Do loop filter per-row rather than per-frame
[libav.git] / libavcodec / vp3.c
CommitLineData
d86053a4 1/*
67335dbc 2 * Copyright (C) 2003-2004 the ffmpeg project
d86053a4 3 *
b78e7197
DB
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
d86053a4
MM
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
b78e7197 9 * version 2.1 of the License, or (at your option) any later version.
d86053a4 10 *
b78e7197 11 * FFmpeg is distributed in the hope that it will be useful,
d86053a4
MM
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
b78e7197 17 * License along with FFmpeg; if not, write to the Free Software
5509bffa 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
d86053a4
MM
19 */
20
21/**
bad5537e 22 * @file libavcodec/vp3.c
d86053a4 23 * On2 VP3 Video Decoder
0ad72bdd
MM
24 *
25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26 * For more information about the VP3 coding process, visit:
9db5bdfa 27 * http://wiki.multimedia.cx/index.php?title=On2_VP3
0ad72bdd
MM
28 *
29 * Theora decoder by Alex Beregszaszi
d86053a4
MM
30 */
31
32#include <stdio.h>
33#include <stdlib.h>
34#include <string.h>
d86053a4 35
d86053a4
MM
36#include "avcodec.h"
37#include "dsputil.h"
9106a698 38#include "get_bits.h"
d86053a4
MM
39
40#include "vp3data.h"
da91ed59 41#include "xiph.h"
d86053a4
MM
42
43#define FRAGMENT_PIXELS 8
44
0efbd068
MM
45static av_cold int vp3_decode_end(AVCodecContext *avctx);
46
7beddb12
MN
47typedef struct Coeff {
48 struct Coeff *next;
49 DCTELEM coeff;
50 uint8_t index;
51} Coeff;
52
53//FIXME split things out into their own arrays
d86053a4 54typedef struct Vp3Fragment {
7beddb12 55 Coeff *next_coeff;
d86053a4
MM
56 /* address of first pixel taking into account which plane the fragment
57 * lives on as well as the plane stride */
58 int first_pixel;
59 /* this is the macroblock that the fragment belongs to */
288774bb
MN
60 uint16_t macroblock;
61 uint8_t coding_method;
288774bb
MN
62 int8_t motion_x;
63 int8_t motion_y;
f2264fa5 64 uint8_t qpi;
d86053a4
MM
65} Vp3Fragment;
66
67#define SB_NOT_CODED 0
68#define SB_PARTIALLY_CODED 1
69#define SB_FULLY_CODED 2
70
71#define MODE_INTER_NO_MV 0
72#define MODE_INTRA 1
73#define MODE_INTER_PLUS_MV 2
74#define MODE_INTER_LAST_MV 3
75#define MODE_INTER_PRIOR_LAST 4
76#define MODE_USING_GOLDEN 5
77#define MODE_GOLDEN_MV 6
78#define MODE_INTER_FOURMV 7
79#define CODING_MODE_COUNT 8
80
81/* special internal mode */
82#define MODE_COPY 8
83
84/* There are 6 preset schemes, plus a free-form scheme */
e8e47435 85static const int ModeAlphabet[6][CODING_MODE_COUNT] =
d86053a4 86{
d86053a4 87 /* scheme 1: Last motion vector dominates */
115329f1 88 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
d86053a4 89 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
115329f1 90 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
91 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
92
93 /* scheme 2 */
115329f1 94 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
d86053a4 95 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
115329f1 96 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
97 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
98
99 /* scheme 3 */
115329f1 100 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
d86053a4 101 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
115329f1 102 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
103 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
104
105 /* scheme 4 */
115329f1 106 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
d86053a4 107 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
115329f1 108 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
109 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
110
111 /* scheme 5: No motion vector dominates */
115329f1 112 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
d86053a4 113 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
115329f1 114 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
115 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
116
117 /* scheme 6 */
115329f1 118 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
d86053a4 119 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
115329f1 120 MODE_INTER_PLUS_MV, MODE_INTRA,
d86053a4
MM
121 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
122
123};
124
125#define MIN_DEQUANT_VAL 2
126
127typedef struct Vp3DecodeContext {
128 AVCodecContext *avctx;
f44ee2c3 129 int theora, theora_tables;
3c3f113e 130 int version;
d86053a4 131 int width, height;
d86053a4
MM
132 AVFrame golden_frame;
133 AVFrame last_frame;
134 AVFrame current_frame;
135 int keyframe;
136 DSPContext dsp;
9a7ad925 137 int flipped_image;
d86053a4 138
f2264fa5
DC
139 int qps[3];
140 int nqps;
141 int last_qps[3];
d86053a4
MM
142
143 int superblock_count;
892fc83e
MM
144 int y_superblock_width;
145 int y_superblock_height;
146 int c_superblock_width;
147 int c_superblock_height;
d86053a4
MM
148 int u_superblock_start;
149 int v_superblock_start;
150 unsigned char *superblock_coding;
151
152 int macroblock_count;
153 int macroblock_width;
154 int macroblock_height;
155
156 int fragment_count;
157 int fragment_width;
158 int fragment_height;
159
160 Vp3Fragment *all_fragments;
36e16253 161 uint8_t *coeff_counts;
7beddb12
MN
162 Coeff *coeffs;
163 Coeff *next_coeff;
1abbf64e 164 int fragment_start[3];
115329f1 165
36af0c95 166 ScanTable scantable;
115329f1 167
f44ee2c3
AB
168 /* tables */
169 uint16_t coded_dc_scale_factor[64];
67335dbc 170 uint32_t coded_ac_scale_factor[64];
ae1dd8e1
MN
171 uint8_t base_matrix[384][64];
172 uint8_t qr_count[2][3];
173 uint8_t qr_size [2][3][64];
174 uint16_t qr_base[2][3][64];
d86053a4 175
f4433de9 176 /* this is a list of indexes into the all_fragments array indicating
d86053a4
MM
177 * which of the fragments are coded */
178 int *coded_fragment_list;
179 int coded_fragment_list_index;
5e534865 180 int pixel_addresses_initialized;
d86053a4 181
098523eb
MM
182 /* track which fragments have already been decoded; called 'fast'
183 * because this data structure avoids having to iterate through every
184 * fragment in coded_fragment_list; once a fragment has been fully
185 * decoded, it is removed from this list */
186 int *fast_fragment_list;
187 int fragment_list_y_head;
188 int fragment_list_c_head;
189
d86053a4
MM
190 VLC dc_vlc[16];
191 VLC ac_vlc_1[16];
192 VLC ac_vlc_2[16];
193 VLC ac_vlc_3[16];
194 VLC ac_vlc_4[16];
195
0ad72bdd
MM
196 VLC superblock_run_length_vlc;
197 VLC fragment_run_length_vlc;
198 VLC mode_code_vlc;
199 VLC motion_vector_vlc;
200
38acbc3c
MM
201 /* these arrays need to be on 16-byte boundaries since SSE2 operations
202 * index into them */
c6727809 203 DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
d86053a4
MM
204
205 /* This table contains superblock_count * 16 entries. Each set of 16
f4433de9 206 * numbers corresponds to the fragment indexes 0..15 of the superblock.
d86053a4
MM
207 * An entry will be -1 to indicate that no entry corresponds to that
208 * index. */
209 int *superblock_fragments;
210
211 /* This table contains superblock_count * 4 entries. Each set of 4
f4433de9 212 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
d86053a4
MM
213 * An entry will be -1 to indicate that no entry corresponds to that
214 * index. */
215 int *superblock_macroblocks;
216
217 /* This table contains macroblock_count * 6 entries. Each set of 6
f4433de9 218 * numbers corresponds to the fragment indexes 0..5 which comprise
d86053a4
MM
219 * the macroblock (4 Y fragments and 2 C fragments). */
220 int *macroblock_fragments;
115329f1 221 /* This is an array that indicates how a particular macroblock
74c0ac12 222 * is coded. */
96a7e73b 223 unsigned char *macroblock_coding;
d86053a4 224
04331882
MM
225 int first_coded_y_fragment;
226 int first_coded_c_fragment;
227 int last_coded_y_fragment;
228 int last_coded_c_fragment;
229
a2f11b3c 230 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
191e8ca7 231 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
39922395 232
f44b08a5
MM
233 /* Huffman decode */
234 int hti;
235 unsigned int hbits;
236 int entries;
237 int huff_code_size;
238 uint16_t huffman_table[80][32][2];
239
51ace577 240 uint8_t filter_limit_values[64];
c6727809 241 DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
d86053a4
MM
242} Vp3DecodeContext;
243
244/************************************************************************
245 * VP3 specific functions
246 ************************************************************************/
247
248/*
249 * This function sets up all of the various blocks mappings:
250 * superblocks <-> fragments, macroblocks <-> fragments,
251 * superblocks <-> macroblocks
892fc83e
MM
252 *
253 * Returns 0 is successful; returns 1 if *anything* went wrong.
d86053a4 254 */
115329f1 255static int init_block_mapping(Vp3DecodeContext *s)
d86053a4
MM
256{
257 int i, j;
d86053a4
MM
258 signed int hilbert_walk_mb[4];
259
260 int current_fragment = 0;
261 int current_width = 0;
262 int current_height = 0;
263 int right_edge = 0;
264 int bottom_edge = 0;
265 int superblock_row_inc = 0;
d86053a4
MM
266 int mapping_index = 0;
267
268 int current_macroblock;
269 int c_fragment;
270
bb991087 271 static const signed char travel_width[16] = {
115329f1 272 1, 1, 0, -1,
d86053a4
MM
273 0, 0, 1, 0,
274 1, 0, 1, 0,
275 0, -1, 0, 1
276 };
277
bb991087 278 static const signed char travel_height[16] = {
d86053a4
MM
279 0, 0, 1, 0,
280 1, 1, 0, -1,
281 0, 1, 0, -1,
282 -1, 0, -1, 0
283 };
284
bb991087 285 static const signed char travel_width_mb[4] = {
d86053a4
MM
286 1, 0, 1, 0
287 };
288
bb991087 289 static const signed char travel_height_mb[4] = {
d86053a4
MM
290 0, 1, 0, -1
291 };
292
d86053a4
MM
293 hilbert_walk_mb[0] = 1;
294 hilbert_walk_mb[1] = s->macroblock_width;
295 hilbert_walk_mb[2] = 1;
296 hilbert_walk_mb[3] = -s->macroblock_width;
297
298 /* iterate through each superblock (all planes) and map the fragments */
299 for (i = 0; i < s->superblock_count; i++) {
d86053a4
MM
300 /* time to re-assign the limits? */
301 if (i == 0) {
302
303 /* start of Y superblocks */
304 right_edge = s->fragment_width;
305 bottom_edge = s->fragment_height;
892fc83e 306 current_width = -1;
d86053a4 307 current_height = 0;
115329f1 308 superblock_row_inc = 3 * s->fragment_width -
892fc83e 309 (s->y_superblock_width * 4 - s->fragment_width);
d86053a4
MM
310
311 /* the first operation for this variable is to advance by 1 */
312 current_fragment = -1;
313
314 } else if (i == s->u_superblock_start) {
315
316 /* start of U superblocks */
317 right_edge = s->fragment_width / 2;
318 bottom_edge = s->fragment_height / 2;
892fc83e 319 current_width = -1;
d86053a4 320 current_height = 0;
115329f1 321 superblock_row_inc = 3 * (s->fragment_width / 2) -
892fc83e 322 (s->c_superblock_width * 4 - s->fragment_width / 2);
d86053a4
MM
323
324 /* the first operation for this variable is to advance by 1 */
1abbf64e 325 current_fragment = s->fragment_start[1] - 1;
d86053a4
MM
326
327 } else if (i == s->v_superblock_start) {
328
329 /* start of V superblocks */
330 right_edge = s->fragment_width / 2;
331 bottom_edge = s->fragment_height / 2;
892fc83e 332 current_width = -1;
d86053a4 333 current_height = 0;
115329f1 334 superblock_row_inc = 3 * (s->fragment_width / 2) -
892fc83e 335 (s->c_superblock_width * 4 - s->fragment_width / 2);
d86053a4
MM
336
337 /* the first operation for this variable is to advance by 1 */
1abbf64e 338 current_fragment = s->fragment_start[2] - 1;
d86053a4
MM
339
340 }
341
892fc83e 342 if (current_width >= right_edge - 1) {
d86053a4 343 /* reset width and move to next superblock row */
892fc83e 344 current_width = -1;
d86053a4
MM
345 current_height += 4;
346
347 /* fragment is now at the start of a new superblock row */
348 current_fragment += superblock_row_inc;
349 }
350
351 /* iterate through all 16 fragments in a superblock */
352 for (j = 0; j < 16; j++) {
684d9e36 353 current_fragment += travel_width[j] + right_edge * travel_height[j];
892fc83e 354 current_width += travel_width[j];
d86053a4
MM
355 current_height += travel_height[j];
356
357 /* check if the fragment is in bounds */
892fc83e 358 if ((current_width < right_edge) &&
d86053a4
MM
359 (current_height < bottom_edge)) {
360 s->superblock_fragments[mapping_index] = current_fragment;
d86053a4
MM
361 } else {
362 s->superblock_fragments[mapping_index] = -1;
d86053a4
MM
363 }
364
d86053a4
MM
365 mapping_index++;
366 }
367 }
368
369 /* initialize the superblock <-> macroblock mapping; iterate through
370 * all of the Y plane superblocks to build this mapping */
371 right_edge = s->macroblock_width;
372 bottom_edge = s->macroblock_height;
892fc83e 373 current_width = -1;
d86053a4 374 current_height = 0;
892fc83e 375 superblock_row_inc = s->macroblock_width -
cea96420 376 (s->y_superblock_width * 2 - s->macroblock_width);
d86053a4
MM
377 mapping_index = 0;
378 current_macroblock = -1;
379 for (i = 0; i < s->u_superblock_start; i++) {
380
96a7e73b 381 if (current_width >= right_edge - 1) {
d86053a4 382 /* reset width and move to next superblock row */
96a7e73b 383 current_width = -1;
d86053a4
MM
384 current_height += 2;
385
386 /* macroblock is now at the start of a new superblock row */
387 current_macroblock += superblock_row_inc;
388 }
389
390 /* iterate through each potential macroblock in the superblock */
391 for (j = 0; j < 4; j++) {
392 current_macroblock += hilbert_walk_mb[j];
892fc83e 393 current_width += travel_width_mb[j];
d86053a4
MM
394 current_height += travel_height_mb[j];
395
396 /* check if the macroblock is in bounds */
892fc83e 397 if ((current_width < right_edge) &&
d86053a4
MM
398 (current_height < bottom_edge)) {
399 s->superblock_macroblocks[mapping_index] = current_macroblock;
d86053a4
MM
400 } else {
401 s->superblock_macroblocks[mapping_index] = -1;
d86053a4
MM
402 }
403
d86053a4
MM
404 mapping_index++;
405 }
406 }
407
408 /* initialize the macroblock <-> fragment mapping */
409 current_fragment = 0;
410 current_macroblock = 0;
411 mapping_index = 0;
412 for (i = 0; i < s->fragment_height; i += 2) {
413
414 for (j = 0; j < s->fragment_width; j += 2) {
415
d86053a4
MM
416 s->all_fragments[current_fragment].macroblock = current_macroblock;
417 s->macroblock_fragments[mapping_index++] = current_fragment;
d86053a4
MM
418
419 if (j + 1 < s->fragment_width) {
420 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
421 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
d86053a4
MM
422 } else
423 s->macroblock_fragments[mapping_index++] = -1;
424
425 if (i + 1 < s->fragment_height) {
115329f1 426 s->all_fragments[current_fragment + s->fragment_width].macroblock =
d86053a4 427 current_macroblock;
115329f1 428 s->macroblock_fragments[mapping_index++] =
d86053a4 429 current_fragment + s->fragment_width;
d86053a4
MM
430 } else
431 s->macroblock_fragments[mapping_index++] = -1;
432
433 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
115329f1 434 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
d86053a4 435 current_macroblock;
115329f1 436 s->macroblock_fragments[mapping_index++] =
d86053a4 437 current_fragment + s->fragment_width + 1;
d86053a4
MM
438 } else
439 s->macroblock_fragments[mapping_index++] = -1;
440
441 /* C planes */
1abbf64e 442 c_fragment = s->fragment_start[1] +
d86053a4 443 (i * s->fragment_width / 4) + (j / 2);
892fc83e 444 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
d86053a4 445 s->macroblock_fragments[mapping_index++] = c_fragment;
d86053a4 446
1abbf64e 447 c_fragment = s->fragment_start[2] +
d86053a4 448 (i * s->fragment_width / 4) + (j / 2);
892fc83e 449 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
d86053a4 450 s->macroblock_fragments[mapping_index++] = c_fragment;
d86053a4
MM
451
452 if (j + 2 <= s->fragment_width)
453 current_fragment += 2;
115329f1 454 else
d86053a4
MM
455 current_fragment++;
456 current_macroblock++;
457 }
458
459 current_fragment += s->fragment_width;
460 }
892fc83e
MM
461
462 return 0; /* successful path out */
d86053a4
MM
463}
464
465/*
d86053a4
MM
466 * This function wipes out all of the fragment data.
467 */
468static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
469{
470 int i;
471
472 /* zero out all of the fragment information */
473 s->coded_fragment_list_index = 0;
474 for (i = 0; i < s->fragment_count; i++) {
36e16253 475 s->coeff_counts[i] = 0;
7dc9ed11
MM
476 s->all_fragments[i].motion_x = 127;
477 s->all_fragments[i].motion_y = 127;
478 s->all_fragments[i].next_coeff= NULL;
f2264fa5 479 s->all_fragments[i].qpi = 0;
7beddb12
MN
480 s->coeffs[i].index=
481 s->coeffs[i].coeff=0;
482 s->coeffs[i].next= NULL;
d86053a4
MM
483 }
484}
485
486/*
f44b08a5 487 * This function sets up the dequantization tables used for a particular
d86053a4
MM
488 * frame.
489 */
f2264fa5 490static void init_dequantizer(Vp3DecodeContext *s, int qpi)
d86053a4 491{
f2264fa5
DC
492 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
493 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
36c32bdd 494 int i, plane, inter, qri, bmi, bmj, qistart;
d86053a4 495
ae1dd8e1
MN
496 for(inter=0; inter<2; inter++){
497 for(plane=0; plane<3; plane++){
498 int sum=0;
499 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
500 sum+= s->qr_size[inter][plane][qri];
f2264fa5 501 if(s->qps[qpi] <= sum)
ae1dd8e1
MN
502 break;
503 }
504 qistart= sum - s->qr_size[inter][plane][qri];
505 bmi= s->qr_base[inter][plane][qri ];
506 bmj= s->qr_base[inter][plane][qri+1];
507 for(i=0; i<64; i++){
f2264fa5
DC
508 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
509 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
ae1dd8e1
MN
510 + s->qr_size[inter][plane][qri])
511 / (2*s->qr_size[inter][plane][qri]);
512
a14ab4e4 513 int qmin= 8<<(inter + !i);
ae1dd8e1
MN
514 int qscale= i ? ac_scale_factor : dc_scale_factor;
515
f2264fa5 516 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
ae1dd8e1 517 }
f2264fa5
DC
518 // all DC coefficients use the same quant so as not to interfere with DC prediction
519 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
ae1dd8e1 520 }
d86053a4 521 }
115329f1 522
f2264fa5 523 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
d86053a4
MM
524}
525
526/*
f44b08a5
MM
527 * This function initializes the loop filter boundary limits if the frame's
528 * quality index is different from the previous frame's.
7fa5f999
RD
529 *
530 * The filter_limit_values may not be larger than 127.
f44b08a5
MM
531 */
532static void init_loop_filter(Vp3DecodeContext *s)
533{
534 int *bounding_values= s->bounding_values_array+127;
535 int filter_limit;
536 int x;
7fa5f999 537 int value;
f44b08a5 538
f2264fa5 539 filter_limit = s->filter_limit_values[s->qps[0]];
f44b08a5
MM
540
541 /* set up the bounding values */
542 memset(s->bounding_values_array, 0, 256 * sizeof(int));
543 for (x = 0; x < filter_limit; x++) {
f44b08a5
MM
544 bounding_values[-x] = -x;
545 bounding_values[x] = x;
f44b08a5 546 }
7fa5f999
RD
547 for (x = value = filter_limit; x < 128 && value; x++, value--) {
548 bounding_values[ x] = value;
549 bounding_values[-x] = -value;
550 }
551 if (value)
552 bounding_values[128] = value;
357f45d9 553 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
f44b08a5
MM
554}
555
556/*
115329f1 557 * This function unpacks all of the superblock/macroblock/fragment coding
d86053a4
MM
558 * information from the bitstream.
559 */
892fc83e 560static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
561{
562 int bit = 0;
563 int current_superblock = 0;
564 int current_run = 0;
565 int decode_fully_flags = 0;
566 int decode_partial_blocks = 0;
22493ab9 567 int first_c_fragment_seen;
d86053a4
MM
568
569 int i, j;
570 int current_fragment;
571
d86053a4 572 if (s->keyframe) {
d86053a4
MM
573 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
574
575 } else {
576
577 /* unpack the list of partially-coded superblocks */
5fc32c27 578 bit = get_bits1(gb);
115329f1 579 /* toggle the bit because as soon as the first run length is
d86053a4
MM
580 * fetched the bit will be toggled again */
581 bit ^= 1;
582 while (current_superblock < s->superblock_count) {
b5da3635 583 if (current_run-- == 0) {
d86053a4 584 bit ^= 1;
115329f1 585 current_run = get_vlc2(gb,
b5da3635
MN
586 s->superblock_run_length_vlc.table, 6, 2);
587 if (current_run == 33)
d8278bab 588 current_run += get_bits(gb, 12);
d86053a4
MM
589
590 /* if any of the superblocks are not partially coded, flag
591 * a boolean to decode the list of fully-coded superblocks */
642d7e84 592 if (bit == 0) {
d86053a4 593 decode_fully_flags = 1;
642d7e84 594 } else {
d86053a4 595
642d7e84
MM
596 /* make a note of the fact that there are partially coded
597 * superblocks */
598 decode_partial_blocks = 1;
599 }
d86053a4 600 }
b5da3635 601 s->superblock_coding[current_superblock++] = bit;
d86053a4
MM
602 }
603
604 /* unpack the list of fully coded superblocks if any of the blocks were
605 * not marked as partially coded in the previous step */
606 if (decode_fully_flags) {
607
608 current_superblock = 0;
609 current_run = 0;
5fc32c27 610 bit = get_bits1(gb);
115329f1 611 /* toggle the bit because as soon as the first run length is
d86053a4
MM
612 * fetched the bit will be toggled again */
613 bit ^= 1;
614 while (current_superblock < s->superblock_count) {
615
616 /* skip any superblocks already marked as partially coded */
617 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
618
b5da3635 619 if (current_run-- == 0) {
d86053a4 620 bit ^= 1;
115329f1 621 current_run = get_vlc2(gb,
b5da3635
MN
622 s->superblock_run_length_vlc.table, 6, 2);
623 if (current_run == 33)
d8278bab 624 current_run += get_bits(gb, 12);
d86053a4 625 }
b5da3635 626 s->superblock_coding[current_superblock] = 2*bit;
d86053a4
MM
627 }
628 current_superblock++;
629 }
630 }
631
632 /* if there were partial blocks, initialize bitstream for
633 * unpacking fragment codings */
634 if (decode_partial_blocks) {
635
636 current_run = 0;
5fc32c27 637 bit = get_bits1(gb);
115329f1 638 /* toggle the bit because as soon as the first run length is
d86053a4
MM
639 * fetched the bit will be toggled again */
640 bit ^= 1;
641 }
642 }
643
644 /* figure out which fragments are coded; iterate through each
645 * superblock (all planes) */
646 s->coded_fragment_list_index = 0;
7beddb12 647 s->next_coeff= s->coeffs + s->fragment_count;
04331882
MM
648 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
649 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
22493ab9 650 first_c_fragment_seen = 0;
96a7e73b 651 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
d86053a4
MM
652 for (i = 0; i < s->superblock_count; i++) {
653
654 /* iterate through all 16 fragments in a superblock */
655 for (j = 0; j < 16; j++) {
656
657 /* if the fragment is in bounds, check its coding status */
658 current_fragment = s->superblock_fragments[i * 16 + j];
892fc83e 659 if (current_fragment >= s->fragment_count) {
9b879566 660 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
892fc83e
MM
661 current_fragment, s->fragment_count);
662 return 1;
663 }
d86053a4
MM
664 if (current_fragment != -1) {
665 if (s->superblock_coding[i] == SB_NOT_CODED) {
666
667 /* copy all the fragments from the prior frame */
115329f1 668 s->all_fragments[current_fragment].coding_method =
d86053a4
MM
669 MODE_COPY;
670
671 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
672
673 /* fragment may or may not be coded; this is the case
674 * that cares about the fragment coding runs */
b5da3635 675 if (current_run-- == 0) {
d86053a4 676 bit ^= 1;
115329f1 677 current_run = get_vlc2(gb,
b5da3635 678 s->fragment_run_length_vlc.table, 5, 2);
d86053a4
MM
679 }
680
681 if (bit) {
115329f1 682 /* default mode; actual mode will be decoded in
22493ab9 683 * the next phase */
115329f1 684 s->all_fragments[current_fragment].coding_method =
d86053a4 685 MODE_INTER_NO_MV;
7beddb12 686 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
115329f1 687 s->coded_fragment_list[s->coded_fragment_list_index] =
d86053a4 688 current_fragment;
1abbf64e 689 if ((current_fragment >= s->fragment_start[1]) &&
22493ab9
MM
690 (s->last_coded_y_fragment == -1) &&
691 (!first_c_fragment_seen)) {
04331882
MM
692 s->first_coded_c_fragment = s->coded_fragment_list_index;
693 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
22493ab9 694 first_c_fragment_seen = 1;
04331882
MM
695 }
696 s->coded_fragment_list_index++;
96a7e73b 697 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
d86053a4
MM
698 } else {
699 /* not coded; copy this fragment from the prior frame */
700 s->all_fragments[current_fragment].coding_method =
701 MODE_COPY;
d86053a4
MM
702 }
703
d86053a4
MM
704 } else {
705
706 /* fragments are fully coded in this superblock; actual
707 * coding will be determined in next step */
115329f1 708 s->all_fragments[current_fragment].coding_method =
d86053a4 709 MODE_INTER_NO_MV;
7beddb12 710 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
115329f1 711 s->coded_fragment_list[s->coded_fragment_list_index] =
d86053a4 712 current_fragment;
1abbf64e 713 if ((current_fragment >= s->fragment_start[1]) &&
22493ab9
MM
714 (s->last_coded_y_fragment == -1) &&
715 (!first_c_fragment_seen)) {
04331882
MM
716 s->first_coded_c_fragment = s->coded_fragment_list_index;
717 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
22493ab9 718 first_c_fragment_seen = 1;
04331882
MM
719 }
720 s->coded_fragment_list_index++;
96a7e73b 721 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
d86053a4
MM
722 }
723 }
724 }
725 }
04331882 726
22493ab9
MM
727 if (!first_c_fragment_seen)
728 /* only Y fragments coded in this frame */
04331882 729 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
115329f1 730 else
642d7e84 731 /* end the list of coded C fragments */
04331882 732 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
22493ab9 733
098523eb
MM
734 for (i = 0; i < s->fragment_count - 1; i++) {
735 s->fast_fragment_list[i] = i + 1;
736 }
737 s->fast_fragment_list[s->fragment_count - 1] = -1;
738
739 if (s->last_coded_y_fragment == -1)
740 s->fragment_list_y_head = -1;
741 else {
742 s->fragment_list_y_head = s->first_coded_y_fragment;
743 s->fast_fragment_list[s->last_coded_y_fragment] = -1;
744 }
745
746 if (s->last_coded_c_fragment == -1)
747 s->fragment_list_c_head = -1;
748 else {
749 s->fragment_list_c_head = s->first_coded_c_fragment;
750 s->fast_fragment_list[s->last_coded_c_fragment] = -1;
751 }
752
892fc83e 753 return 0;
d86053a4
MM
754}
755
756/*
757 * This function unpacks all the coding mode data for individual macroblocks
758 * from the bitstream.
759 */
892fc83e 760static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
761{
762 int i, j, k;
763 int scheme;
764 int current_macroblock;
765 int current_fragment;
766 int coding_mode;
e8e47435 767 int custom_mode_alphabet[CODING_MODE_COUNT];
d86053a4 768
d86053a4 769 if (s->keyframe) {
d86053a4
MM
770 for (i = 0; i < s->fragment_count; i++)
771 s->all_fragments[i].coding_method = MODE_INTRA;
772
773 } else {
774
775 /* fetch the mode coding scheme for this frame */
776 scheme = get_bits(gb, 3);
d86053a4
MM
777
778 /* is it a custom coding scheme? */
779 if (scheme == 0) {
d86053a4 780 for (i = 0; i < 8; i++)
2c823b3c
AC
781 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
782 for (i = 0; i < 8; i++)
e8e47435 783 custom_mode_alphabet[get_bits(gb, 3)] = i;
d86053a4
MM
784 }
785
d86053a4
MM
786 /* iterate through all of the macroblocks that contain 1 or more
787 * coded fragments */
788 for (i = 0; i < s->u_superblock_start; i++) {
789
790 for (j = 0; j < 4; j++) {
791 current_macroblock = s->superblock_macroblocks[i * 4 + j];
792 if ((current_macroblock == -1) ||
96a7e73b 793 (s->macroblock_coding[current_macroblock] == MODE_COPY))
d86053a4 794 continue;
892fc83e 795 if (current_macroblock >= s->macroblock_count) {
9b879566 796 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
892fc83e
MM
797 current_macroblock, s->macroblock_count);
798 return 1;
799 }
d86053a4
MM
800
801 /* mode 7 means get 3 bits for each coding mode */
802 if (scheme == 7)
803 coding_mode = get_bits(gb, 3);
e8e47435
SG
804 else if(scheme == 0)
805 coding_mode = custom_mode_alphabet
806 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
d86053a4 807 else
e8e47435 808 coding_mode = ModeAlphabet[scheme-1]
0ad72bdd 809 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
d86053a4 810
96a7e73b 811 s->macroblock_coding[current_macroblock] = coding_mode;
d86053a4 812 for (k = 0; k < 6; k++) {
115329f1 813 current_fragment =
d86053a4 814 s->macroblock_fragments[current_macroblock * 6 + k];
892fc83e
MM
815 if (current_fragment == -1)
816 continue;
817 if (current_fragment >= s->fragment_count) {
9b879566 818 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
892fc83e
MM
819 current_fragment, s->fragment_count);
820 return 1;
821 }
115329f1 822 if (s->all_fragments[current_fragment].coding_method !=
d86053a4
MM
823 MODE_COPY)
824 s->all_fragments[current_fragment].coding_method =
825 coding_mode;
826 }
d86053a4
MM
827 }
828 }
829 }
892fc83e
MM
830
831 return 0;
44ae98dd
MM
832}
833
834/*
d86053a4
MM
835 * This function unpacks all the motion vectors for the individual
836 * macroblocks from the bitstream.
837 */
892fc83e 838static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 839{
feaf1a73 840 int i, j, k, l;
d86053a4
MM
841 int coding_mode;
842 int motion_x[6];
843 int motion_y[6];
844 int last_motion_x = 0;
845 int last_motion_y = 0;
846 int prior_last_motion_x = 0;
847 int prior_last_motion_y = 0;
848 int current_macroblock;
849 int current_fragment;
850
6599e2a7 851 if (s->keyframe)
6298f49f 852 return 0;
10f38380 853
1ae4518d
DC
854 memset(motion_x, 0, 6 * sizeof(int));
855 memset(motion_y, 0, 6 * sizeof(int));
d86053a4 856
1ae4518d
DC
857 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
858 coding_mode = get_bits1(gb);
d86053a4 859
1ae4518d
DC
860 /* iterate through all of the macroblocks that contain 1 or more
861 * coded fragments */
862 for (i = 0; i < s->u_superblock_start; i++) {
d86053a4 863
1ae4518d
DC
864 for (j = 0; j < 4; j++) {
865 current_macroblock = s->superblock_macroblocks[i * 4 + j];
866 if ((current_macroblock == -1) ||
867 (s->macroblock_coding[current_macroblock] == MODE_COPY))
868 continue;
869 if (current_macroblock >= s->macroblock_count) {
870 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
871 current_macroblock, s->macroblock_count);
872 return 1;
873 }
d86053a4 874
1ae4518d
DC
875 current_fragment = s->macroblock_fragments[current_macroblock * 6];
876 if (current_fragment >= s->fragment_count) {
877 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
878 current_fragment, s->fragment_count);
879 return 1;
880 }
881 switch (s->macroblock_coding[current_macroblock]) {
882
883 case MODE_INTER_PLUS_MV:
884 case MODE_GOLDEN_MV:
885 /* all 6 fragments use the same motion vector */
886 if (coding_mode == 0) {
887 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
888 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
889 } else {
890 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
891 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
892fc83e 892 }
7f9926a4 893
1ae4518d
DC
894 /* vector maintenance, only on MODE_INTER_PLUS_MV */
895 if (s->macroblock_coding[current_macroblock] ==
896 MODE_INTER_PLUS_MV) {
e32e2d56
AJ
897 prior_last_motion_x = last_motion_x;
898 prior_last_motion_y = last_motion_y;
1ae4518d
DC
899 last_motion_x = motion_x[0];
900 last_motion_y = motion_y[0];
901 }
902 break;
903
904 case MODE_INTER_FOURMV:
905 /* vector maintenance */
906 prior_last_motion_x = last_motion_x;
907 prior_last_motion_y = last_motion_y;
908
909 /* fetch 4 vectors from the bitstream, one for each
910 * Y fragment, then average for the C fragment vectors */
911 motion_x[4] = motion_y[4] = 0;
912 for (k = 0; k < 4; k++) {
913 for (l = 0; l < s->coded_fragment_list_index; l++)
914 if (s->coded_fragment_list[l] == s->macroblock_fragments[6*current_macroblock + k])
915 break;
916 if (l < s->coded_fragment_list_index) {
917 if (coding_mode == 0) {
918 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
919 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
feaf1a73 920 } else {
1ae4518d
DC
921 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
922 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
feaf1a73 923 }
1ae4518d
DC
924 last_motion_x = motion_x[k];
925 last_motion_y = motion_y[k];
926 } else {
927 motion_x[k] = 0;
928 motion_y[k] = 0;
d86053a4 929 }
1ae4518d
DC
930 motion_x[4] += motion_x[k];
931 motion_y[4] += motion_y[k];
932 }
d86053a4 933
1ae4518d
DC
934 motion_x[5]=
935 motion_x[4]= RSHIFT(motion_x[4], 2);
936 motion_y[5]=
937 motion_y[4]= RSHIFT(motion_y[4], 2);
938 break;
939
940 case MODE_INTER_LAST_MV:
941 /* all 6 fragments use the last motion vector */
942 motion_x[0] = last_motion_x;
943 motion_y[0] = last_motion_y;
d86053a4 944
1ae4518d
DC
945 /* no vector maintenance (last vector remains the
946 * last vector) */
947 break;
948
949 case MODE_INTER_PRIOR_LAST:
950 /* all 6 fragments use the motion vector prior to the
951 * last motion vector */
952 motion_x[0] = prior_last_motion_x;
953 motion_y[0] = prior_last_motion_y;
d86053a4 954
1ae4518d
DC
955 /* vector maintenance */
956 prior_last_motion_x = last_motion_x;
957 prior_last_motion_y = last_motion_y;
958 last_motion_x = motion_x[0];
959 last_motion_y = motion_y[0];
960 break;
44ae98dd 961
1ae4518d
DC
962 default:
963 /* covers intra, inter without MV, golden without MV */
e6e32bdc
MM
964 motion_x[0] = 0;
965 motion_y[0] = 0;
44ae98dd 966
1ae4518d
DC
967 /* no vector maintenance */
968 break;
969 }
d86053a4 970
1ae4518d
DC
971 /* assign the motion vectors to the correct fragments */
972 for (k = 0; k < 6; k++) {
973 current_fragment =
974 s->macroblock_fragments[current_macroblock * 6 + k];
975 if (current_fragment == -1)
976 continue;
977 if (current_fragment >= s->fragment_count) {
978 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
979 current_fragment, s->fragment_count);
980 return 1;
d86053a4 981 }
e6e32bdc 982 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
da8af938
MM
983 s->all_fragments[current_fragment].motion_x = motion_x[k];
984 s->all_fragments[current_fragment].motion_y = motion_y[k];
e6e32bdc
MM
985 } else {
986 s->all_fragments[current_fragment].motion_x = motion_x[0];
987 s->all_fragments[current_fragment].motion_y = motion_y[0];
988 }
d86053a4
MM
989 }
990 }
1ae4518d 991 }
892fc83e
MM
992
993 return 0;
d86053a4
MM
994}
995
f2264fa5
DC
996static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
997{
998 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
999 int num_blocks = s->coded_fragment_list_index;
1000
1001 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
1002 i = blocks_decoded = num_blocks_at_qpi = 0;
1003
1004 bit = get_bits1(gb);
1005
1006 do {
1007 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
1008 if (run_length == 34)
1009 run_length += get_bits(gb, 12);
1010 blocks_decoded += run_length;
1011
1012 if (!bit)
1013 num_blocks_at_qpi += run_length;
1014
1015 for (j = 0; j < run_length; i++) {
310afddf 1016 if (i >= s->coded_fragment_list_index)
f2264fa5
DC
1017 return -1;
1018
1019 if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
1020 s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
1021 j++;
1022 }
1023 }
1024
1025 if (run_length == 4129)
1026 bit = get_bits1(gb);
1027 else
1028 bit ^= 1;
1029 } while (blocks_decoded < num_blocks);
1030
1031 num_blocks -= num_blocks_at_qpi;
1032 }
1033
1034 return 0;
1035}
1036
115329f1 1037/*
d86053a4
MM
1038 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1039 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1040 * data. This function unpacks all the VLCs for either the Y plane or both
1041 * C planes, and is called for DC coefficients or different AC coefficient
1042 * levels (since different coefficient types require different VLC tables.
1043 *
1044 * This function returns a residual eob run. E.g, if a particular token gave
1045 * instructions to EOB the next 5 fragments and there were only 2 fragments
1046 * left in the current fragment range, 3 would be returned so that it could
1047 * be passed into the next call to this same function.
1048 */
1049static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1050 VLC *table, int coeff_index,
098523eb 1051 int y_plane,
d86053a4
MM
1052 int eob_run)
1053{
1054 int i;
1055 int token;
d3076955
MM
1056 int zero_run = 0;
1057 DCTELEM coeff = 0;
d86053a4 1058 Vp3Fragment *fragment;
d3076955 1059 int bits_to_get;
098523eb
MM
1060 int next_fragment;
1061 int previous_fragment;
1062 int fragment_num;
1063 int *list_head;
d86053a4 1064
ee3d7f58
MM
1065 /* local references to structure members to avoid repeated deferences */
1066 uint8_t *perm= s->scantable.permutated;
1067 int *coded_fragment_list = s->coded_fragment_list;
1068 Vp3Fragment *all_fragments = s->all_fragments;
1069 uint8_t *coeff_counts = s->coeff_counts;
1070 VLC_TYPE (*vlc_table)[2] = table->table;
098523eb 1071 int *fast_fragment_list = s->fast_fragment_list;
ee3d7f58 1072
098523eb
MM
1073 if (y_plane) {
1074 next_fragment = s->fragment_list_y_head;
1075 list_head = &s->fragment_list_y_head;
1076 } else {
1077 next_fragment = s->fragment_list_c_head;
1078 list_head = &s->fragment_list_c_head;
74c0ac12
MM
1079 }
1080
098523eb
MM
1081 i = next_fragment;
1082 previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */
1083 while (i != -1) {
1084 fragment_num = coded_fragment_list[i];
d86053a4 1085
098523eb
MM
1086 if (coeff_counts[fragment_num] > coeff_index) {
1087 previous_fragment = i;
1088 i = fast_fragment_list[i];
d86053a4 1089 continue;
098523eb 1090 }
ee3d7f58 1091 fragment = &all_fragments[fragment_num];
d86053a4
MM
1092
1093 if (!eob_run) {
1094 /* decode a VLC into a token */
ee3d7f58 1095 token = get_vlc2(gb, vlc_table, 5, 3);
d86053a4 1096 /* use the token to get a zero run, a coefficient, and an eob run */
d3076955
MM
1097 if (token <= 6) {
1098 eob_run = eob_run_base[token];
1099 if (eob_run_get_bits[token])
1100 eob_run += get_bits(gb, eob_run_get_bits[token]);
1101 coeff = zero_run = 0;
1102 } else {
1103 bits_to_get = coeff_get_bits[token];
428984b0
MM
1104 if (bits_to_get)
1105 bits_to_get = get_bits(gb, bits_to_get);
1106 coeff = coeff_tables[token][bits_to_get];
d3076955
MM
1107
1108 zero_run = zero_run_base[token];
1109 if (zero_run_get_bits[token])
1110 zero_run += get_bits(gb, zero_run_get_bits[token]);
1111 }
d86053a4
MM
1112 }
1113
1114 if (!eob_run) {
ee3d7f58
MM
1115 coeff_counts[fragment_num] += zero_run;
1116 if (coeff_counts[fragment_num] < 64){
7beddb12 1117 fragment->next_coeff->coeff= coeff;
ee3d7f58 1118 fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
7beddb12
MN
1119 fragment->next_coeff->next= s->next_coeff;
1120 s->next_coeff->next=NULL;
1121 fragment->next_coeff= s->next_coeff++;
1122 }
098523eb
MM
1123 /* previous fragment is now this fragment */
1124 previous_fragment = i;
d86053a4 1125 } else {
ee3d7f58 1126 coeff_counts[fragment_num] |= 128;
d86053a4 1127 eob_run--;
098523eb
MM
1128 /* remove this fragment from the list */
1129 if (previous_fragment != -1)
1130 fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1131 else
1132 *list_head = fast_fragment_list[i];
1133 /* previous fragment remains unchanged */
d86053a4 1134 }
098523eb
MM
1135
1136 i = fast_fragment_list[i];
d86053a4
MM
1137 }
1138
1139 return eob_run;
1140}
1141
138fe832
MM
1142static void reverse_dc_prediction(Vp3DecodeContext *s,
1143 int first_fragment,
1144 int fragment_width,
1145 int fragment_height);
d86053a4
MM
1146/*
1147 * This function unpacks all of the DCT coefficient data from the
1148 * bitstream.
1149 */
892fc83e 1150static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
1151{
1152 int i;
1153 int dc_y_table;
1154 int dc_c_table;
1155 int ac_y_table;
1156 int ac_c_table;
1157 int residual_eob_run = 0;
9d8bb031
MM
1158 VLC *y_tables[64];
1159 VLC *c_tables[64];
d86053a4 1160
f4433de9 1161 /* fetch the DC table indexes */
d86053a4
MM
1162 dc_y_table = get_bits(gb, 4);
1163 dc_c_table = get_bits(gb, 4);
1164
1165 /* unpack the Y plane DC coefficients */
115329f1 1166 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
098523eb 1167 1, residual_eob_run);
d86053a4 1168
138fe832
MM
1169 /* reverse prediction of the Y-plane DC coefficients */
1170 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1171
d86053a4 1172 /* unpack the C plane DC coefficients */
d86053a4 1173 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
098523eb 1174 0, residual_eob_run);
d86053a4 1175
138fe832
MM
1176 /* reverse prediction of the C-plane DC coefficients */
1177 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1178 {
1179 reverse_dc_prediction(s, s->fragment_start[1],
1180 s->fragment_width / 2, s->fragment_height / 2);
1181 reverse_dc_prediction(s, s->fragment_start[2],
1182 s->fragment_width / 2, s->fragment_height / 2);
1183 }
1184
f4433de9 1185 /* fetch the AC table indexes */
d86053a4
MM
1186 ac_y_table = get_bits(gb, 4);
1187 ac_c_table = get_bits(gb, 4);
1188
9d8bb031 1189 /* build tables of AC VLC tables */
d86053a4 1190 for (i = 1; i <= 5; i++) {
9d8bb031
MM
1191 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1192 c_tables[i] = &s->ac_vlc_1[ac_c_table];
d86053a4 1193 }
d86053a4 1194 for (i = 6; i <= 14; i++) {
9d8bb031
MM
1195 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1196 c_tables[i] = &s->ac_vlc_2[ac_c_table];
d86053a4 1197 }
d86053a4 1198 for (i = 15; i <= 27; i++) {
9d8bb031
MM
1199 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1200 c_tables[i] = &s->ac_vlc_3[ac_c_table];
d86053a4 1201 }
d86053a4 1202 for (i = 28; i <= 63; i++) {
9d8bb031
MM
1203 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1204 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1205 }
1206
1207 /* decode all AC coefficents */
1208 for (i = 1; i <= 63; i++) {
1209 if (s->fragment_list_y_head != -1)
1210 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1211 1, residual_eob_run);
d86053a4 1212
9d8bb031
MM
1213 if (s->fragment_list_c_head != -1)
1214 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1215 0, residual_eob_run);
d86053a4 1216 }
892fc83e
MM
1217
1218 return 0;
d86053a4
MM
1219}
1220
1221/*
1222 * This function reverses the DC prediction for each coded fragment in
115329f1 1223 * the frame. Much of this function is adapted directly from the original
d86053a4
MM
1224 * VP3 source code.
1225 */
1226#define COMPATIBLE_FRAME(x) \
1227 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
7beddb12 1228#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
d86053a4
MM
1229
1230static void reverse_dc_prediction(Vp3DecodeContext *s,
1231 int first_fragment,
1232 int fragment_width,
115329f1 1233 int fragment_height)
d86053a4
MM
1234{
1235
1236#define PUL 8
1237#define PU 4
1238#define PUR 2
1239#define PL 1
1240
1241 int x, y;
1242 int i = first_fragment;
1243
59ef342b 1244 int predicted_dc;
d86053a4 1245
d86053a4
MM
1246 /* DC values for the left, up-left, up, and up-right fragments */
1247 int vl, vul, vu, vur;
1248
f4433de9 1249 /* indexes for the left, up-left, up, and up-right fragments */
d86053a4
MM
1250 int l, ul, u, ur;
1251
115329f1 1252 /*
d86053a4
MM
1253 * The 6 fields mean:
1254 * 0: up-left multiplier
1255 * 1: up multiplier
1256 * 2: up-right multiplier
1257 * 3: left multiplier
d86053a4 1258 */
bb991087 1259 static const int predictor_transform[16][4] = {
006ff1ca
MN
1260 { 0, 0, 0, 0},
1261 { 0, 0, 0,128}, // PL
1262 { 0, 0,128, 0}, // PUR
1263 { 0, 0, 53, 75}, // PUR|PL
1264 { 0,128, 0, 0}, // PU
1265 { 0, 64, 0, 64}, // PU|PL
1266 { 0,128, 0, 0}, // PU|PUR
1267 { 0, 0, 53, 75}, // PU|PUR|PL
1268 {128, 0, 0, 0}, // PUL
1269 { 0, 0, 0,128}, // PUL|PL
1270 { 64, 0, 64, 0}, // PUL|PUR
1271 { 0, 0, 53, 75}, // PUL|PUR|PL
1272 { 0,128, 0, 0}, // PUL|PU
1273 {-104,116, 0,116}, // PUL|PU|PL
1274 { 24, 80, 24, 0}, // PUL|PU|PUR
1275 {-104,116, 0,116} // PUL|PU|PUR|PL
d86053a4
MM
1276 };
1277
1278 /* This table shows which types of blocks can use other blocks for
1279 * prediction. For example, INTRA is the only mode in this table to
1280 * have a frame number of 0. That means INTRA blocks can only predict
115329f1 1281 * from other INTRA blocks. There are 2 golden frame coding types;
d86053a4
MM
1282 * blocks encoding in these modes can only predict from other blocks
1283 * that were encoded with these 1 of these 2 modes. */
50ba3fd7 1284 static const unsigned char compatible_frame[9] = {
d86053a4
MM
1285 1, /* MODE_INTER_NO_MV */
1286 0, /* MODE_INTRA */
1287 1, /* MODE_INTER_PLUS_MV */
1288 1, /* MODE_INTER_LAST_MV */
1289 1, /* MODE_INTER_PRIOR_MV */
1290 2, /* MODE_USING_GOLDEN */
1291 2, /* MODE_GOLDEN_MV */
50ba3fd7
JGG
1292 1, /* MODE_INTER_FOUR_MV */
1293 3 /* MODE_COPY */
d86053a4
MM
1294 };
1295 int current_frame_type;
1296
1297 /* there is a last DC predictor for each of the 3 frame types */
1298 short last_dc[3];
1299
1300 int transform = 0;
1301
d86053a4
MM
1302 vul = vu = vur = vl = 0;
1303 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1304
1305 /* for each fragment row... */
1306 for (y = 0; y < fragment_height; y++) {
1307
1308 /* for each fragment in a row... */
1309 for (x = 0; x < fragment_width; x++, i++) {
1310
1311 /* reverse prediction if this block was coded */
1312 if (s->all_fragments[i].coding_method != MODE_COPY) {
1313
115329f1 1314 current_frame_type =
d86053a4 1315 compatible_frame[s->all_fragments[i].coding_method];
d86053a4 1316
f72f8a77
MN
1317 transform= 0;
1318 if(x){
1319 l= i-1;
7beddb12 1320 vl = DC_COEFF(l);
50ba3fd7 1321 if(COMPATIBLE_FRAME(l))
006ff1ca 1322 transform |= PL;
f72f8a77
MN
1323 }
1324 if(y){
1325 u= i-fragment_width;
7beddb12 1326 vu = DC_COEFF(u);
50ba3fd7 1327 if(COMPATIBLE_FRAME(u))
006ff1ca 1328 transform |= PU;
f72f8a77
MN
1329 if(x){
1330 ul= i-fragment_width-1;
1331 vul = DC_COEFF(ul);
50ba3fd7 1332 if(COMPATIBLE_FRAME(ul))
006ff1ca 1333 transform |= PUL;
f72f8a77
MN
1334 }
1335 if(x + 1 < fragment_width){
1336 ur= i-fragment_width+1;
1337 vur = DC_COEFF(ur);
50ba3fd7 1338 if(COMPATIBLE_FRAME(ur))
006ff1ca 1339 transform |= PUR;
f72f8a77 1340 }
d86053a4
MM
1341 }
1342
d86053a4
MM
1343 if (transform == 0) {
1344
1345 /* if there were no fragments to predict from, use last
1346 * DC saved */
7beddb12 1347 predicted_dc = last_dc[current_frame_type];
d86053a4
MM
1348 } else {
1349
1350 /* apply the appropriate predictor transform */
1351 predicted_dc =
1352 (predictor_transform[transform][0] * vul) +
1353 (predictor_transform[transform][1] * vu) +
1354 (predictor_transform[transform][2] * vur) +
1355 (predictor_transform[transform][3] * vl);
1356
684d9e36 1357 predicted_dc /= 128;
d86053a4
MM
1358
1359 /* check for outranging on the [ul u l] and
1360 * [ul u ur l] predictors */
c11cb375 1361 if ((transform == 15) || (transform == 13)) {
c26abfa5 1362 if (FFABS(predicted_dc - vu) > 128)
d86053a4 1363 predicted_dc = vu;
c26abfa5 1364 else if (FFABS(predicted_dc - vl) > 128)
d86053a4 1365 predicted_dc = vl;
c26abfa5 1366 else if (FFABS(predicted_dc - vul) > 128)
d86053a4
MM
1367 predicted_dc = vul;
1368 }
d86053a4
MM
1369 }
1370
7beddb12
MN
1371 /* at long last, apply the predictor */
1372 if(s->coeffs[i].index){
1373 *s->next_coeff= s->coeffs[i];
1374 s->coeffs[i].index=0;
1375 s->coeffs[i].coeff=0;
1376 s->coeffs[i].next= s->next_coeff++;
1377 }
1378 s->coeffs[i].coeff += predicted_dc;
d86053a4 1379 /* save the DC */
7beddb12 1380 last_dc[current_frame_type] = DC_COEFF(i);
36e16253
RD
1381 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1382 s->coeff_counts[i]= 129;
7beddb12
MN
1383// s->all_fragments[i].next_coeff= s->next_coeff;
1384 s->coeffs[i].next= s->next_coeff;
1385 (s->next_coeff++)->next=NULL;
1386 }
d86053a4
MM
1387 }
1388 }
1389 }
1390}
1391
256c0662 1392static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
fe313556 1393{
fe313556
DC
1394 int x, y;
1395 int *bounding_values= s->bounding_values_array+127;
1396
fe313556
DC
1397 int width = s->fragment_width >> !!plane;
1398 int height = s->fragment_height >> !!plane;
256c0662 1399 int fragment = s->fragment_start [plane] + ystart * width;
fe313556
DC
1400 int stride = s->current_frame.linesize[plane];
1401 uint8_t *plane_data = s->current_frame.data [plane];
1402 if (!s->flipped_image) stride = -stride;
1403
256c0662 1404 for (y = ystart; y < yend; y++) {
fe313556
DC
1405
1406 for (x = 0; x < width; x++) {
1407 /* This code basically just deblocks on the edges of coded blocks.
1408 * However, it has to be much more complicated because of the
1409 * braindamaged deblock ordering used in VP3/Theora. Order matters
1410 * because some pixels get filtered twice. */
1411 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1412 {
1413 /* do not perform left edge filter for left columns frags */
1414 if (x > 0) {
1415 s->dsp.vp3_h_loop_filter(
1416 plane_data + s->all_fragments[fragment].first_pixel,
1417 stride, bounding_values);
1418 }
1419
1420 /* do not perform top edge filter for top row fragments */
1421 if (y > 0) {
1422 s->dsp.vp3_v_loop_filter(
1423 plane_data + s->all_fragments[fragment].first_pixel,
1424 stride, bounding_values);
1425 }
1426
1427 /* do not perform right edge filter for right column
1428 * fragments or if right fragment neighbor is also coded
1429 * in this frame (it will be filtered in next iteration) */
1430 if ((x < width - 1) &&
1431 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1432 s->dsp.vp3_h_loop_filter(
1433 plane_data + s->all_fragments[fragment + 1].first_pixel,
1434 stride, bounding_values);
1435 }
1436
1437 /* do not perform bottom edge filter for bottom row
1438 * fragments or if bottom fragment neighbor is also coded
1439 * in this frame (it will be filtered in the next row) */
1440 if ((y < height - 1) &&
1441 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1442 s->dsp.vp3_v_loop_filter(
1443 plane_data + s->all_fragments[fragment + width].first_pixel,
1444 stride, bounding_values);
1445 }
1446 }
1447
1448 fragment++;
1449 }
1450 }
fe313556
DC
1451}
1452
d86053a4 1453/*
dc4b78d9
MM
1454 * Perform the final rendering for a particular slice of data.
1455 * The slice number ranges from 0..(macroblock_height - 1).
1456 */
1457static void render_slice(Vp3DecodeContext *s, int slice)
1458{
1abbf64e 1459 int x;
dc4b78d9 1460 int16_t *dequantizer;
c6727809 1461 DECLARE_ALIGNED_16(DCTELEM, block)[64];
dc4b78d9 1462 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
dc4b78d9
MM
1463 int motion_halfpel_index;
1464 uint8_t *motion_source;
1465 int plane;
dc4b78d9
MM
1466 int current_macroblock_entry = slice * s->macroblock_width * 6;
1467
1468 if (slice >= s->macroblock_height)
1469 return;
1470
1471 for (plane = 0; plane < 3; plane++) {
1abbf64e
MN
1472 uint8_t *output_plane = s->current_frame.data [plane];
1473 uint8_t * last_plane = s-> last_frame.data [plane];
1474 uint8_t *golden_plane = s-> golden_frame.data [plane];
1475 int stride = s->current_frame.linesize[plane];
1476 int plane_width = s->width >> !!plane;
1477 int plane_height = s->height >> !!plane;
1478 int y = slice * FRAGMENT_PIXELS << !plane ;
1479 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1480 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1481
1482 if (!s->flipped_image) stride = -stride;
dc4b78d9 1483
115329f1 1484
c26abfa5 1485 if(FFABS(stride) > 2048)
dc4b78d9
MM
1486 return; //various tables are fixed size
1487
1488 /* for each fragment row in the slice (both of them)... */
1489 for (; y < slice_height; y += 8) {
1490
1491 /* for each fragment in a row... */
1492 for (x = 0; x < plane_width; x += 8, i++) {
1493
1494 if ((i < 0) || (i >= s->fragment_count)) {
1495 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1496 return;
1497 }
1498
1499 /* transform if this block was coded */
1500 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1501 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1502
1503 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1504 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1505 motion_source= golden_plane;
115329f1 1506 else
dc4b78d9
MM
1507 motion_source= last_plane;
1508
1509 motion_source += s->all_fragments[i].first_pixel;
1510 motion_halfpel_index = 0;
1511
1512 /* sort out the motion vector if this fragment is coded
1513 * using a motion vector method */
1514 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1515 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1516 int src_x, src_y;
1517 motion_x = s->all_fragments[i].motion_x;
1518 motion_y = s->all_fragments[i].motion_y;
1519 if(plane){
1520 motion_x= (motion_x>>1) | (motion_x&1);
1521 motion_y= (motion_y>>1) | (motion_y&1);
1522 }
1523
1524 src_x= (motion_x>>1) + x;
1525 src_y= (motion_y>>1) + y;
1526 if ((motion_x == 127) || (motion_y == 127))
1527 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1528
1529 motion_halfpel_index = motion_x & 0x01;
1530 motion_source += (motion_x >> 1);
1531
1532 motion_halfpel_index |= (motion_y & 0x01) << 1;
1533 motion_source += ((motion_y >> 1) * stride);
1534
1535 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1536 uint8_t *temp= s->edge_emu_buffer;
1537 if(stride<0) temp -= 9*stride;
1538 else temp += 9*stride;
1539
1540 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1541 motion_source= temp;
1542 }
1543 }
115329f1 1544
dc4b78d9
MM
1545
1546 /* first, take care of copying a block from either the
1547 * previous or the golden frame */
1548 if (s->all_fragments[i].coding_method != MODE_INTRA) {
115329f1
DB
1549 /* Note, it is possible to implement all MC cases with
1550 put_no_rnd_pixels_l2 which would look more like the
1551 VP3 source but this would be slower as
dc4b78d9
MM
1552 put_no_rnd_pixels_tab is better optimzed */
1553 if(motion_halfpel_index != 3){
1554 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1555 output_plane + s->all_fragments[i].first_pixel,
1556 motion_source, stride, 8);
1557 }else{
1558 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1559 s->dsp.put_no_rnd_pixels_l2[1](
1560 output_plane + s->all_fragments[i].first_pixel,
115329f1
DB
1561 motion_source - d,
1562 motion_source + stride + 1 + d,
dc4b78d9
MM
1563 stride, 8);
1564 }
f2264fa5 1565 dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
dc4b78d9 1566 }else{
f2264fa5 1567 dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
dc4b78d9
MM
1568 }
1569
1570 /* dequantize the DCT coefficients */
dc4b78d9
MM
1571 if(s->avctx->idct_algo==FF_IDCT_VP3){
1572 Coeff *coeff= s->coeffs + i;
5fecfb7d 1573 s->dsp.clear_block(block);
dc4b78d9
MM
1574 while(coeff->next){
1575 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1576 coeff= coeff->next;
1577 }
1578 }else{
1579 Coeff *coeff= s->coeffs + i;
5fecfb7d 1580 s->dsp.clear_block(block);
dc4b78d9
MM
1581 while(coeff->next){
1582 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1583 coeff= coeff->next;
1584 }
1585 }
1586
1587 /* invert DCT and place (or add) in final output */
115329f1 1588
dc4b78d9
MM
1589 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1590 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1591 block[0] += 128<<3;
1592 s->dsp.idct_put(
1593 output_plane + s->all_fragments[i].first_pixel,
1594 stride,
1595 block);
1596 } else {
1597 s->dsp.idct_add(
1598 output_plane + s->all_fragments[i].first_pixel,
1599 stride,
1600 block);
1601 }
dc4b78d9
MM
1602 } else {
1603
1604 /* copy directly from the previous frame */
1605 s->dsp.put_pixels_tab[1][0](
1606 output_plane + s->all_fragments[i].first_pixel,
1607 last_plane + s->all_fragments[i].first_pixel,
1608 stride, 8);
1609
1610 }
1611 }
256c0662
DC
1612 // Filter the previous block row. We can't filter the current row yet
1613 // since it needs pixels from the next row
1614 if (y > 0)
1615 apply_loop_filter(s, plane, (y>>3)-1, (y>>3));
dc4b78d9
MM
1616 }
1617 }
1618
dc4b78d9
MM
1619 /* this looks like a good place for slice dispatch... */
1620 /* algorithm:
dc4b78d9 1621 * if (slice == s->macroblock_height - 1)
f44b08a5
MM
1622 * dispatch (both last slice & 2nd-to-last slice);
1623 * else if (slice > 0)
1624 * dispatch (slice - 1);
dc4b78d9
MM
1625 */
1626
1627 emms_c();
1628}
1629
115329f1 1630/*
d86053a4
MM
1631 * This function computes the first pixel addresses for each fragment.
1632 * This function needs to be invoked after the first frame is allocated
1633 * so that it has access to the plane strides.
1634 */
115329f1 1635static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
d86053a4 1636{
fc961819
DC
1637#define Y_INITIAL(chroma_shift) s->flipped_image ? 1 : s->fragment_height >> chroma_shift
1638#define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
d86053a4
MM
1639
1640 int i, x, y;
fc961819 1641 const int y_inc = s->flipped_image ? 1 : -1;
d86053a4
MM
1642
1643 /* figure out the first pixel addresses for each of the fragments */
1644 /* Y plane */
1645 i = 0;
fc961819 1646 for (y = Y_INITIAL(0); Y_FINISHED(0); y += y_inc) {
d86053a4 1647 for (x = 0; x < s->fragment_width; x++) {
115329f1 1648 s->all_fragments[i++].first_pixel =
d86053a4
MM
1649 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1650 s->golden_frame.linesize[0] +
1651 x * FRAGMENT_PIXELS;
d86053a4
MM
1652 }
1653 }
1654
1655 /* U plane */
1abbf64e 1656 i = s->fragment_start[1];
fc961819 1657 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
d86053a4 1658 for (x = 0; x < s->fragment_width / 2; x++) {
115329f1 1659 s->all_fragments[i++].first_pixel =
d86053a4
MM
1660 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1661 s->golden_frame.linesize[1] +
1662 x * FRAGMENT_PIXELS;
d86053a4
MM
1663 }
1664 }
1665
1666 /* V plane */
1abbf64e 1667 i = s->fragment_start[2];
fc961819 1668 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
9a7ad925 1669 for (x = 0; x < s->fragment_width / 2; x++) {
115329f1 1670 s->all_fragments[i++].first_pixel =
9a7ad925
AB
1671 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1672 s->golden_frame.linesize[2] +
1673 x * FRAGMENT_PIXELS;
9a7ad925
AB
1674 }
1675 }
1676}
1677
d86053a4
MM
1678/*
1679 * This is the ffmpeg/libavcodec API init function.
1680 */
98a6fff9 1681static av_cold int vp3_decode_init(AVCodecContext *avctx)
d86053a4
MM
1682{
1683 Vp3DecodeContext *s = avctx->priv_data;
ae1dd8e1 1684 int i, inter, plane;
892fc83e
MM
1685 int c_width;
1686 int c_height;
1687 int y_superblock_count;
1688 int c_superblock_count;
d86053a4 1689
3c3f113e 1690 if (avctx->codec_tag == MKTAG('V','P','3','0'))
bb270c08 1691 s->version = 0;
3c3f113e 1692 else
bb270c08 1693 s->version = 1;
3c3f113e 1694
d86053a4 1695 s->avctx = avctx;
ef516f73
DC
1696 s->width = FFALIGN(avctx->width, 16);
1697 s->height = FFALIGN(avctx->height, 16);
d86053a4 1698 avctx->pix_fmt = PIX_FMT_YUV420P;
580a7465 1699 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
8b6103da
MN
1700 if(avctx->idct_algo==FF_IDCT_AUTO)
1701 avctx->idct_algo=FF_IDCT_VP3;
d86053a4 1702 dsputil_init(&s->dsp, avctx);
115329f1 1703
36af0c95 1704 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
d86053a4
MM
1705
1706 /* initialize to an impossible value which will force a recalculation
1707 * in the first frame decode */
f2264fa5
DC
1708 for (i = 0; i < 3; i++)
1709 s->qps[i] = -1;
d86053a4 1710
892fc83e
MM
1711 s->y_superblock_width = (s->width + 31) / 32;
1712 s->y_superblock_height = (s->height + 31) / 32;
1713 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1714
1715 /* work out the dimensions for the C planes */
1716 c_width = s->width / 2;
1717 c_height = s->height / 2;
1718 s->c_superblock_width = (c_width + 31) / 32;
1719 s->c_superblock_height = (c_height + 31) / 32;
1720 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1721
1722 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1723 s->u_superblock_start = y_superblock_count;
1724 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
d86053a4
MM
1725 s->superblock_coding = av_malloc(s->superblock_count);
1726
1727 s->macroblock_width = (s->width + 15) / 16;
1728 s->macroblock_height = (s->height + 15) / 16;
1729 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1730
1731 s->fragment_width = s->width / FRAGMENT_PIXELS;
1732 s->fragment_height = s->height / FRAGMENT_PIXELS;
1733
1734 /* fragment count covers all 8x8 blocks for all 3 planes */
1735 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1abbf64e
MN
1736 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1737 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
d86053a4 1738
d86053a4 1739 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
36e16253 1740 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
7beddb12 1741 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
d86053a4 1742 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
098523eb 1743 s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
5e534865 1744 s->pixel_addresses_initialized = 0;
0efbd068 1745 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
098523eb 1746 !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
0efbd068
MM
1747 vp3_decode_end(avctx);
1748 return -1;
1749 }
d86053a4 1750
f44ee2c3
AB
1751 if (!s->theora_tables)
1752 {
2287c100 1753 for (i = 0; i < 64; i++) {
bb270c08 1754 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
bb270c08 1755 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
ae1dd8e1 1756 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
ae1dd8e1 1757 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
ae1dd8e1 1758 s->base_matrix[2][i] = vp31_inter_dequant[i];
bb270c08 1759 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2287c100 1760 }
f44ee2c3 1761
ae1dd8e1
MN
1762 for(inter=0; inter<2; inter++){
1763 for(plane=0; plane<3; plane++){
1764 s->qr_count[inter][plane]= 1;
1765 s->qr_size [inter][plane][0]= 63;
1766 s->qr_base [inter][plane][0]=
1767 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1768 }
1769 }
1770
39922395
MM
1771 /* init VLC tables */
1772 for (i = 0; i < 16; i++) {
1773
1774 /* DC histograms */
1775 init_vlc(&s->dc_vlc[i], 5, 32,
1776 &dc_bias[i][0][1], 4, 2,
1777 &dc_bias[i][0][0], 4, 2, 0);
1778
1779 /* group 1 AC histograms */
1780 init_vlc(&s->ac_vlc_1[i], 5, 32,
1781 &ac_bias_0[i][0][1], 4, 2,
1782 &ac_bias_0[i][0][0], 4, 2, 0);
1783
1784 /* group 2 AC histograms */
1785 init_vlc(&s->ac_vlc_2[i], 5, 32,
1786 &ac_bias_1[i][0][1], 4, 2,
1787 &ac_bias_1[i][0][0], 4, 2, 0);
1788
1789 /* group 3 AC histograms */
1790 init_vlc(&s->ac_vlc_3[i], 5, 32,
1791 &ac_bias_2[i][0][1], 4, 2,
1792 &ac_bias_2[i][0][0], 4, 2, 0);
1793
1794 /* group 4 AC histograms */
1795 init_vlc(&s->ac_vlc_4[i], 5, 32,
1796 &ac_bias_3[i][0][1], 4, 2,
1797 &ac_bias_3[i][0][0], 4, 2, 0);
1798 }
1799 } else {
1800 for (i = 0; i < 16; i++) {
1801
1802 /* DC histograms */
c4b7b8bf 1803 if (init_vlc(&s->dc_vlc[i], 5, 32,
39922395 1804 &s->huffman_table[i][0][1], 4, 2,
c4b7b8bf
RD
1805 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1806 goto vlc_fail;
39922395
MM
1807
1808 /* group 1 AC histograms */
c4b7b8bf 1809 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
39922395 1810 &s->huffman_table[i+16][0][1], 4, 2,
c4b7b8bf
RD
1811 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1812 goto vlc_fail;
39922395
MM
1813
1814 /* group 2 AC histograms */
c4b7b8bf 1815 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
39922395 1816 &s->huffman_table[i+16*2][0][1], 4, 2,
c4b7b8bf
RD
1817 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1818 goto vlc_fail;
39922395
MM
1819
1820 /* group 3 AC histograms */
c4b7b8bf 1821 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
39922395 1822 &s->huffman_table[i+16*3][0][1], 4, 2,
c4b7b8bf
RD
1823 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1824 goto vlc_fail;
39922395
MM
1825
1826 /* group 4 AC histograms */
c4b7b8bf 1827 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
39922395 1828 &s->huffman_table[i+16*4][0][1], 4, 2,
c4b7b8bf
RD
1829 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1830 goto vlc_fail;
39922395 1831 }
d86053a4
MM
1832 }
1833
d8278bab
MM
1834 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1835 &superblock_run_length_vlc_table[0][1], 4, 2,
1836 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1837
dd36b667 1838 init_vlc(&s->fragment_run_length_vlc, 5, 30,
0ad72bdd
MM
1839 &fragment_run_length_vlc_table[0][1], 4, 2,
1840 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1841
1842 init_vlc(&s->mode_code_vlc, 3, 8,
1843 &mode_code_vlc_table[0][1], 2, 1,
1844 &mode_code_vlc_table[0][0], 2, 1, 0);
1845
1846 init_vlc(&s->motion_vector_vlc, 6, 63,
1847 &motion_vector_vlc_table[0][1], 2, 1,
1848 &motion_vector_vlc_table[0][0], 2, 1, 0);
1849
d86053a4
MM
1850 /* work out the block mapping tables */
1851 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1852 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
1853 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
96a7e73b 1854 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
0efbd068
MM
1855 if (!s->superblock_fragments || !s->superblock_macroblocks ||
1856 !s->macroblock_fragments || !s->macroblock_coding) {
1857 vp3_decode_end(avctx);
1858 return -1;
1859 }
d86053a4
MM
1860 init_block_mapping(s);
1861
44ae98dd
MM
1862 for (i = 0; i < 3; i++) {
1863 s->current_frame.data[i] = NULL;
1864 s->last_frame.data[i] = NULL;
1865 s->golden_frame.data[i] = NULL;
61873c4a
MM
1866 }
1867
d86053a4 1868 return 0;
c4b7b8bf
RD
1869
1870vlc_fail:
1871 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1872 return -1;
d86053a4
MM
1873}
1874
1875/*
1876 * This is the ffmpeg/libavcodec API frame decode function.
1877 */
115329f1 1878static int vp3_decode_frame(AVCodecContext *avctx,
d86053a4 1879 void *data, int *data_size,
7a00bbad 1880 AVPacket *avpkt)
d86053a4 1881{
7a00bbad
TB
1882 const uint8_t *buf = avpkt->data;
1883 int buf_size = avpkt->size;
d86053a4
MM
1884 Vp3DecodeContext *s = avctx->priv_data;
1885 GetBitContext gb;
1886 static int counter = 0;
dc4b78d9 1887 int i;
d86053a4 1888
d86053a4 1889 init_get_bits(&gb, buf, buf_size * 8);
115329f1 1890
f44ee2c3
AB
1891 if (s->theora && get_bits1(&gb))
1892 {
bb270c08
DB
1893 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1894 return -1;
f44ee2c3 1895 }
3c3f113e
AB
1896
1897 s->keyframe = !get_bits1(&gb);
1898 if (!s->theora)
bb270c08 1899 skip_bits(&gb, 1);
f2264fa5
DC
1900 for (i = 0; i < 3; i++)
1901 s->last_qps[i] = s->qps[i];
efea8528 1902
f2264fa5 1903 s->nqps=0;
efea8528 1904 do{
f2264fa5
DC
1905 s->qps[s->nqps++]= get_bits(&gb, 6);
1906 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1907 for (i = s->nqps; i < 3; i++)
1908 s->qps[i] = -1;
d86053a4 1909
f8830383 1910 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
bb270c08 1911 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
f2264fa5 1912 s->keyframe?"key":"", counter, s->qps[0]);
d86053a4
MM
1913 counter++;
1914
f2264fa5 1915 if (s->qps[0] != s->last_qps[0])
f44b08a5 1916 init_loop_filter(s);
f2264fa5
DC
1917
1918 for (i = 0; i < s->nqps; i++)
1919 // reinit all dequantizers if the first one changed, because
1920 // the DC of the first quantizer must be used for all matrices
1921 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1922 init_dequantizer(s, i);
642d7e84 1923
068e82ba
DC
1924 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1925 return buf_size;
1926
d86053a4 1927 if (s->keyframe) {
bb270c08
DB
1928 if (!s->theora)
1929 {
1930 skip_bits(&gb, 4); /* width code */
1931 skip_bits(&gb, 4); /* height code */
1932 if (s->version)
1933 {
1934 s->version = get_bits(&gb, 5);
1935 if (counter == 1)
1936 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1937 }
1938 }
1939 if (s->version || s->theora)
1940 {
1941 if (get_bits1(&gb))
1942 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1943 skip_bits(&gb, 2); /* reserved? */
1944 }
3c3f113e 1945
74c0ac12
MM
1946 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1947 if (s->golden_frame.data[0])
1948 avctx->release_buffer(avctx, &s->golden_frame);
8e39d4a7 1949 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
74c0ac12
MM
1950 } else {
1951 if (s->golden_frame.data[0])
1952 avctx->release_buffer(avctx, &s->golden_frame);
1953 if (s->last_frame.data[0])
1954 avctx->release_buffer(avctx, &s->last_frame);
1955 }
d86053a4 1956
8e39d4a7 1957 s->golden_frame.reference = 3;
d86053a4 1958 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
9b879566 1959 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
d86053a4
MM
1960 return -1;
1961 }
1962
d86053a4 1963 /* golden frame is also the current frame */
4501e678 1964 s->current_frame= s->golden_frame;
d86053a4
MM
1965
1966 /* time to figure out pixel addresses? */
5e534865 1967 if (!s->pixel_addresses_initialized)
bb270c08 1968 {
ba31df31 1969 vp3_calculate_pixel_addresses(s);
5e534865 1970 s->pixel_addresses_initialized = 1;
bb270c08 1971 }
d86053a4 1972 } else {
d86053a4 1973 /* allocate a new current frame */
8e39d4a7 1974 s->current_frame.reference = 3;
5e534865 1975 if (!s->pixel_addresses_initialized) {
bc185f72
RD
1976 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1977 return -1;
1978 }
d86053a4 1979 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
9b879566 1980 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
d86053a4
MM
1981 return -1;
1982 }
d86053a4
MM
1983 }
1984
b928ec64
MN
1985 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1986 s->current_frame.qstride= 0;
1987
d86053a4
MM
1988 init_frame(s, &gb);
1989
220a6f40
MN
1990 if (unpack_superblocks(s, &gb)){
1991 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1992 return -1;
1993 }
220a6f40
MN
1994 if (unpack_modes(s, &gb)){
1995 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1996 return -1;
1997 }
220a6f40
MN
1998 if (unpack_vectors(s, &gb)){
1999 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2000 return -1;
2001 }
f2264fa5
DC
2002 if (unpack_block_qpis(s, &gb)){
2003 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2004 return -1;
2005 }
220a6f40
MN
2006 if (unpack_dct_coeffs(s, &gb)){
2007 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
892fc83e
MM
2008 return -1;
2009 }
d86053a4 2010
dc4b78d9
MM
2011 for (i = 0; i < s->macroblock_height; i++)
2012 render_slice(s, i);
d86053a4 2013
256c0662
DC
2014 // filter the last row
2015 for (i = 0; i < 3; i++) {
2016 int row = (s->height >> (3+!!i)) - 1;
2017 apply_loop_filter(s, i, row, row+1);
2018 }
892fc83e 2019
d86053a4
MM
2020 *data_size=sizeof(AVFrame);
2021 *(AVFrame*)data= s->current_frame;
2022
44ae98dd
MM
2023 /* release the last frame, if it is allocated and if it is not the
2024 * golden frame */
2025 if ((s->last_frame.data[0]) &&
2026 (s->last_frame.data[0] != s->golden_frame.data[0]))
2027 avctx->release_buffer(avctx, &s->last_frame);
d86053a4 2028
61873c4a 2029 /* shuffle frames (last = current) */
4501e678 2030 s->last_frame= s->current_frame;
8e39d4a7 2031 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
d86053a4
MM
2032
2033 return buf_size;
2034}
2035
2036/*
2037 * This is the ffmpeg/libavcodec API module cleanup function.
2038 */
98a6fff9 2039static av_cold int vp3_decode_end(AVCodecContext *avctx)
d86053a4
MM
2040{
2041 Vp3DecodeContext *s = avctx->priv_data;
6f4e2b5a 2042 int i;
d86053a4 2043
6f4e2b5a 2044 av_free(s->superblock_coding);
d86053a4 2045 av_free(s->all_fragments);
36e16253 2046 av_free(s->coeff_counts);
a2df5a50 2047 av_free(s->coeffs);
d86053a4 2048 av_free(s->coded_fragment_list);
098523eb 2049 av_free(s->fast_fragment_list);
d86053a4
MM
2050 av_free(s->superblock_fragments);
2051 av_free(s->superblock_macroblocks);
2052 av_free(s->macroblock_fragments);
96a7e73b 2053 av_free(s->macroblock_coding);
115329f1 2054
6f4e2b5a
MR
2055 for (i = 0; i < 16; i++) {
2056 free_vlc(&s->dc_vlc[i]);
2057 free_vlc(&s->ac_vlc_1[i]);
2058 free_vlc(&s->ac_vlc_2[i]);
2059 free_vlc(&s->ac_vlc_3[i]);
2060 free_vlc(&s->ac_vlc_4[i]);
2061 }
2062
2063 free_vlc(&s->superblock_run_length_vlc);
2064 free_vlc(&s->fragment_run_length_vlc);
2065 free_vlc(&s->mode_code_vlc);
2066 free_vlc(&s->motion_vector_vlc);
2067
d86053a4 2068 /* release all frames */
8e39d4a7 2069 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
892fc83e
MM
2070 avctx->release_buffer(avctx, &s->golden_frame);
2071 if (s->last_frame.data[0])
2072 avctx->release_buffer(avctx, &s->last_frame);
2073 /* no need to release the current_frame since it will always be pointing
2074 * to the same frame as either the golden or last frame */
d86053a4
MM
2075
2076 return 0;
2077}
2078
39922395
MM
2079static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2080{
2081 Vp3DecodeContext *s = avctx->priv_data;
2082
5fc32c27 2083 if (get_bits1(gb)) {
39922395
MM
2084 int token;
2085 if (s->entries >= 32) { /* overflow */
2086 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2087 return -1;
2088 }
2089 token = get_bits(gb, 5);
2090 //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);
2091 s->huffman_table[s->hti][token][0] = s->hbits;
2092 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2093 s->entries++;
2094 }
2095 else {
2096 if (s->huff_code_size >= 32) {/* overflow */
2097 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2098 return -1;
2099 }
2100 s->huff_code_size++;
2101 s->hbits <<= 1;
00bbe276
AC
2102 if (read_huffman_tree(avctx, gb))
2103 return -1;
39922395 2104 s->hbits |= 1;
00bbe276
AC
2105 if (read_huffman_tree(avctx, gb))
2106 return -1;
39922395
MM
2107 s->hbits >>= 1;
2108 s->huff_code_size--;
2109 }
2110 return 0;
2111}
2112
b250f9c6 2113#if CONFIG_THEORA_DECODER
e278056f 2114static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
2115{
2116 Vp3DecodeContext *s = avctx->priv_data;
ea3c2d53 2117 int visible_width, visible_height, colorspace;
9a7ad925 2118
e278056f 2119 s->theora = get_bits_long(gb, 24);
356306ac 2120 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
105c3d25 2121
ba7ee4a4 2122 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
9a7ad925 2123 /* but previous versions have the image flipped relative to vp3 */
ba7ee4a4 2124 if (s->theora < 0x030200)
9a7ad925 2125 {
bb270c08 2126 s->flipped_image = 1;
9a7ad925
AB
2127 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2128 }
f44ee2c3 2129
277e3e53
DC
2130 visible_width = s->width = get_bits(gb, 16) << 4;
2131 visible_height = s->height = get_bits(gb, 16) << 4;
115329f1 2132
0ecca7a4 2133 if(avcodec_check_dimensions(avctx, s->width, s->height)){
7146d2c2 2134 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
0ecca7a4
MN
2135 s->width= s->height= 0;
2136 return -1;
2137 }
7146d2c2 2138
277e3e53 2139 if (s->theora >= 0x030200) {
a0ce2d1b
DC
2140 visible_width = get_bits_long(gb, 24);
2141 visible_height = get_bits_long(gb, 24);
c0f716b8 2142
ba4816a0
AJ
2143 skip_bits(gb, 8); /* offset x */
2144 skip_bits(gb, 8); /* offset y */
2145 }
f44ee2c3 2146
e278056f
MN
2147 skip_bits(gb, 32); /* fps numerator */
2148 skip_bits(gb, 32); /* fps denumerator */
2149 skip_bits(gb, 24); /* aspect numerator */
2150 skip_bits(gb, 24); /* aspect denumerator */
115329f1 2151
ba7ee4a4 2152 if (s->theora < 0x030200)
e278056f 2153 skip_bits(gb, 5); /* keyframe frequency force */
ea3c2d53 2154 colorspace = get_bits(gb, 8);
e278056f 2155 skip_bits(gb, 24); /* bitrate */
f44ee2c3 2156
e278056f 2157 skip_bits(gb, 6); /* quality hint */
115329f1 2158
ba7ee4a4 2159 if (s->theora >= 0x030200)
105c3d25 2160 {
e278056f 2161 skip_bits(gb, 5); /* keyframe frequency force */
337f5c6e
DC
2162 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2163 skip_bits(gb, 3); /* reserved */
105c3d25 2164 }
115329f1 2165
e278056f 2166// align_get_bits(gb);
115329f1 2167
c0f716b8
AJ
2168 if ( visible_width <= s->width && visible_width > s->width-16
2169 && visible_height <= s->height && visible_height > s->height-16)
2170 avcodec_set_dimensions(avctx, visible_width, visible_height);
2171 else
2172 avcodec_set_dimensions(avctx, s->width, s->height);
f44ee2c3 2173
ea3c2d53
DC
2174 if (colorspace == 1) {
2175 avctx->color_primaries = AVCOL_PRI_BT470M;
2176 } else if (colorspace == 2) {
2177 avctx->color_primaries = AVCOL_PRI_BT470BG;
2178 }
2179 if (colorspace == 1 || colorspace == 2) {
2180 avctx->colorspace = AVCOL_SPC_BT470BG;
2181 avctx->color_trc = AVCOL_TRC_BT709;
2182 }
2183
f44ee2c3
AB
2184 return 0;
2185}
2186
e278056f 2187static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
2188{
2189 Vp3DecodeContext *s = avctx->priv_data;
ae1dd8e1 2190 int i, n, matrices, inter, plane;
ba7ee4a4
MC
2191
2192 if (s->theora >= 0x030200) {
e278056f 2193 n = get_bits(gb, 3);
9c7154c7 2194 /* loop filter limit values table */
e13cca4b 2195 for (i = 0; i < 64; i++) {
e278056f 2196 s->filter_limit_values[i] = get_bits(gb, n);
e13cca4b
RD
2197 if (s->filter_limit_values[i] > 127) {
2198 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2199 s->filter_limit_values[i] = 127;
2200 }
2201 }
ba7ee4a4 2202 }
115329f1 2203
ba7ee4a4 2204 if (s->theora >= 0x030200)
e278056f 2205 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2206 else
2207 n = 16;
f44ee2c3
AB
2208 /* quality threshold table */
2209 for (i = 0; i < 64; i++)
e278056f 2210 s->coded_ac_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2211
ba7ee4a4 2212 if (s->theora >= 0x030200)
e278056f 2213 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2214 else
2215 n = 16;
f44ee2c3
AB
2216 /* dc scale factor table */
2217 for (i = 0; i < 64; i++)
e278056f 2218 s->coded_dc_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2219
ba7ee4a4 2220 if (s->theora >= 0x030200)
e278056f 2221 matrices = get_bits(gb, 9) + 1;
ba7ee4a4 2222 else
2da2ba03 2223 matrices = 3;
f44ee2c3 2224
ae1dd8e1
MN
2225 if(matrices > 384){
2226 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2227 return -1;
2228 }
3c3f113e 2229
ae1dd8e1 2230 for(n=0; n<matrices; n++){
bb270c08 2231 for (i = 0; i < 64; i++)
ae1dd8e1
MN
2232 s->base_matrix[n][i]= get_bits(gb, 8);
2233 }
2da2ba03 2234
ae1dd8e1
MN
2235 for (inter = 0; inter <= 1; inter++) {
2236 for (plane = 0; plane <= 2; plane++) {
2237 int newqr= 1;
2238 if (inter || plane > 0)
5fc32c27 2239 newqr = get_bits1(gb);
39922395 2240 if (!newqr) {
ae1dd8e1 2241 int qtj, plj;
5fc32c27 2242 if(inter && get_bits1(gb)){
ae1dd8e1
MN
2243 qtj = 0;
2244 plj = plane;
2245 }else{
2246 qtj= (3*inter + plane - 1) / 3;
2247 plj= (plane + 2) % 3;
2248 }
2249 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2250 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2251 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2252 } else {
2253 int qri= 0;
39922395 2254 int qi = 0;
ae1dd8e1
MN
2255
2256 for(;;){
2257 i= get_bits(gb, av_log2(matrices-1)+1);
2258 if(i>= matrices){
2259 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2260 return -1;
2261 }
2262 s->qr_base[inter][plane][qri]= i;
2263 if(qi >= 63)
2264 break;
2265 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2266 s->qr_size[inter][plane][qri++]= i;
2267 qi += i;
39922395 2268 }
ae1dd8e1 2269
2da2ba03 2270 if (qi > 63) {
7146d2c2 2271 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
bb270c08
DB
2272 return -1;
2273 }
ae1dd8e1 2274 s->qr_count[inter][plane]= qri;
39922395
MM
2275 }
2276 }
2277 }
2278
2da2ba03 2279 /* Huffman tables */
39922395
MM
2280 for (s->hti = 0; s->hti < 80; s->hti++) {
2281 s->entries = 0;
2282 s->huff_code_size = 1;
5fc32c27 2283 if (!get_bits1(gb)) {
39922395 2284 s->hbits = 0;
00bbe276
AC
2285 if(read_huffman_tree(avctx, gb))
2286 return -1;
39922395 2287 s->hbits = 1;
00bbe276
AC
2288 if(read_huffman_tree(avctx, gb))
2289 return -1;
39922395
MM
2290 }
2291 }
115329f1 2292
f44ee2c3 2293 s->theora_tables = 1;
115329f1 2294
f44ee2c3
AB
2295 return 0;
2296}
2297
5ef251e5 2298static av_cold int theora_decode_init(AVCodecContext *avctx)
f44ee2c3
AB
2299{
2300 Vp3DecodeContext *s = avctx->priv_data;
2301 GetBitContext gb;
2302 int ptype;
da91ed59
AJ
2303 uint8_t *header_start[3];
2304 int header_len[3];
2305 int i;
115329f1 2306
f44ee2c3
AB
2307 s->theora = 1;
2308
2309 if (!avctx->extradata_size)
7146d2c2
AB
2310 {
2311 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
bb270c08 2312 return -1;
7146d2c2 2313 }
f44ee2c3 2314
da91ed59
AJ
2315 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2316 42, header_start, header_len) < 0) {
2317 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2318 return -1;
2319 }
ee89b2b9 2320
da91ed59 2321 for(i=0;i<3;i++) {
fa6f2751 2322 init_get_bits(&gb, header_start[i], header_len[i] * 8);
f44ee2c3
AB
2323
2324 ptype = get_bits(&gb, 8);
115329f1 2325
7146d2c2
AB
2326 if (!(ptype & 0x80))
2327 {
2328 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
e278056f 2329// return -1;
115329f1 2330 }
7146d2c2 2331
3700dab4 2332 // FIXME: Check for this as well.
0a8dedc9 2333 skip_bits_long(&gb, 6*8); /* "theora" */
115329f1 2334
f44ee2c3
AB
2335 switch(ptype)
2336 {
2337 case 0x80:
e278056f 2338 theora_decode_header(avctx, &gb);
bb270c08
DB
2339 break;
2340 case 0x81:
2da2ba03 2341// FIXME: is this needed? it breaks sometimes
bb270c08
DB
2342// theora_decode_comments(avctx, gb);
2343 break;
2344 case 0x82:
00bbe276
AC
2345 if (theora_decode_tables(avctx, &gb))
2346 return -1;
bb270c08
DB
2347 break;
2348 default:
2349 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2350 break;
f44ee2c3 2351 }
12ce1f3f
DC
2352 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2353 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
116d866c
MC
2354 if (s->theora < 0x030200)
2355 break;
ee89b2b9 2356 }
f44ee2c3 2357
c79c960a 2358 return vp3_decode_init(avctx);
f44ee2c3
AB
2359}
2360
6f6a3e2a
DB
2361AVCodec theora_decoder = {
2362 "theora",
d86053a4 2363 CODEC_TYPE_VIDEO,
6f6a3e2a 2364 CODEC_ID_THEORA,
d86053a4 2365 sizeof(Vp3DecodeContext),
6f6a3e2a 2366 theora_decode_init,
d86053a4
MM
2367 NULL,
2368 vp3_decode_end,
2369 vp3_decode_frame,
7f57905d 2370 CODEC_CAP_DR1,
d5202e4f 2371 NULL,
fe4bf374 2372 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
d86053a4 2373};
6f6a3e2a 2374#endif
f44ee2c3 2375
6f6a3e2a
DB
2376AVCodec vp3_decoder = {
2377 "vp3",
f44ee2c3 2378 CODEC_TYPE_VIDEO,
6f6a3e2a 2379 CODEC_ID_VP3,
f44ee2c3 2380 sizeof(Vp3DecodeContext),
6f6a3e2a 2381 vp3_decode_init,
f44ee2c3
AB
2382 NULL,
2383 vp3_decode_end,
2384 vp3_decode_frame,
7f57905d 2385 CODEC_CAP_DR1,
d5202e4f 2386 NULL,
fe4bf374 2387 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
f44ee2c3 2388};