
1872 return -1;
1873 }
1874
1875 if (s != s1) {
1876 // init tables if the first frame hasn't been decoded
1877 if (!s->current_frame.data[0]) {
1878 int y_fragment_count, c_fragment_count;
1879 s->avctx = dst;
1880 err = allocate_tables(dst);
1881 if (err)
1882 return err;
1883 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1884 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1885 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1886 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1887 }
1888
1889 // copy previous frame data
1890 copy_fields(s, s1, golden_frame, dsp);
1891
1892 // copy qscale data if necessary
1893 for (i = 0; i < 3; i++) {
1894 if (s->qps[i] != s1->qps[1]) {
1895 qps_changed = 1;
1896 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1897 }
1898 }
1899
1900 if (s->qps[0] != s1->qps[0])
1901 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
1902
1903 if (qps_changed)
1904 copy_fields(s, s1, qps, superblock_count);
1905 #undef copy_fields
1906 }
1907
1908 update_frames(dst);
1909
1910 return 0;
1911 }
1912
1913 static int vp3_decode_frame(AVCodecContext *avctx,
1914 void *data, int *got_frame,
1915 AVPacket *avpkt)
1916 {
1917 const uint8_t *buf = avpkt->data;
1918 int buf_size = avpkt->size;
1919 Vp3DecodeContext *s = avctx->priv_data;
1920 GetBitContext gb;
1921 int i;
1922
1923 init_get_bits(&gb, buf, buf_size * 8);
1924
1925 if (s->theora && get_bits1(&gb))
1926 {
1927 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1928 return -1;
1929 }
1930
1931 s->keyframe = !get_bits1(&gb);
1932 if (!s->theora)
1933 skip_bits(&gb, 1);
1934 for (i = 0; i < 3; i++)
1935 s->last_qps[i] = s->qps[i];
1936
1937 s->nqps=0;
1938 do{
1939 s->qps[s->nqps++]= get_bits(&gb, 6);
1940 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1941 for (i = s->nqps; i < 3; i++)
1942 s->qps[i] = -1;
1943
1944 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1945 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1946 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1947
1948 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1949 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1950
1951 if (s->qps[0] != s->last_qps[0])
1952 init_loop_filter(s);
1953
1954 for (i = 0; i < s->nqps; i++)
1955 // reinit all dequantizers if the first one changed, because
1956 // the DC of the first quantizer must be used for all matrices
1957 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1958 init_dequantizer(s, i);
1959
1960 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1961 return buf_size;
1962
1963 s->current_frame.reference = 3;
1964 s->current_frame.pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1965 if (ff_thread_get_buffer(avctx, &s->current_frame) < 0) {
1966 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1967 goto error;
1968 }
1969
1970 if (!s->edge_emu_buffer)
1971 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.linesize[0]));
1972
1973 if (s->keyframe) {
1974 if (!s->theora)
1975 {
1976 skip_bits(&gb, 4); /* width code */
1977 skip_bits(&gb, 4); /* height code */
1978 if (s->version)
1979 {
1980 s->version = get_bits(&gb, 5);
1981 if (avctx->frame_number == 0)
1982 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1983 }
1984 }
1985 if (s->version || s->theora)
1986 {
1987 if (get_bits1(&gb))
1988 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1989 skip_bits(&gb, 2); /* reserved? */
1990 }
1991 } else {
1992 if (!s->golden_frame.data[0]) {
1993 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1994
1995 s->golden_frame.reference = 3;
1996 s->golden_frame.pict_type = AV_PICTURE_TYPE_I;
1997 if (ff_thread_get_buffer(avctx, &s->golden_frame) < 0) {
1998 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1999 goto error;
2000 }
2001 s->last_frame = s->golden_frame;
2002 s->last_frame.type = FF_BUFFER_TYPE_COPY;
2003 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
2004 }
2005 }
2006
2007 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
2008 ff_thread_finish_setup(avctx);
2009
2010 if (unpack_superblocks(s, &gb)){
2011 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2012 goto error;
2013 }
2014 if (unpack_modes(s, &gb)){
2015 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2016 goto error;
2017 }
2018 if (unpack_vectors(s, &gb)){
2019 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2020 goto error;
2021 }
2022 if (unpack_block_qpis(s, &gb)){
2023 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2024 goto error;
2025 }
2026 if (unpack_dct_coeffs(s, &gb)){
2027 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2028 goto error;
2029 }
2030
2031 for (i = 0; i < 3; i++) {
2032 int height = s->height >> (i && s->chroma_y_shift);
2033 if (s->flipped_image)
2034 s->data_offset[i] = 0;
2035 else
2036 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
2037 }
2038
2039 s->last_slice_end = 0;
2040 for (i = 0; i < s->c_superblock_height; i++)
2041 render_slice(s, i);
2042
2043 // filter the last row
2044 for (i = 0; i < 3; i++) {
2045 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
2046 apply_loop_filter(s, i, row, row+1);
2047 }
2048 vp3_draw_horiz_band(s, s->avctx->height);
2049
2050 *got_frame = 1;
2051 *(AVFrame*)data= s->current_frame;
2052
2053 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2054 update_frames(avctx);
2055
2056 return buf_size;
2057
2058 error:
2059 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2060
2061 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
2062 avctx->release_buffer(avctx, &s->current_frame);
2063
2064 return -1;
2065 }
2066
2067 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2068 {
2069 Vp3DecodeContext *s = avctx->priv_data;
2070
2071 if (get_bits1(gb)) {
2072 int token;
2073 if (s->entries >= 32) { /* overflow */
2074 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2075 return -1;
2076 }
2077 token = get_bits(gb, 5);
2078 av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2079 s->hti, s->hbits, token, s->entries, s->huff_code_size);
2080 s->huffman_table[s->hti][token][0] = s->hbits;
2081 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2082 s->entries++;
2083 }
2084 else {
2085 if (s->huff_code_size >= 32) {/* overflow */
2086 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2087 return -1;
2088 }
2089 s->huff_code_size++;
2090 s->hbits <<= 1;
2091 if (read_huffman_tree(avctx, gb))
2092 return -1;
2093 s->hbits |= 1;
2094 if (read_huffman_tree(avctx, gb))
2095 return -1;
2096 s->hbits >>= 1;
2097 s->huff_code_size--;
2098 }
2099 return 0;
2100 }
2101
2102 static int vp3_init_thread_copy(AVCodecContext *avctx)
2103 {
2104 Vp3DecodeContext *s = avctx->priv_data;
2105
2106 s->superblock_coding = NULL;
2107 s->all_fragments = NULL;
2108 s->coded_fragment_list[0] = NULL;
2109 s->dct_tokens_base = NULL;
2110 s->superblock_fragments = NULL;
2111 s->macroblock_coding = NULL;
2112 s->motion_val[0] = NULL;
2113 s->motion_val[1] = NULL;
2114 s->edge_emu_buffer = NULL;
2115
2116 return 0;
2117 }
2118
2119 #if CONFIG_THEORA_DECODER
2120 static const enum AVPixelFormat theora_pix_fmts[4] = {
2121 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
2122 };
2123
2124 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2125 {
2126 Vp3DecodeContext *s = avctx->priv_data;
2127 int visible_width, visible_height, colorspace;
2128 int offset_x = 0, offset_y = 0;
2129 AVRational fps, aspect;
2130
2131 s->theora = get_bits_long(gb, 24);
2132 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2133
2134 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2135 /* but previous versions have the image flipped relative to vp3 */
2136 if (s->theora < 0x030200)
2137 {
2138 s->flipped_image = 1;
2139 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2140 }
2141
2142 visible_width = s->width = get_bits(gb, 16) << 4;
2143 visible_height = s->height = get_bits(gb, 16) << 4;
2144
2145 if(av_image_check_size(s->width, s->height, 0, avctx)){
2146 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2147 s->width= s->height= 0;
2148 return -1;
2149 }
2150
2151 if (s->theora >= 0x030200) {
2152 visible_width = get_bits_long(gb, 24);
2153 visible_height = get_bits_long(gb, 24);
2154
2155 offset_x = get_bits(gb, 8); /* offset x */
2156 offset_y = get_bits(gb, 8); /* offset y, from bottom */
2157 }
2158
2159 fps.num = get_bits_long(gb, 32);
2160 fps.den = get_bits_long(gb, 32);
2161 if (fps.num && fps.den) {
2162 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2163 fps.den, fps.num, 1<<30);
2164 }
2165
2166 aspect.num = get_bits_long(gb, 24);
2167 aspect.den = get_bits_long(gb, 24);
2168 if (aspect.num && aspect.den) {
2169 av_reduce(&avctx->sample_aspect_ratio.num,
2170 &avctx->sample_aspect_ratio.den,
2171 aspect.num, aspect.den, 1<<30);
2172 }
2173
2174 if (s->theora < 0x030200)
2175 skip_bits(gb, 5); /* keyframe frequency force */
2176 colorspace = get_bits(gb, 8);
2177 skip_bits(gb, 24); /* bitrate */
2178
2179 skip_bits(gb, 6); /* quality hint */
2180
2181 if (s->theora >= 0x030200)
2182 {
2183 skip_bits(gb, 5); /* keyframe frequency force */
2184 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2185 skip_bits(gb, 3); /* reserved */
2186 }
2187
2188 // align_get_bits(gb);
2189
2190 if ( visible_width <= s->width && visible_width > s->width-16
2191 && visible_height <= s->height && visible_height > s->height-16
2192 && !offset_x && (offset_y == s->height - visible_height))
2193 avcodec_set_dimensions(avctx, visible_width, visible_height);
2194 else
2195 avcodec_set_dimensions(avctx, s->width, s->height);
2196
2197 if (colorspace == 1) {
2198 avctx->color_primaries = AVCOL_PRI_BT470M;
2199 } else if (colorspace == 2) {
2200 avctx->color_primaries = AVCOL_PRI_BT470BG;
2201 }
2202 if (colorspace == 1 || colorspace == 2) {
2203 avctx->colorspace = AVCOL_SPC_BT470BG;
2204 avctx->color_trc = AVCOL_TRC_BT709;
2205 }
2206
2207 return 0;
2208 }
2209
2210 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2211 {
2212 Vp3DecodeContext *s = avctx->priv_data;
2213 int i, n, matrices, inter, plane;
2214
2215 if (s->theora >= 0x030200) {
2216 n = get_bits(gb, 3);
2217 /* loop filter limit values table */
2218 if (n)
2219 for (i = 0; i < 64; i++)
2220 s->filter_limit_values[i] = get_bits(gb, n);
2221 }
2222
2223 if (s->theora >= 0x030200)
2224 n = get_bits(gb, 4) + 1;
2225 else
2226 n = 16;
2227 /* quality threshold table */
2228 for (i = 0; i < 64; i++)
2229 s->coded_ac_scale_factor[i] = get_bits(gb, n);
2230
2231 if (s->theora >= 0x030200)
2232 n = get_bits(gb, 4) + 1;
2233 else
2234 n = 16;
2235 /* dc scale factor table */
2236 for (i = 0; i < 64; i++)
2237 s->coded_dc_scale_factor[i] = get_bits(gb, n);
2238
2239 if (s->theora >= 0x030200)
2240 matrices = get_bits(gb, 9) + 1;
2241 else
2242 matrices = 3;
2243
2244 if(matrices > 384){
2245 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2246 return -1;
2247 }
2248
2249 for(n=0; n<matrices; n++){
2250 for (i = 0; i < 64; i++)
2251 s->base_matrix[n][i]= get_bits(gb, 8);
2252 }
2253
2254 for (inter = 0; inter <= 1; inter++) {
2255 for (plane = 0; plane <= 2; plane++) {
2256 int newqr= 1;
2257 if (inter || plane > 0)
2258 newqr = get_bits1(gb);
2259 if (!newqr) {
2260 int qtj, plj;
2261 if(inter && get_bits1(gb)){
2262 qtj = 0;
2263 plj = plane;
2264 }else{
2265 qtj= (3*inter + plane - 1) / 3;
2266 plj= (plane + 2) % 3;
2267 }
2268 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2269 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2270 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2271 } else {
2272 int qri= 0;
2273 int qi = 0;
2274
2275 for(;;){
2276 i= get_bits(gb, av_log2(matrices-1)+1);
2277 if(i>= matrices){
2278 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2279 return -1;
2280 }
2281 s->qr_base[inter][plane][qri]= i;
2282 if(qi >= 63)
2283 break;
2284 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2285 s->qr_size[inter][plane][qri++]= i;
2286 qi += i;
2287 }
2288
2289 if (qi > 63) {
2290 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2291 return -1;
2292 }
2293 s->qr_count[inter][plane]= qri;
2294 }
2295 }
2296 }
2297
2298 /* Huffman tables */
2299 for (s->hti = 0; s->hti < 80; s->hti++) {
2300 s->entries = 0;
2301 s->huff_code_size = 1;
2302 if (!get_bits1(gb)) {
2303 s->hbits = 0;
2304 if(read_huffman_tree(avctx, gb))
2305 return -1;
2306 s->hbits = 1;
2307 if(read_huffman_tree(avctx, gb))
2308 return -1;
2309 }
2310 }
2311
2312 s->theora_tables = 1;
2313
2314 return 0;
2315 }
2316
2317 static av_cold int theora_decode_init(AVCodecContext *avctx)
2318 {
2319 Vp3DecodeContext *s = avctx->priv_data;
2320 GetBitContext gb;
2321 int ptype;
2322 uint8_t *header_start[3];
2323 int header_len[3];
2324 int i;
2325
2326 s->theora = 1;
2327
2328 if (!avctx->extradata_size)
2329 {
2330 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2331 return -1;
2332 }
2333
2334 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2335 42, header_start, header_len) < 0) {
2336 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2337 return -1;
2338 }
2339
2340 for(i=0;i<3;i++) {
2341 if (header_len[i] <= 0)
2342 continue;
2343 init_get_bits(&gb, header_start[i], header_len[i] * 8);
2344
2345 ptype = get_bits(&gb, 8);
2346
2347 if (!(ptype & 0x80))
2348 {
2349 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2350 // return -1;
2351 }
2352
2353 // FIXME: Check for this as well.
2354 skip_bits_long(&gb, 6*8); /* "theora" */
2355
2356 switch(ptype)
2357 {
2358 case 0x80:
2359 theora_decode_header(avctx, &gb);
2360 break;
2361 case 0x81:
2362 // FIXME: is this needed? it breaks sometimes
2363 // theora_decode_comments(avctx, gb);
2364 break;
2365 case 0x82:
2366 if (theora_decode_tables(avctx, &gb))
2367 return -1;
2368 break;
2369 default:
2370 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2371 break;
2372 }
2373 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2374 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2375 if (s->theora < 0x030200)
2376 break;
2377 }
2378
2379 return vp3_decode_init(avctx);
2380 }
2381
2382 AVCodec ff_theora_decoder = {
2383 .name = "theora",
2384 .type = AVMEDIA_TYPE_VIDEO,
2385 .id = AV_CODEC_ID_THEORA,
2386 .priv_data_size = sizeof(Vp3DecodeContext),
2387 .init = theora_decode_init,
2388 .close = vp3_decode_end,
2389 .decode = vp3_decode_frame,
2390 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2391 CODEC_CAP_FRAME_THREADS,
2392 .flush = vp3_decode_flush,
2393 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2394 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2395 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
2396 };
2397 #endif
2398
2399 AVCodec ff_vp3_decoder = {
2400 .name = "vp3",
2401 .type = AVMEDIA_TYPE_VIDEO,
2402 .id = AV_CODEC_ID_VP3,
2403 .priv_data_size = sizeof(Vp3DecodeContext),
2404 .init = vp3_decode_init,
2405 .close = vp3_decode_end,
2406 .decode = vp3_decode_frame,
2407 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2408 CODEC_CAP_FRAME_THREADS,
2409 .flush = vp3_decode_flush,
2410 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2411 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
2412 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
2413 };
A couple of patches to make building with MPlayer easier