Implement CODEC_CAP_DRAW_HORIZ_BAND for VP3 decoder
[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;
a8de3901 138 int last_slice_end;
d86053a4 139
f2264fa5
DC
140 int qps[3];
141 int nqps;
142 int last_qps[3];
d86053a4
MM
143
144 int superblock_count;
892fc83e
MM
145 int y_superblock_width;
146 int y_superblock_height;
147 int c_superblock_width;
148 int c_superblock_height;
d86053a4
MM
149 int u_superblock_start;
150 int v_superblock_start;
151 unsigned char *superblock_coding;
152
153 int macroblock_count;
154 int macroblock_width;
155 int macroblock_height;
156
157 int fragment_count;
158 int fragment_width;
159 int fragment_height;
160
161 Vp3Fragment *all_fragments;
36e16253 162 uint8_t *coeff_counts;
7beddb12
MN
163 Coeff *coeffs;
164 Coeff *next_coeff;
1abbf64e 165 int fragment_start[3];
115329f1 166
36af0c95 167 ScanTable scantable;
115329f1 168
f44ee2c3
AB
169 /* tables */
170 uint16_t coded_dc_scale_factor[64];
67335dbc 171 uint32_t coded_ac_scale_factor[64];
ae1dd8e1
MN
172 uint8_t base_matrix[384][64];
173 uint8_t qr_count[2][3];
174 uint8_t qr_size [2][3][64];
175 uint16_t qr_base[2][3][64];
d86053a4 176
f4433de9 177 /* this is a list of indexes into the all_fragments array indicating
d86053a4
MM
178 * which of the fragments are coded */
179 int *coded_fragment_list;
180 int coded_fragment_list_index;
5e534865 181 int pixel_addresses_initialized;
d86053a4 182
098523eb
MM
183 /* track which fragments have already been decoded; called 'fast'
184 * because this data structure avoids having to iterate through every
185 * fragment in coded_fragment_list; once a fragment has been fully
186 * decoded, it is removed from this list */
187 int *fast_fragment_list;
188 int fragment_list_y_head;
189 int fragment_list_c_head;
190
d86053a4
MM
191 VLC dc_vlc[16];
192 VLC ac_vlc_1[16];
193 VLC ac_vlc_2[16];
194 VLC ac_vlc_3[16];
195 VLC ac_vlc_4[16];
196
0ad72bdd
MM
197 VLC superblock_run_length_vlc;
198 VLC fragment_run_length_vlc;
199 VLC mode_code_vlc;
200 VLC motion_vector_vlc;
201
38acbc3c
MM
202 /* these arrays need to be on 16-byte boundaries since SSE2 operations
203 * index into them */
c6727809 204 DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
d86053a4
MM
205
206 /* This table contains superblock_count * 16 entries. Each set of 16
f4433de9 207 * numbers corresponds to the fragment indexes 0..15 of the superblock.
d86053a4
MM
208 * An entry will be -1 to indicate that no entry corresponds to that
209 * index. */
210 int *superblock_fragments;
211
212 /* This table contains superblock_count * 4 entries. Each set of 4
f4433de9 213 * numbers corresponds to the macroblock indexes 0..3 of the superblock.
d86053a4
MM
214 * An entry will be -1 to indicate that no entry corresponds to that
215 * index. */
216 int *superblock_macroblocks;
217
218 /* This table contains macroblock_count * 6 entries. Each set of 6
f4433de9 219 * numbers corresponds to the fragment indexes 0..5 which comprise
d86053a4
MM
220 * the macroblock (4 Y fragments and 2 C fragments). */
221 int *macroblock_fragments;
115329f1 222 /* This is an array that indicates how a particular macroblock
74c0ac12 223 * is coded. */
96a7e73b 224 unsigned char *macroblock_coding;
d86053a4 225
04331882
MM
226 int first_coded_y_fragment;
227 int first_coded_c_fragment;
228 int last_coded_y_fragment;
229 int last_coded_c_fragment;
230
a2f11b3c 231 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
191e8ca7 232 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
39922395 233
f44b08a5
MM
234 /* Huffman decode */
235 int hti;
236 unsigned int hbits;
237 int entries;
238 int huff_code_size;
239 uint16_t huffman_table[80][32][2];
240
51ace577 241 uint8_t filter_limit_values[64];
c6727809 242 DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
d86053a4
MM
243} Vp3DecodeContext;
244
245/************************************************************************
246 * VP3 specific functions
247 ************************************************************************/
248
249/*
250 * This function sets up all of the various blocks mappings:
251 * superblocks <-> fragments, macroblocks <-> fragments,
252 * superblocks <-> macroblocks
892fc83e
MM
253 *
254 * Returns 0 is successful; returns 1 if *anything* went wrong.
d86053a4 255 */
115329f1 256static int init_block_mapping(Vp3DecodeContext *s)
d86053a4
MM
257{
258 int i, j;
d86053a4
MM
259 signed int hilbert_walk_mb[4];
260
261 int current_fragment = 0;
262 int current_width = 0;
263 int current_height = 0;
264 int right_edge = 0;
265 int bottom_edge = 0;
266 int superblock_row_inc = 0;
d86053a4
MM
267 int mapping_index = 0;
268
269 int current_macroblock;
270 int c_fragment;
271
bb991087 272 static const signed char travel_width[16] = {
115329f1 273 1, 1, 0, -1,
d86053a4
MM
274 0, 0, 1, 0,
275 1, 0, 1, 0,
276 0, -1, 0, 1
277 };
278
bb991087 279 static const signed char travel_height[16] = {
d86053a4
MM
280 0, 0, 1, 0,
281 1, 1, 0, -1,
282 0, 1, 0, -1,
283 -1, 0, -1, 0
284 };
285
bb991087 286 static const signed char travel_width_mb[4] = {
d86053a4
MM
287 1, 0, 1, 0
288 };
289
bb991087 290 static const signed char travel_height_mb[4] = {
d86053a4
MM
291 0, 1, 0, -1
292 };
293
d86053a4
MM
294 hilbert_walk_mb[0] = 1;
295 hilbert_walk_mb[1] = s->macroblock_width;
296 hilbert_walk_mb[2] = 1;
297 hilbert_walk_mb[3] = -s->macroblock_width;
298
299 /* iterate through each superblock (all planes) and map the fragments */
300 for (i = 0; i < s->superblock_count; i++) {
d86053a4
MM
301 /* time to re-assign the limits? */
302 if (i == 0) {
303
304 /* start of Y superblocks */
305 right_edge = s->fragment_width;
306 bottom_edge = s->fragment_height;
892fc83e 307 current_width = -1;
d86053a4 308 current_height = 0;
115329f1 309 superblock_row_inc = 3 * s->fragment_width -
892fc83e 310 (s->y_superblock_width * 4 - s->fragment_width);
d86053a4
MM
311
312 /* the first operation for this variable is to advance by 1 */
313 current_fragment = -1;
314
315 } else if (i == s->u_superblock_start) {
316
317 /* start of U superblocks */
318 right_edge = s->fragment_width / 2;
319 bottom_edge = s->fragment_height / 2;
892fc83e 320 current_width = -1;
d86053a4 321 current_height = 0;
115329f1 322 superblock_row_inc = 3 * (s->fragment_width / 2) -
892fc83e 323 (s->c_superblock_width * 4 - s->fragment_width / 2);
d86053a4
MM
324
325 /* the first operation for this variable is to advance by 1 */
1abbf64e 326 current_fragment = s->fragment_start[1] - 1;
d86053a4
MM
327
328 } else if (i == s->v_superblock_start) {
329
330 /* start of V superblocks */
331 right_edge = s->fragment_width / 2;
332 bottom_edge = s->fragment_height / 2;
892fc83e 333 current_width = -1;
d86053a4 334 current_height = 0;
115329f1 335 superblock_row_inc = 3 * (s->fragment_width / 2) -
892fc83e 336 (s->c_superblock_width * 4 - s->fragment_width / 2);
d86053a4
MM
337
338 /* the first operation for this variable is to advance by 1 */
1abbf64e 339 current_fragment = s->fragment_start[2] - 1;
d86053a4
MM
340
341 }
342
892fc83e 343 if (current_width >= right_edge - 1) {
d86053a4 344 /* reset width and move to next superblock row */
892fc83e 345 current_width = -1;
d86053a4
MM
346 current_height += 4;
347
348 /* fragment is now at the start of a new superblock row */
349 current_fragment += superblock_row_inc;
350 }
351
352 /* iterate through all 16 fragments in a superblock */
353 for (j = 0; j < 16; j++) {
684d9e36 354 current_fragment += travel_width[j] + right_edge * travel_height[j];
892fc83e 355 current_width += travel_width[j];
d86053a4
MM
356 current_height += travel_height[j];
357
358 /* check if the fragment is in bounds */
892fc83e 359 if ((current_width < right_edge) &&
d86053a4
MM
360 (current_height < bottom_edge)) {
361 s->superblock_fragments[mapping_index] = current_fragment;
d86053a4
MM
362 } else {
363 s->superblock_fragments[mapping_index] = -1;
d86053a4
MM
364 }
365
d86053a4
MM
366 mapping_index++;
367 }
368 }
369
370 /* initialize the superblock <-> macroblock mapping; iterate through
371 * all of the Y plane superblocks to build this mapping */
372 right_edge = s->macroblock_width;
373 bottom_edge = s->macroblock_height;
892fc83e 374 current_width = -1;
d86053a4 375 current_height = 0;
892fc83e 376 superblock_row_inc = s->macroblock_width -
cea96420 377 (s->y_superblock_width * 2 - s->macroblock_width);
d86053a4
MM
378 mapping_index = 0;
379 current_macroblock = -1;
380 for (i = 0; i < s->u_superblock_start; i++) {
381
96a7e73b 382 if (current_width >= right_edge - 1) {
d86053a4 383 /* reset width and move to next superblock row */
96a7e73b 384 current_width = -1;
d86053a4
MM
385 current_height += 2;
386
387 /* macroblock is now at the start of a new superblock row */
388 current_macroblock += superblock_row_inc;
389 }
390
391 /* iterate through each potential macroblock in the superblock */
392 for (j = 0; j < 4; j++) {
393 current_macroblock += hilbert_walk_mb[j];
892fc83e 394 current_width += travel_width_mb[j];
d86053a4
MM
395 current_height += travel_height_mb[j];
396
397 /* check if the macroblock is in bounds */
892fc83e 398 if ((current_width < right_edge) &&
d86053a4
MM
399 (current_height < bottom_edge)) {
400 s->superblock_macroblocks[mapping_index] = current_macroblock;
d86053a4
MM
401 } else {
402 s->superblock_macroblocks[mapping_index] = -1;
d86053a4
MM
403 }
404
d86053a4
MM
405 mapping_index++;
406 }
407 }
408
409 /* initialize the macroblock <-> fragment mapping */
410 current_fragment = 0;
411 current_macroblock = 0;
412 mapping_index = 0;
413 for (i = 0; i < s->fragment_height; i += 2) {
414
415 for (j = 0; j < s->fragment_width; j += 2) {
416
d86053a4
MM
417 s->all_fragments[current_fragment].macroblock = current_macroblock;
418 s->macroblock_fragments[mapping_index++] = current_fragment;
d86053a4
MM
419
420 if (j + 1 < s->fragment_width) {
421 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
422 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
d86053a4
MM
423 } else
424 s->macroblock_fragments[mapping_index++] = -1;
425
426 if (i + 1 < s->fragment_height) {
115329f1 427 s->all_fragments[current_fragment + s->fragment_width].macroblock =
d86053a4 428 current_macroblock;
115329f1 429 s->macroblock_fragments[mapping_index++] =
d86053a4 430 current_fragment + s->fragment_width;
d86053a4
MM
431 } else
432 s->macroblock_fragments[mapping_index++] = -1;
433
434 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
115329f1 435 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
d86053a4 436 current_macroblock;
115329f1 437 s->macroblock_fragments[mapping_index++] =
d86053a4 438 current_fragment + s->fragment_width + 1;
d86053a4
MM
439 } else
440 s->macroblock_fragments[mapping_index++] = -1;
441
442 /* C planes */
1abbf64e 443 c_fragment = s->fragment_start[1] +
d86053a4 444 (i * s->fragment_width / 4) + (j / 2);
892fc83e 445 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
d86053a4 446 s->macroblock_fragments[mapping_index++] = c_fragment;
d86053a4 447
1abbf64e 448 c_fragment = s->fragment_start[2] +
d86053a4 449 (i * s->fragment_width / 4) + (j / 2);
892fc83e 450 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
d86053a4 451 s->macroblock_fragments[mapping_index++] = c_fragment;
d86053a4
MM
452
453 if (j + 2 <= s->fragment_width)
454 current_fragment += 2;
115329f1 455 else
d86053a4
MM
456 current_fragment++;
457 current_macroblock++;
458 }
459
460 current_fragment += s->fragment_width;
461 }
892fc83e
MM
462
463 return 0; /* successful path out */
d86053a4
MM
464}
465
466/*
d86053a4
MM
467 * This function wipes out all of the fragment data.
468 */
469static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
470{
471 int i;
472
473 /* zero out all of the fragment information */
474 s->coded_fragment_list_index = 0;
475 for (i = 0; i < s->fragment_count; i++) {
36e16253 476 s->coeff_counts[i] = 0;
7dc9ed11
MM
477 s->all_fragments[i].motion_x = 127;
478 s->all_fragments[i].motion_y = 127;
479 s->all_fragments[i].next_coeff= NULL;
f2264fa5 480 s->all_fragments[i].qpi = 0;
7beddb12
MN
481 s->coeffs[i].index=
482 s->coeffs[i].coeff=0;
483 s->coeffs[i].next= NULL;
d86053a4
MM
484 }
485}
486
487/*
f44b08a5 488 * This function sets up the dequantization tables used for a particular
d86053a4
MM
489 * frame.
490 */
f2264fa5 491static void init_dequantizer(Vp3DecodeContext *s, int qpi)
d86053a4 492{
f2264fa5
DC
493 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
494 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
36c32bdd 495 int i, plane, inter, qri, bmi, bmj, qistart;
d86053a4 496
ae1dd8e1
MN
497 for(inter=0; inter<2; inter++){
498 for(plane=0; plane<3; plane++){
499 int sum=0;
500 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
501 sum+= s->qr_size[inter][plane][qri];
f2264fa5 502 if(s->qps[qpi] <= sum)
ae1dd8e1
MN
503 break;
504 }
505 qistart= sum - s->qr_size[inter][plane][qri];
506 bmi= s->qr_base[inter][plane][qri ];
507 bmj= s->qr_base[inter][plane][qri+1];
508 for(i=0; i<64; i++){
f2264fa5
DC
509 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
510 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
ae1dd8e1
MN
511 + s->qr_size[inter][plane][qri])
512 / (2*s->qr_size[inter][plane][qri]);
513
a14ab4e4 514 int qmin= 8<<(inter + !i);
ae1dd8e1
MN
515 int qscale= i ? ac_scale_factor : dc_scale_factor;
516
f2264fa5 517 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
ae1dd8e1 518 }
f2264fa5
DC
519 // all DC coefficients use the same quant so as not to interfere with DC prediction
520 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
ae1dd8e1 521 }
d86053a4 522 }
115329f1 523
f2264fa5 524 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
d86053a4
MM
525}
526
527/*
f44b08a5
MM
528 * This function initializes the loop filter boundary limits if the frame's
529 * quality index is different from the previous frame's.
7fa5f999
RD
530 *
531 * The filter_limit_values may not be larger than 127.
f44b08a5
MM
532 */
533static void init_loop_filter(Vp3DecodeContext *s)
534{
535 int *bounding_values= s->bounding_values_array+127;
536 int filter_limit;
537 int x;
7fa5f999 538 int value;
f44b08a5 539
f2264fa5 540 filter_limit = s->filter_limit_values[s->qps[0]];
f44b08a5
MM
541
542 /* set up the bounding values */
543 memset(s->bounding_values_array, 0, 256 * sizeof(int));
544 for (x = 0; x < filter_limit; x++) {
f44b08a5
MM
545 bounding_values[-x] = -x;
546 bounding_values[x] = x;
f44b08a5 547 }
7fa5f999
RD
548 for (x = value = filter_limit; x < 128 && value; x++, value--) {
549 bounding_values[ x] = value;
550 bounding_values[-x] = -value;
551 }
552 if (value)
553 bounding_values[128] = value;
357f45d9 554 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
f44b08a5
MM
555}
556
557/*
115329f1 558 * This function unpacks all of the superblock/macroblock/fragment coding
d86053a4
MM
559 * information from the bitstream.
560 */
892fc83e 561static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
562{
563 int bit = 0;
564 int current_superblock = 0;
565 int current_run = 0;
566 int decode_fully_flags = 0;
567 int decode_partial_blocks = 0;
22493ab9 568 int first_c_fragment_seen;
d86053a4
MM
569
570 int i, j;
571 int current_fragment;
572
d86053a4 573 if (s->keyframe) {
d86053a4
MM
574 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
575
576 } else {
577
578 /* unpack the list of partially-coded superblocks */
5fc32c27 579 bit = get_bits1(gb);
115329f1 580 /* toggle the bit because as soon as the first run length is
d86053a4
MM
581 * fetched the bit will be toggled again */
582 bit ^= 1;
583 while (current_superblock < s->superblock_count) {
b5da3635 584 if (current_run-- == 0) {
d86053a4 585 bit ^= 1;
115329f1 586 current_run = get_vlc2(gb,
b5da3635
MN
587 s->superblock_run_length_vlc.table, 6, 2);
588 if (current_run == 33)
d8278bab 589 current_run += get_bits(gb, 12);
d86053a4
MM
590
591 /* if any of the superblocks are not partially coded, flag
592 * a boolean to decode the list of fully-coded superblocks */
642d7e84 593 if (bit == 0) {
d86053a4 594 decode_fully_flags = 1;
642d7e84 595 } else {
d86053a4 596
642d7e84
MM
597 /* make a note of the fact that there are partially coded
598 * superblocks */
599 decode_partial_blocks = 1;
600 }
d86053a4 601 }
b5da3635 602 s->superblock_coding[current_superblock++] = bit;
d86053a4
MM
603 }
604
605 /* unpack the list of fully coded superblocks if any of the blocks were
606 * not marked as partially coded in the previous step */
607 if (decode_fully_flags) {
608
609 current_superblock = 0;
610 current_run = 0;
5fc32c27 611 bit = get_bits1(gb);
115329f1 612 /* toggle the bit because as soon as the first run length is
d86053a4
MM
613 * fetched the bit will be toggled again */
614 bit ^= 1;
615 while (current_superblock < s->superblock_count) {
616
617 /* skip any superblocks already marked as partially coded */
618 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
619
b5da3635 620 if (current_run-- == 0) {
d86053a4 621 bit ^= 1;
115329f1 622 current_run = get_vlc2(gb,
b5da3635
MN
623 s->superblock_run_length_vlc.table, 6, 2);
624 if (current_run == 33)
d8278bab 625 current_run += get_bits(gb, 12);
d86053a4 626 }
b5da3635 627 s->superblock_coding[current_superblock] = 2*bit;
d86053a4
MM
628 }
629 current_superblock++;
630 }
631 }
632
633 /* if there were partial blocks, initialize bitstream for
634 * unpacking fragment codings */
635 if (decode_partial_blocks) {
636
637 current_run = 0;
5fc32c27 638 bit = get_bits1(gb);
115329f1 639 /* toggle the bit because as soon as the first run length is
d86053a4
MM
640 * fetched the bit will be toggled again */
641 bit ^= 1;
642 }
643 }
644
645 /* figure out which fragments are coded; iterate through each
646 * superblock (all planes) */
647 s->coded_fragment_list_index = 0;
7beddb12 648 s->next_coeff= s->coeffs + s->fragment_count;
04331882
MM
649 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
650 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
22493ab9 651 first_c_fragment_seen = 0;
96a7e73b 652 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
d86053a4
MM
653 for (i = 0; i < s->superblock_count; i++) {
654
655 /* iterate through all 16 fragments in a superblock */
656 for (j = 0; j < 16; j++) {
657
658 /* if the fragment is in bounds, check its coding status */
659 current_fragment = s->superblock_fragments[i * 16 + j];
892fc83e 660 if (current_fragment >= s->fragment_count) {
9b879566 661 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
892fc83e
MM
662 current_fragment, s->fragment_count);
663 return 1;
664 }
d86053a4
MM
665 if (current_fragment != -1) {
666 if (s->superblock_coding[i] == SB_NOT_CODED) {
667
668 /* copy all the fragments from the prior frame */
115329f1 669 s->all_fragments[current_fragment].coding_method =
d86053a4
MM
670 MODE_COPY;
671
672 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
673
674 /* fragment may or may not be coded; this is the case
675 * that cares about the fragment coding runs */
b5da3635 676 if (current_run-- == 0) {
d86053a4 677 bit ^= 1;
115329f1 678 current_run = get_vlc2(gb,
b5da3635 679 s->fragment_run_length_vlc.table, 5, 2);
d86053a4
MM
680 }
681
682 if (bit) {
115329f1 683 /* default mode; actual mode will be decoded in
22493ab9 684 * the next phase */
115329f1 685 s->all_fragments[current_fragment].coding_method =
d86053a4 686 MODE_INTER_NO_MV;
7beddb12 687 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
115329f1 688 s->coded_fragment_list[s->coded_fragment_list_index] =
d86053a4 689 current_fragment;
1abbf64e 690 if ((current_fragment >= s->fragment_start[1]) &&
22493ab9
MM
691 (s->last_coded_y_fragment == -1) &&
692 (!first_c_fragment_seen)) {
04331882
MM
693 s->first_coded_c_fragment = s->coded_fragment_list_index;
694 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
22493ab9 695 first_c_fragment_seen = 1;
04331882
MM
696 }
697 s->coded_fragment_list_index++;
96a7e73b 698 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
d86053a4
MM
699 } else {
700 /* not coded; copy this fragment from the prior frame */
701 s->all_fragments[current_fragment].coding_method =
702 MODE_COPY;
d86053a4
MM
703 }
704
d86053a4
MM
705 } else {
706
707 /* fragments are fully coded in this superblock; actual
708 * coding will be determined in next step */
115329f1 709 s->all_fragments[current_fragment].coding_method =
d86053a4 710 MODE_INTER_NO_MV;
7beddb12 711 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
115329f1 712 s->coded_fragment_list[s->coded_fragment_list_index] =
d86053a4 713 current_fragment;
1abbf64e 714 if ((current_fragment >= s->fragment_start[1]) &&
22493ab9
MM
715 (s->last_coded_y_fragment == -1) &&
716 (!first_c_fragment_seen)) {
04331882
MM
717 s->first_coded_c_fragment = s->coded_fragment_list_index;
718 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
22493ab9 719 first_c_fragment_seen = 1;
04331882
MM
720 }
721 s->coded_fragment_list_index++;
96a7e73b 722 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
d86053a4
MM
723 }
724 }
725 }
726 }
04331882 727
22493ab9
MM
728 if (!first_c_fragment_seen)
729 /* only Y fragments coded in this frame */
04331882 730 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
115329f1 731 else
642d7e84 732 /* end the list of coded C fragments */
04331882 733 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
22493ab9 734
098523eb
MM
735 for (i = 0; i < s->fragment_count - 1; i++) {
736 s->fast_fragment_list[i] = i + 1;
737 }
738 s->fast_fragment_list[s->fragment_count - 1] = -1;
739
740 if (s->last_coded_y_fragment == -1)
741 s->fragment_list_y_head = -1;
742 else {
743 s->fragment_list_y_head = s->first_coded_y_fragment;
744 s->fast_fragment_list[s->last_coded_y_fragment] = -1;
745 }
746
747 if (s->last_coded_c_fragment == -1)
748 s->fragment_list_c_head = -1;
749 else {
750 s->fragment_list_c_head = s->first_coded_c_fragment;
751 s->fast_fragment_list[s->last_coded_c_fragment] = -1;
752 }
753
892fc83e 754 return 0;
d86053a4
MM
755}
756
757/*
758 * This function unpacks all the coding mode data for individual macroblocks
759 * from the bitstream.
760 */
892fc83e 761static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
762{
763 int i, j, k;
764 int scheme;
765 int current_macroblock;
766 int current_fragment;
767 int coding_mode;
e8e47435 768 int custom_mode_alphabet[CODING_MODE_COUNT];
d86053a4 769
d86053a4 770 if (s->keyframe) {
d86053a4
MM
771 for (i = 0; i < s->fragment_count; i++)
772 s->all_fragments[i].coding_method = MODE_INTRA;
773
774 } else {
775
776 /* fetch the mode coding scheme for this frame */
777 scheme = get_bits(gb, 3);
d86053a4
MM
778
779 /* is it a custom coding scheme? */
780 if (scheme == 0) {
d86053a4 781 for (i = 0; i < 8; i++)
2c823b3c
AC
782 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
783 for (i = 0; i < 8; i++)
e8e47435 784 custom_mode_alphabet[get_bits(gb, 3)] = i;
d86053a4
MM
785 }
786
d86053a4
MM
787 /* iterate through all of the macroblocks that contain 1 or more
788 * coded fragments */
789 for (i = 0; i < s->u_superblock_start; i++) {
790
791 for (j = 0; j < 4; j++) {
792 current_macroblock = s->superblock_macroblocks[i * 4 + j];
793 if ((current_macroblock == -1) ||
96a7e73b 794 (s->macroblock_coding[current_macroblock] == MODE_COPY))
d86053a4 795 continue;
892fc83e 796 if (current_macroblock >= s->macroblock_count) {
9b879566 797 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
892fc83e
MM
798 current_macroblock, s->macroblock_count);
799 return 1;
800 }
d86053a4
MM
801
802 /* mode 7 means get 3 bits for each coding mode */
803 if (scheme == 7)
804 coding_mode = get_bits(gb, 3);
e8e47435
SG
805 else if(scheme == 0)
806 coding_mode = custom_mode_alphabet
807 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
d86053a4 808 else
e8e47435 809 coding_mode = ModeAlphabet[scheme-1]
0ad72bdd 810 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
d86053a4 811
96a7e73b 812 s->macroblock_coding[current_macroblock] = coding_mode;
d86053a4 813 for (k = 0; k < 6; k++) {
115329f1 814 current_fragment =
d86053a4 815 s->macroblock_fragments[current_macroblock * 6 + k];
892fc83e
MM
816 if (current_fragment == -1)
817 continue;
818 if (current_fragment >= s->fragment_count) {
9b879566 819 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
892fc83e
MM
820 current_fragment, s->fragment_count);
821 return 1;
822 }
115329f1 823 if (s->all_fragments[current_fragment].coding_method !=
d86053a4
MM
824 MODE_COPY)
825 s->all_fragments[current_fragment].coding_method =
826 coding_mode;
827 }
d86053a4
MM
828 }
829 }
830 }
892fc83e
MM
831
832 return 0;
44ae98dd
MM
833}
834
835/*
d86053a4
MM
836 * This function unpacks all the motion vectors for the individual
837 * macroblocks from the bitstream.
838 */
892fc83e 839static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 840{
feaf1a73 841 int i, j, k, l;
d86053a4
MM
842 int coding_mode;
843 int motion_x[6];
844 int motion_y[6];
845 int last_motion_x = 0;
846 int last_motion_y = 0;
847 int prior_last_motion_x = 0;
848 int prior_last_motion_y = 0;
849 int current_macroblock;
850 int current_fragment;
851
6599e2a7 852 if (s->keyframe)
6298f49f 853 return 0;
10f38380 854
1ae4518d
DC
855 memset(motion_x, 0, 6 * sizeof(int));
856 memset(motion_y, 0, 6 * sizeof(int));
d86053a4 857
1ae4518d
DC
858 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
859 coding_mode = get_bits1(gb);
d86053a4 860
1ae4518d
DC
861 /* iterate through all of the macroblocks that contain 1 or more
862 * coded fragments */
863 for (i = 0; i < s->u_superblock_start; i++) {
d86053a4 864
1ae4518d
DC
865 for (j = 0; j < 4; j++) {
866 current_macroblock = s->superblock_macroblocks[i * 4 + j];
867 if ((current_macroblock == -1) ||
868 (s->macroblock_coding[current_macroblock] == MODE_COPY))
869 continue;
870 if (current_macroblock >= s->macroblock_count) {
871 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
872 current_macroblock, s->macroblock_count);
873 return 1;
874 }
d86053a4 875
1ae4518d
DC
876 current_fragment = s->macroblock_fragments[current_macroblock * 6];
877 if (current_fragment >= s->fragment_count) {
878 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
879 current_fragment, s->fragment_count);
880 return 1;
881 }
882 switch (s->macroblock_coding[current_macroblock]) {
883
884 case MODE_INTER_PLUS_MV:
885 case MODE_GOLDEN_MV:
886 /* all 6 fragments use the same motion vector */
887 if (coding_mode == 0) {
888 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
889 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
890 } else {
891 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
892 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
892fc83e 893 }
7f9926a4 894
1ae4518d
DC
895 /* vector maintenance, only on MODE_INTER_PLUS_MV */
896 if (s->macroblock_coding[current_macroblock] ==
897 MODE_INTER_PLUS_MV) {
e32e2d56
AJ
898 prior_last_motion_x = last_motion_x;
899 prior_last_motion_y = last_motion_y;
1ae4518d
DC
900 last_motion_x = motion_x[0];
901 last_motion_y = motion_y[0];
902 }
903 break;
904
905 case MODE_INTER_FOURMV:
906 /* vector maintenance */
907 prior_last_motion_x = last_motion_x;
908 prior_last_motion_y = last_motion_y;
909
910 /* fetch 4 vectors from the bitstream, one for each
911 * Y fragment, then average for the C fragment vectors */
912 motion_x[4] = motion_y[4] = 0;
913 for (k = 0; k < 4; k++) {
914 for (l = 0; l < s->coded_fragment_list_index; l++)
915 if (s->coded_fragment_list[l] == s->macroblock_fragments[6*current_macroblock + k])
916 break;
917 if (l < s->coded_fragment_list_index) {
918 if (coding_mode == 0) {
919 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
920 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
feaf1a73 921 } else {
1ae4518d
DC
922 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
923 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
feaf1a73 924 }
1ae4518d
DC
925 last_motion_x = motion_x[k];
926 last_motion_y = motion_y[k];
927 } else {
928 motion_x[k] = 0;
929 motion_y[k] = 0;
d86053a4 930 }
1ae4518d
DC
931 motion_x[4] += motion_x[k];
932 motion_y[4] += motion_y[k];
933 }
d86053a4 934
1ae4518d
DC
935 motion_x[5]=
936 motion_x[4]= RSHIFT(motion_x[4], 2);
937 motion_y[5]=
938 motion_y[4]= RSHIFT(motion_y[4], 2);
939 break;
940
941 case MODE_INTER_LAST_MV:
942 /* all 6 fragments use the last motion vector */
943 motion_x[0] = last_motion_x;
944 motion_y[0] = last_motion_y;
d86053a4 945
1ae4518d
DC
946 /* no vector maintenance (last vector remains the
947 * last vector) */
948 break;
949
950 case MODE_INTER_PRIOR_LAST:
951 /* all 6 fragments use the motion vector prior to the
952 * last motion vector */
953 motion_x[0] = prior_last_motion_x;
954 motion_y[0] = prior_last_motion_y;
d86053a4 955
1ae4518d
DC
956 /* vector maintenance */
957 prior_last_motion_x = last_motion_x;
958 prior_last_motion_y = last_motion_y;
959 last_motion_x = motion_x[0];
960 last_motion_y = motion_y[0];
961 break;
44ae98dd 962
1ae4518d
DC
963 default:
964 /* covers intra, inter without MV, golden without MV */
e6e32bdc
MM
965 motion_x[0] = 0;
966 motion_y[0] = 0;
44ae98dd 967
1ae4518d
DC
968 /* no vector maintenance */
969 break;
970 }
d86053a4 971
1ae4518d
DC
972 /* assign the motion vectors to the correct fragments */
973 for (k = 0; k < 6; k++) {
974 current_fragment =
975 s->macroblock_fragments[current_macroblock * 6 + k];
976 if (current_fragment == -1)
977 continue;
978 if (current_fragment >= s->fragment_count) {
979 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
980 current_fragment, s->fragment_count);
981 return 1;
d86053a4 982 }
e6e32bdc 983 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
da8af938
MM
984 s->all_fragments[current_fragment].motion_x = motion_x[k];
985 s->all_fragments[current_fragment].motion_y = motion_y[k];
e6e32bdc
MM
986 } else {
987 s->all_fragments[current_fragment].motion_x = motion_x[0];
988 s->all_fragments[current_fragment].motion_y = motion_y[0];
989 }
d86053a4
MM
990 }
991 }
1ae4518d 992 }
892fc83e
MM
993
994 return 0;
d86053a4
MM
995}
996
f2264fa5
DC
997static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
998{
999 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
1000 int num_blocks = s->coded_fragment_list_index;
1001
1002 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
1003 i = blocks_decoded = num_blocks_at_qpi = 0;
1004
1005 bit = get_bits1(gb);
1006
1007 do {
1008 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
1009 if (run_length == 34)
1010 run_length += get_bits(gb, 12);
1011 blocks_decoded += run_length;
1012
1013 if (!bit)
1014 num_blocks_at_qpi += run_length;
1015
1016 for (j = 0; j < run_length; i++) {
310afddf 1017 if (i >= s->coded_fragment_list_index)
f2264fa5
DC
1018 return -1;
1019
1020 if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
1021 s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
1022 j++;
1023 }
1024 }
1025
1026 if (run_length == 4129)
1027 bit = get_bits1(gb);
1028 else
1029 bit ^= 1;
1030 } while (blocks_decoded < num_blocks);
1031
1032 num_blocks -= num_blocks_at_qpi;
1033 }
1034
1035 return 0;
1036}
1037
115329f1 1038/*
d86053a4
MM
1039 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1040 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1041 * data. This function unpacks all the VLCs for either the Y plane or both
1042 * C planes, and is called for DC coefficients or different AC coefficient
1043 * levels (since different coefficient types require different VLC tables.
1044 *
1045 * This function returns a residual eob run. E.g, if a particular token gave
1046 * instructions to EOB the next 5 fragments and there were only 2 fragments
1047 * left in the current fragment range, 3 would be returned so that it could
1048 * be passed into the next call to this same function.
1049 */
1050static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1051 VLC *table, int coeff_index,
098523eb 1052 int y_plane,
d86053a4
MM
1053 int eob_run)
1054{
1055 int i;
1056 int token;
d3076955
MM
1057 int zero_run = 0;
1058 DCTELEM coeff = 0;
d86053a4 1059 Vp3Fragment *fragment;
d3076955 1060 int bits_to_get;
098523eb
MM
1061 int next_fragment;
1062 int previous_fragment;
1063 int fragment_num;
1064 int *list_head;
d86053a4 1065
ee3d7f58
MM
1066 /* local references to structure members to avoid repeated deferences */
1067 uint8_t *perm= s->scantable.permutated;
1068 int *coded_fragment_list = s->coded_fragment_list;
1069 Vp3Fragment *all_fragments = s->all_fragments;
1070 uint8_t *coeff_counts = s->coeff_counts;
1071 VLC_TYPE (*vlc_table)[2] = table->table;
098523eb 1072 int *fast_fragment_list = s->fast_fragment_list;
ee3d7f58 1073
098523eb
MM
1074 if (y_plane) {
1075 next_fragment = s->fragment_list_y_head;
1076 list_head = &s->fragment_list_y_head;
1077 } else {
1078 next_fragment = s->fragment_list_c_head;
1079 list_head = &s->fragment_list_c_head;
74c0ac12
MM
1080 }
1081
098523eb
MM
1082 i = next_fragment;
1083 previous_fragment = -1; /* this indicates that the previous fragment is actually the list head */
1084 while (i != -1) {
1085 fragment_num = coded_fragment_list[i];
d86053a4 1086
098523eb
MM
1087 if (coeff_counts[fragment_num] > coeff_index) {
1088 previous_fragment = i;
1089 i = fast_fragment_list[i];
d86053a4 1090 continue;
098523eb 1091 }
ee3d7f58 1092 fragment = &all_fragments[fragment_num];
d86053a4
MM
1093
1094 if (!eob_run) {
1095 /* decode a VLC into a token */
ee3d7f58 1096 token = get_vlc2(gb, vlc_table, 5, 3);
d86053a4 1097 /* use the token to get a zero run, a coefficient, and an eob run */
d3076955
MM
1098 if (token <= 6) {
1099 eob_run = eob_run_base[token];
1100 if (eob_run_get_bits[token])
1101 eob_run += get_bits(gb, eob_run_get_bits[token]);
1102 coeff = zero_run = 0;
1103 } else {
1104 bits_to_get = coeff_get_bits[token];
428984b0
MM
1105 if (bits_to_get)
1106 bits_to_get = get_bits(gb, bits_to_get);
1107 coeff = coeff_tables[token][bits_to_get];
d3076955
MM
1108
1109 zero_run = zero_run_base[token];
1110 if (zero_run_get_bits[token])
1111 zero_run += get_bits(gb, zero_run_get_bits[token]);
1112 }
d86053a4
MM
1113 }
1114
1115 if (!eob_run) {
ee3d7f58
MM
1116 coeff_counts[fragment_num] += zero_run;
1117 if (coeff_counts[fragment_num] < 64){
7beddb12 1118 fragment->next_coeff->coeff= coeff;
ee3d7f58 1119 fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
7beddb12
MN
1120 fragment->next_coeff->next= s->next_coeff;
1121 s->next_coeff->next=NULL;
1122 fragment->next_coeff= s->next_coeff++;
1123 }
098523eb
MM
1124 /* previous fragment is now this fragment */
1125 previous_fragment = i;
d86053a4 1126 } else {
ee3d7f58 1127 coeff_counts[fragment_num] |= 128;
d86053a4 1128 eob_run--;
098523eb
MM
1129 /* remove this fragment from the list */
1130 if (previous_fragment != -1)
1131 fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1132 else
1133 *list_head = fast_fragment_list[i];
1134 /* previous fragment remains unchanged */
d86053a4 1135 }
098523eb
MM
1136
1137 i = fast_fragment_list[i];
d86053a4
MM
1138 }
1139
1140 return eob_run;
1141}
1142
138fe832
MM
1143static void reverse_dc_prediction(Vp3DecodeContext *s,
1144 int first_fragment,
1145 int fragment_width,
1146 int fragment_height);
d86053a4
MM
1147/*
1148 * This function unpacks all of the DCT coefficient data from the
1149 * bitstream.
1150 */
892fc83e 1151static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
1152{
1153 int i;
1154 int dc_y_table;
1155 int dc_c_table;
1156 int ac_y_table;
1157 int ac_c_table;
1158 int residual_eob_run = 0;
9d8bb031
MM
1159 VLC *y_tables[64];
1160 VLC *c_tables[64];
d86053a4 1161
f4433de9 1162 /* fetch the DC table indexes */
d86053a4
MM
1163 dc_y_table = get_bits(gb, 4);
1164 dc_c_table = get_bits(gb, 4);
1165
1166 /* unpack the Y plane DC coefficients */
115329f1 1167 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
098523eb 1168 1, residual_eob_run);
d86053a4 1169
138fe832
MM
1170 /* reverse prediction of the Y-plane DC coefficients */
1171 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1172
d86053a4 1173 /* unpack the C plane DC coefficients */
d86053a4 1174 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
098523eb 1175 0, residual_eob_run);
d86053a4 1176
138fe832
MM
1177 /* reverse prediction of the C-plane DC coefficients */
1178 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1179 {
1180 reverse_dc_prediction(s, s->fragment_start[1],
1181 s->fragment_width / 2, s->fragment_height / 2);
1182 reverse_dc_prediction(s, s->fragment_start[2],
1183 s->fragment_width / 2, s->fragment_height / 2);
1184 }
1185
f4433de9 1186 /* fetch the AC table indexes */
d86053a4
MM
1187 ac_y_table = get_bits(gb, 4);
1188 ac_c_table = get_bits(gb, 4);
1189
9d8bb031 1190 /* build tables of AC VLC tables */
d86053a4 1191 for (i = 1; i <= 5; i++) {
9d8bb031
MM
1192 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1193 c_tables[i] = &s->ac_vlc_1[ac_c_table];
d86053a4 1194 }
d86053a4 1195 for (i = 6; i <= 14; i++) {
9d8bb031
MM
1196 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1197 c_tables[i] = &s->ac_vlc_2[ac_c_table];
d86053a4 1198 }
d86053a4 1199 for (i = 15; i <= 27; i++) {
9d8bb031
MM
1200 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1201 c_tables[i] = &s->ac_vlc_3[ac_c_table];
d86053a4 1202 }
d86053a4 1203 for (i = 28; i <= 63; i++) {
9d8bb031
MM
1204 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1205 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1206 }
1207
1208 /* decode all AC coefficents */
1209 for (i = 1; i <= 63; i++) {
1210 if (s->fragment_list_y_head != -1)
1211 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1212 1, residual_eob_run);
d86053a4 1213
9d8bb031
MM
1214 if (s->fragment_list_c_head != -1)
1215 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1216 0, residual_eob_run);
d86053a4 1217 }
892fc83e
MM
1218
1219 return 0;
d86053a4
MM
1220}
1221
1222/*
1223 * This function reverses the DC prediction for each coded fragment in
115329f1 1224 * the frame. Much of this function is adapted directly from the original
d86053a4
MM
1225 * VP3 source code.
1226 */
1227#define COMPATIBLE_FRAME(x) \
1228 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
7beddb12 1229#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
d86053a4
MM
1230
1231static void reverse_dc_prediction(Vp3DecodeContext *s,
1232 int first_fragment,
1233 int fragment_width,
115329f1 1234 int fragment_height)
d86053a4
MM
1235{
1236
1237#define PUL 8
1238#define PU 4
1239#define PUR 2
1240#define PL 1
1241
1242 int x, y;
1243 int i = first_fragment;
1244
59ef342b 1245 int predicted_dc;
d86053a4 1246
d86053a4
MM
1247 /* DC values for the left, up-left, up, and up-right fragments */
1248 int vl, vul, vu, vur;
1249
f4433de9 1250 /* indexes for the left, up-left, up, and up-right fragments */
d86053a4
MM
1251 int l, ul, u, ur;
1252
115329f1 1253 /*
d86053a4
MM
1254 * The 6 fields mean:
1255 * 0: up-left multiplier
1256 * 1: up multiplier
1257 * 2: up-right multiplier
1258 * 3: left multiplier
d86053a4 1259 */
bb991087 1260 static const int predictor_transform[16][4] = {
006ff1ca
MN
1261 { 0, 0, 0, 0},
1262 { 0, 0, 0,128}, // PL
1263 { 0, 0,128, 0}, // PUR
1264 { 0, 0, 53, 75}, // PUR|PL
1265 { 0,128, 0, 0}, // PU
1266 { 0, 64, 0, 64}, // PU|PL
1267 { 0,128, 0, 0}, // PU|PUR
1268 { 0, 0, 53, 75}, // PU|PUR|PL
1269 {128, 0, 0, 0}, // PUL
1270 { 0, 0, 0,128}, // PUL|PL
1271 { 64, 0, 64, 0}, // PUL|PUR
1272 { 0, 0, 53, 75}, // PUL|PUR|PL
1273 { 0,128, 0, 0}, // PUL|PU
1274 {-104,116, 0,116}, // PUL|PU|PL
1275 { 24, 80, 24, 0}, // PUL|PU|PUR
1276 {-104,116, 0,116} // PUL|PU|PUR|PL
d86053a4
MM
1277 };
1278
1279 /* This table shows which types of blocks can use other blocks for
1280 * prediction. For example, INTRA is the only mode in this table to
1281 * have a frame number of 0. That means INTRA blocks can only predict
115329f1 1282 * from other INTRA blocks. There are 2 golden frame coding types;
d86053a4
MM
1283 * blocks encoding in these modes can only predict from other blocks
1284 * that were encoded with these 1 of these 2 modes. */
50ba3fd7 1285 static const unsigned char compatible_frame[9] = {
d86053a4
MM
1286 1, /* MODE_INTER_NO_MV */
1287 0, /* MODE_INTRA */
1288 1, /* MODE_INTER_PLUS_MV */
1289 1, /* MODE_INTER_LAST_MV */
1290 1, /* MODE_INTER_PRIOR_MV */
1291 2, /* MODE_USING_GOLDEN */
1292 2, /* MODE_GOLDEN_MV */
50ba3fd7
JGG
1293 1, /* MODE_INTER_FOUR_MV */
1294 3 /* MODE_COPY */
d86053a4
MM
1295 };
1296 int current_frame_type;
1297
1298 /* there is a last DC predictor for each of the 3 frame types */
1299 short last_dc[3];
1300
1301 int transform = 0;
1302
d86053a4
MM
1303 vul = vu = vur = vl = 0;
1304 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1305
1306 /* for each fragment row... */
1307 for (y = 0; y < fragment_height; y++) {
1308
1309 /* for each fragment in a row... */
1310 for (x = 0; x < fragment_width; x++, i++) {
1311
1312 /* reverse prediction if this block was coded */
1313 if (s->all_fragments[i].coding_method != MODE_COPY) {
1314
115329f1 1315 current_frame_type =
d86053a4 1316 compatible_frame[s->all_fragments[i].coding_method];
d86053a4 1317
f72f8a77
MN
1318 transform= 0;
1319 if(x){
1320 l= i-1;
7beddb12 1321 vl = DC_COEFF(l);
50ba3fd7 1322 if(COMPATIBLE_FRAME(l))
006ff1ca 1323 transform |= PL;
f72f8a77
MN
1324 }
1325 if(y){
1326 u= i-fragment_width;
7beddb12 1327 vu = DC_COEFF(u);
50ba3fd7 1328 if(COMPATIBLE_FRAME(u))
006ff1ca 1329 transform |= PU;
f72f8a77
MN
1330 if(x){
1331 ul= i-fragment_width-1;
1332 vul = DC_COEFF(ul);
50ba3fd7 1333 if(COMPATIBLE_FRAME(ul))
006ff1ca 1334 transform |= PUL;
f72f8a77
MN
1335 }
1336 if(x + 1 < fragment_width){
1337 ur= i-fragment_width+1;
1338 vur = DC_COEFF(ur);
50ba3fd7 1339 if(COMPATIBLE_FRAME(ur))
006ff1ca 1340 transform |= PUR;
f72f8a77 1341 }
d86053a4
MM
1342 }
1343
d86053a4
MM
1344 if (transform == 0) {
1345
1346 /* if there were no fragments to predict from, use last
1347 * DC saved */
7beddb12 1348 predicted_dc = last_dc[current_frame_type];
d86053a4
MM
1349 } else {
1350
1351 /* apply the appropriate predictor transform */
1352 predicted_dc =
1353 (predictor_transform[transform][0] * vul) +
1354 (predictor_transform[transform][1] * vu) +
1355 (predictor_transform[transform][2] * vur) +
1356 (predictor_transform[transform][3] * vl);
1357
684d9e36 1358 predicted_dc /= 128;
d86053a4
MM
1359
1360 /* check for outranging on the [ul u l] and
1361 * [ul u ur l] predictors */
c11cb375 1362 if ((transform == 15) || (transform == 13)) {
c26abfa5 1363 if (FFABS(predicted_dc - vu) > 128)
d86053a4 1364 predicted_dc = vu;
c26abfa5 1365 else if (FFABS(predicted_dc - vl) > 128)
d86053a4 1366 predicted_dc = vl;
c26abfa5 1367 else if (FFABS(predicted_dc - vul) > 128)
d86053a4
MM
1368 predicted_dc = vul;
1369 }
d86053a4
MM
1370 }
1371
7beddb12
MN
1372 /* at long last, apply the predictor */
1373 if(s->coeffs[i].index){
1374 *s->next_coeff= s->coeffs[i];
1375 s->coeffs[i].index=0;
1376 s->coeffs[i].coeff=0;
1377 s->coeffs[i].next= s->next_coeff++;
1378 }
1379 s->coeffs[i].coeff += predicted_dc;
d86053a4 1380 /* save the DC */
7beddb12 1381 last_dc[current_frame_type] = DC_COEFF(i);
36e16253
RD
1382 if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1383 s->coeff_counts[i]= 129;
7beddb12
MN
1384// s->all_fragments[i].next_coeff= s->next_coeff;
1385 s->coeffs[i].next= s->next_coeff;
1386 (s->next_coeff++)->next=NULL;
1387 }
d86053a4
MM
1388 }
1389 }
1390 }
1391}
1392
256c0662 1393static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
fe313556 1394{
fe313556
DC
1395 int x, y;
1396 int *bounding_values= s->bounding_values_array+127;
1397
621f9a40
DC
1398 int width = s->fragment_width >> !!plane;
1399 int height = s->fragment_height >> !!plane;
1400 int fragment = s->fragment_start [plane] + ystart * width;
1401 int stride = s->current_frame.linesize[plane];
1402 uint8_t *plane_data = s->current_frame.data [plane];
1403 if (!s->flipped_image) stride = -stride;
1404
1405 for (y = ystart; y < yend; y++) {
1406
1407 for (x = 0; x < width; x++) {
1408 /* This code basically just deblocks on the edges of coded blocks.
1409 * However, it has to be much more complicated because of the
1410 * braindamaged deblock ordering used in VP3/Theora. Order matters
1411 * because some pixels get filtered twice. */
1412 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1413 {
1414 /* do not perform left edge filter for left columns frags */
1415 if (x > 0) {
1416 s->dsp.vp3_h_loop_filter(
1417 plane_data + s->all_fragments[fragment].first_pixel,
1418 stride, bounding_values);
1419 }
fe313556 1420
621f9a40
DC
1421 /* do not perform top edge filter for top row fragments */
1422 if (y > 0) {
1423 s->dsp.vp3_v_loop_filter(
1424 plane_data + s->all_fragments[fragment].first_pixel,
1425 stride, bounding_values);
1426 }
fe313556 1427
621f9a40
DC
1428 /* do not perform right edge filter for right column
1429 * fragments or if right fragment neighbor is also coded
1430 * in this frame (it will be filtered in next iteration) */
1431 if ((x < width - 1) &&
1432 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1433 s->dsp.vp3_h_loop_filter(
1434 plane_data + s->all_fragments[fragment + 1].first_pixel,
1435 stride, bounding_values);
fe313556
DC
1436 }
1437
621f9a40
DC
1438 /* do not perform bottom edge filter for bottom row
1439 * fragments or if bottom fragment neighbor is also coded
1440 * in this frame (it will be filtered in the next row) */
1441 if ((y < height - 1) &&
1442 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1443 s->dsp.vp3_v_loop_filter(
1444 plane_data + s->all_fragments[fragment + width].first_pixel,
1445 stride, bounding_values);
1446 }
fe313556 1447 }
621f9a40
DC
1448
1449 fragment++;
fe313556 1450 }
621f9a40 1451 }
fe313556
DC
1452}
1453
a8de3901
DC
1454/**
1455 * called when all pixels up to row y are complete
1456 */
1457static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1458{
1459 int h, cy;
1460 int offset[4];
1461
1462 if(s->avctx->draw_horiz_band==NULL)
1463 return;
1464
1465 h= y - s->last_slice_end;
1466 y -= h;
1467
1468 if (!s->flipped_image) {
1469 if (y == 0)
1470 h -= s->height - s->avctx->height; // account for non-mod16
1471 y = s->height - y - h;
1472 }
1473
1474 cy = y >> 1;
1475 offset[0] = s->current_frame.linesize[0]*y;
1476 offset[1] = s->current_frame.linesize[1]*cy;
1477 offset[2] = s->current_frame.linesize[2]*cy;
1478 offset[3] = 0;
1479
1480 emms_c();
1481 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1482 s->last_slice_end= y + h;
1483}
1484
d86053a4 1485/*
dc4b78d9
MM
1486 * Perform the final rendering for a particular slice of data.
1487 * The slice number ranges from 0..(macroblock_height - 1).
1488 */
1489static void render_slice(Vp3DecodeContext *s, int slice)
1490{
1abbf64e 1491 int x;
dc4b78d9 1492 int16_t *dequantizer;
c6727809 1493 DECLARE_ALIGNED_16(DCTELEM, block)[64];
dc4b78d9 1494 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
dc4b78d9
MM
1495 int motion_halfpel_index;
1496 uint8_t *motion_source;
1497 int plane;
dc4b78d9
MM
1498 int current_macroblock_entry = slice * s->macroblock_width * 6;
1499
1500 if (slice >= s->macroblock_height)
1501 return;
1502
1503 for (plane = 0; plane < 3; plane++) {
1abbf64e
MN
1504 uint8_t *output_plane = s->current_frame.data [plane];
1505 uint8_t * last_plane = s-> last_frame.data [plane];
1506 uint8_t *golden_plane = s-> golden_frame.data [plane];
1507 int stride = s->current_frame.linesize[plane];
1508 int plane_width = s->width >> !!plane;
1509 int plane_height = s->height >> !!plane;
1510 int y = slice * FRAGMENT_PIXELS << !plane ;
1511 int slice_height = y + (FRAGMENT_PIXELS << !plane);
1512 int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1513
1514 if (!s->flipped_image) stride = -stride;
dc4b78d9 1515
115329f1 1516
c26abfa5 1517 if(FFABS(stride) > 2048)
dc4b78d9
MM
1518 return; //various tables are fixed size
1519
1520 /* for each fragment row in the slice (both of them)... */
1521 for (; y < slice_height; y += 8) {
1522
1523 /* for each fragment in a row... */
1524 for (x = 0; x < plane_width; x += 8, i++) {
1525
1526 if ((i < 0) || (i >= s->fragment_count)) {
1527 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_slice(): bad fragment number (%d)\n", i);
1528 return;
1529 }
1530
1531 /* transform if this block was coded */
1532 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1533 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1534
1535 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1536 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1537 motion_source= golden_plane;
115329f1 1538 else
dc4b78d9
MM
1539 motion_source= last_plane;
1540
1541 motion_source += s->all_fragments[i].first_pixel;
1542 motion_halfpel_index = 0;
1543
1544 /* sort out the motion vector if this fragment is coded
1545 * using a motion vector method */
1546 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1547 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1548 int src_x, src_y;
1549 motion_x = s->all_fragments[i].motion_x;
1550 motion_y = s->all_fragments[i].motion_y;
1551 if(plane){
1552 motion_x= (motion_x>>1) | (motion_x&1);
1553 motion_y= (motion_y>>1) | (motion_y&1);
1554 }
1555
1556 src_x= (motion_x>>1) + x;
1557 src_y= (motion_y>>1) + y;
1558 if ((motion_x == 127) || (motion_y == 127))
1559 av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1560
1561 motion_halfpel_index = motion_x & 0x01;
1562 motion_source += (motion_x >> 1);
1563
1564 motion_halfpel_index |= (motion_y & 0x01) << 1;
1565 motion_source += ((motion_y >> 1) * stride);
1566
1567 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1568 uint8_t *temp= s->edge_emu_buffer;
1569 if(stride<0) temp -= 9*stride;
1570 else temp += 9*stride;
1571
1572 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1573 motion_source= temp;
1574 }
1575 }
115329f1 1576
dc4b78d9
MM
1577
1578 /* first, take care of copying a block from either the
1579 * previous or the golden frame */
1580 if (s->all_fragments[i].coding_method != MODE_INTRA) {
115329f1
DB
1581 /* Note, it is possible to implement all MC cases with
1582 put_no_rnd_pixels_l2 which would look more like the
1583 VP3 source but this would be slower as
dc4b78d9
MM
1584 put_no_rnd_pixels_tab is better optimzed */
1585 if(motion_halfpel_index != 3){
1586 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1587 output_plane + s->all_fragments[i].first_pixel,
1588 motion_source, stride, 8);
1589 }else{
1590 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1591 s->dsp.put_no_rnd_pixels_l2[1](
1592 output_plane + s->all_fragments[i].first_pixel,
115329f1
DB
1593 motion_source - d,
1594 motion_source + stride + 1 + d,
dc4b78d9
MM
1595 stride, 8);
1596 }
f2264fa5 1597 dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
dc4b78d9 1598 }else{
f2264fa5 1599 dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
dc4b78d9
MM
1600 }
1601
1602 /* dequantize the DCT coefficients */
dc4b78d9
MM
1603 if(s->avctx->idct_algo==FF_IDCT_VP3){
1604 Coeff *coeff= s->coeffs + i;
5fecfb7d 1605 s->dsp.clear_block(block);
dc4b78d9
MM
1606 while(coeff->next){
1607 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1608 coeff= coeff->next;
1609 }
1610 }else{
1611 Coeff *coeff= s->coeffs + i;
5fecfb7d 1612 s->dsp.clear_block(block);
dc4b78d9
MM
1613 while(coeff->next){
1614 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1615 coeff= coeff->next;
1616 }
1617 }
1618
1619 /* invert DCT and place (or add) in final output */
115329f1 1620
dc4b78d9
MM
1621 if (s->all_fragments[i].coding_method == MODE_INTRA) {
1622 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1623 block[0] += 128<<3;
1624 s->dsp.idct_put(
1625 output_plane + s->all_fragments[i].first_pixel,
1626 stride,
1627 block);
1628 } else {
1629 s->dsp.idct_add(
1630 output_plane + s->all_fragments[i].first_pixel,
1631 stride,
1632 block);
1633 }
dc4b78d9
MM
1634 } else {
1635
1636 /* copy directly from the previous frame */
1637 s->dsp.put_pixels_tab[1][0](
1638 output_plane + s->all_fragments[i].first_pixel,
1639 last_plane + s->all_fragments[i].first_pixel,
1640 stride, 8);
1641
1642 }
1643 }
256c0662
DC
1644 // Filter the previous block row. We can't filter the current row yet
1645 // since it needs pixels from the next row
1646 if (y > 0)
1647 apply_loop_filter(s, plane, (y>>3)-1, (y>>3));
dc4b78d9
MM
1648 }
1649 }
1650
dc4b78d9
MM
1651 /* this looks like a good place for slice dispatch... */
1652 /* algorithm:
dc4b78d9 1653 * if (slice == s->macroblock_height - 1)
f44b08a5
MM
1654 * dispatch (both last slice & 2nd-to-last slice);
1655 * else if (slice > 0)
1656 * dispatch (slice - 1);
dc4b78d9
MM
1657 */
1658
a8de3901
DC
1659 // now that we've filtered the last rows, they're safe to display
1660 if (slice)
1661 vp3_draw_horiz_band(s, 16*slice);
dc4b78d9
MM
1662}
1663
115329f1 1664/*
d86053a4
MM
1665 * This function computes the first pixel addresses for each fragment.
1666 * This function needs to be invoked after the first frame is allocated
1667 * so that it has access to the plane strides.
1668 */
115329f1 1669static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
d86053a4 1670{
fc961819
DC
1671#define Y_INITIAL(chroma_shift) s->flipped_image ? 1 : s->fragment_height >> chroma_shift
1672#define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
d86053a4
MM
1673
1674 int i, x, y;
fc961819 1675 const int y_inc = s->flipped_image ? 1 : -1;
d86053a4
MM
1676
1677 /* figure out the first pixel addresses for each of the fragments */
1678 /* Y plane */
1679 i = 0;
fc961819 1680 for (y = Y_INITIAL(0); Y_FINISHED(0); y += y_inc) {
d86053a4 1681 for (x = 0; x < s->fragment_width; x++) {
115329f1 1682 s->all_fragments[i++].first_pixel =
d86053a4
MM
1683 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1684 s->golden_frame.linesize[0] +
1685 x * FRAGMENT_PIXELS;
d86053a4
MM
1686 }
1687 }
1688
1689 /* U plane */
1abbf64e 1690 i = s->fragment_start[1];
fc961819 1691 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
d86053a4 1692 for (x = 0; x < s->fragment_width / 2; x++) {
115329f1 1693 s->all_fragments[i++].first_pixel =
d86053a4
MM
1694 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1695 s->golden_frame.linesize[1] +
1696 x * FRAGMENT_PIXELS;
d86053a4
MM
1697 }
1698 }
1699
1700 /* V plane */
1abbf64e 1701 i = s->fragment_start[2];
fc961819 1702 for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
9a7ad925 1703 for (x = 0; x < s->fragment_width / 2; x++) {
115329f1 1704 s->all_fragments[i++].first_pixel =
9a7ad925
AB
1705 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1706 s->golden_frame.linesize[2] +
1707 x * FRAGMENT_PIXELS;
9a7ad925
AB
1708 }
1709 }
1710}
1711
d86053a4
MM
1712/*
1713 * This is the ffmpeg/libavcodec API init function.
1714 */
98a6fff9 1715static av_cold int vp3_decode_init(AVCodecContext *avctx)
d86053a4
MM
1716{
1717 Vp3DecodeContext *s = avctx->priv_data;
ae1dd8e1 1718 int i, inter, plane;
892fc83e
MM
1719 int c_width;
1720 int c_height;
1721 int y_superblock_count;
1722 int c_superblock_count;
d86053a4 1723
3c3f113e 1724 if (avctx->codec_tag == MKTAG('V','P','3','0'))
bb270c08 1725 s->version = 0;
3c3f113e 1726 else
bb270c08 1727 s->version = 1;
3c3f113e 1728
d86053a4 1729 s->avctx = avctx;
ef516f73
DC
1730 s->width = FFALIGN(avctx->width, 16);
1731 s->height = FFALIGN(avctx->height, 16);
d86053a4 1732 avctx->pix_fmt = PIX_FMT_YUV420P;
580a7465 1733 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
8b6103da
MN
1734 if(avctx->idct_algo==FF_IDCT_AUTO)
1735 avctx->idct_algo=FF_IDCT_VP3;
d86053a4 1736 dsputil_init(&s->dsp, avctx);
115329f1 1737
36af0c95 1738 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
d86053a4
MM
1739
1740 /* initialize to an impossible value which will force a recalculation
1741 * in the first frame decode */
f2264fa5
DC
1742 for (i = 0; i < 3; i++)
1743 s->qps[i] = -1;
d86053a4 1744
892fc83e
MM
1745 s->y_superblock_width = (s->width + 31) / 32;
1746 s->y_superblock_height = (s->height + 31) / 32;
1747 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1748
1749 /* work out the dimensions for the C planes */
1750 c_width = s->width / 2;
1751 c_height = s->height / 2;
1752 s->c_superblock_width = (c_width + 31) / 32;
1753 s->c_superblock_height = (c_height + 31) / 32;
1754 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1755
1756 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1757 s->u_superblock_start = y_superblock_count;
1758 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
d86053a4
MM
1759 s->superblock_coding = av_malloc(s->superblock_count);
1760
1761 s->macroblock_width = (s->width + 15) / 16;
1762 s->macroblock_height = (s->height + 15) / 16;
1763 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1764
1765 s->fragment_width = s->width / FRAGMENT_PIXELS;
1766 s->fragment_height = s->height / FRAGMENT_PIXELS;
1767
1768 /* fragment count covers all 8x8 blocks for all 3 planes */
1769 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1abbf64e
MN
1770 s->fragment_start[1] = s->fragment_width * s->fragment_height;
1771 s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
d86053a4 1772
d86053a4 1773 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
36e16253 1774 s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
7beddb12 1775 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
d86053a4 1776 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
098523eb 1777 s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
5e534865 1778 s->pixel_addresses_initialized = 0;
0efbd068 1779 if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
098523eb 1780 !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
0efbd068
MM
1781 vp3_decode_end(avctx);
1782 return -1;
1783 }
d86053a4 1784
f44ee2c3
AB
1785 if (!s->theora_tables)
1786 {
2287c100 1787 for (i = 0; i < 64; i++) {
bb270c08 1788 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
bb270c08 1789 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
ae1dd8e1 1790 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
ae1dd8e1 1791 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
ae1dd8e1 1792 s->base_matrix[2][i] = vp31_inter_dequant[i];
bb270c08 1793 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2287c100 1794 }
f44ee2c3 1795
ae1dd8e1
MN
1796 for(inter=0; inter<2; inter++){
1797 for(plane=0; plane<3; plane++){
1798 s->qr_count[inter][plane]= 1;
1799 s->qr_size [inter][plane][0]= 63;
1800 s->qr_base [inter][plane][0]=
1801 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1802 }
1803 }
1804
39922395
MM
1805 /* init VLC tables */
1806 for (i = 0; i < 16; i++) {
1807
1808 /* DC histograms */
1809 init_vlc(&s->dc_vlc[i], 5, 32,
1810 &dc_bias[i][0][1], 4, 2,
1811 &dc_bias[i][0][0], 4, 2, 0);
1812
1813 /* group 1 AC histograms */
1814 init_vlc(&s->ac_vlc_1[i], 5, 32,
1815 &ac_bias_0[i][0][1], 4, 2,
1816 &ac_bias_0[i][0][0], 4, 2, 0);
1817
1818 /* group 2 AC histograms */
1819 init_vlc(&s->ac_vlc_2[i], 5, 32,
1820 &ac_bias_1[i][0][1], 4, 2,
1821 &ac_bias_1[i][0][0], 4, 2, 0);
1822
1823 /* group 3 AC histograms */
1824 init_vlc(&s->ac_vlc_3[i], 5, 32,
1825 &ac_bias_2[i][0][1], 4, 2,
1826 &ac_bias_2[i][0][0], 4, 2, 0);
1827
1828 /* group 4 AC histograms */
1829 init_vlc(&s->ac_vlc_4[i], 5, 32,
1830 &ac_bias_3[i][0][1], 4, 2,
1831 &ac_bias_3[i][0][0], 4, 2, 0);
1832 }
1833 } else {
1834 for (i = 0; i < 16; i++) {
1835
1836 /* DC histograms */
c4b7b8bf 1837 if (init_vlc(&s->dc_vlc[i], 5, 32,
39922395 1838 &s->huffman_table[i][0][1], 4, 2,
c4b7b8bf
RD
1839 &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1840 goto vlc_fail;
39922395
MM
1841
1842 /* group 1 AC histograms */
c4b7b8bf 1843 if (init_vlc(&s->ac_vlc_1[i], 5, 32,
39922395 1844 &s->huffman_table[i+16][0][1], 4, 2,
c4b7b8bf
RD
1845 &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1846 goto vlc_fail;
39922395
MM
1847
1848 /* group 2 AC histograms */
c4b7b8bf 1849 if (init_vlc(&s->ac_vlc_2[i], 5, 32,
39922395 1850 &s->huffman_table[i+16*2][0][1], 4, 2,
c4b7b8bf
RD
1851 &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1852 goto vlc_fail;
39922395
MM
1853
1854 /* group 3 AC histograms */
c4b7b8bf 1855 if (init_vlc(&s->ac_vlc_3[i], 5, 32,
39922395 1856 &s->huffman_table[i+16*3][0][1], 4, 2,
c4b7b8bf
RD
1857 &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1858 goto vlc_fail;
39922395
MM
1859
1860 /* group 4 AC histograms */
c4b7b8bf 1861 if (init_vlc(&s->ac_vlc_4[i], 5, 32,
39922395 1862 &s->huffman_table[i+16*4][0][1], 4, 2,
c4b7b8bf
RD
1863 &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1864 goto vlc_fail;
39922395 1865 }
d86053a4
MM
1866 }
1867
d8278bab
MM
1868 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1869 &superblock_run_length_vlc_table[0][1], 4, 2,
1870 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1871
dd36b667 1872 init_vlc(&s->fragment_run_length_vlc, 5, 30,
0ad72bdd
MM
1873 &fragment_run_length_vlc_table[0][1], 4, 2,
1874 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1875
1876 init_vlc(&s->mode_code_vlc, 3, 8,
1877 &mode_code_vlc_table[0][1], 2, 1,
1878 &mode_code_vlc_table[0][0], 2, 1, 0);
1879
1880 init_vlc(&s->motion_vector_vlc, 6, 63,
1881 &motion_vector_vlc_table[0][1], 2, 1,
1882 &motion_vector_vlc_table[0][0], 2, 1, 0);
1883
d86053a4
MM
1884 /* work out the block mapping tables */
1885 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1886 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
1887 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
96a7e73b 1888 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
0efbd068
MM
1889 if (!s->superblock_fragments || !s->superblock_macroblocks ||
1890 !s->macroblock_fragments || !s->macroblock_coding) {
1891 vp3_decode_end(avctx);
1892 return -1;
1893 }
d86053a4
MM
1894 init_block_mapping(s);
1895
44ae98dd
MM
1896 for (i = 0; i < 3; i++) {
1897 s->current_frame.data[i] = NULL;
1898 s->last_frame.data[i] = NULL;
1899 s->golden_frame.data[i] = NULL;
61873c4a
MM
1900 }
1901
d86053a4 1902 return 0;
c4b7b8bf
RD
1903
1904vlc_fail:
1905 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1906 return -1;
d86053a4
MM
1907}
1908
1909/*
1910 * This is the ffmpeg/libavcodec API frame decode function.
1911 */
115329f1 1912static int vp3_decode_frame(AVCodecContext *avctx,
d86053a4 1913 void *data, int *data_size,
7a00bbad 1914 AVPacket *avpkt)
d86053a4 1915{
7a00bbad
TB
1916 const uint8_t *buf = avpkt->data;
1917 int buf_size = avpkt->size;
d86053a4
MM
1918 Vp3DecodeContext *s = avctx->priv_data;
1919 GetBitContext gb;
1920 static int counter = 0;
dc4b78d9 1921 int i;
d86053a4 1922
d86053a4 1923 init_get_bits(&gb, buf, buf_size * 8);
115329f1 1924
f44ee2c3
AB
1925 if (s->theora && get_bits1(&gb))
1926 {
bb270c08
DB
1927 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1928 return -1;
f44ee2c3 1929 }
3c3f113e
AB
1930
1931 s->keyframe = !get_bits1(&gb);
1932 if (!s->theora)
bb270c08 1933 skip_bits(&gb, 1);
f2264fa5
DC
1934 for (i = 0; i < 3; i++)
1935 s->last_qps[i] = s->qps[i];
efea8528 1936
f2264fa5 1937 s->nqps=0;
efea8528 1938 do{
f2264fa5
DC
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;
d86053a4 1943
f8830383 1944 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
bb270c08 1945 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
f2264fa5 1946 s->keyframe?"key":"", counter, s->qps[0]);
d86053a4
MM
1947 counter++;
1948
f2264fa5 1949 if (s->qps[0] != s->last_qps[0])
f44b08a5 1950 init_loop_filter(s);
f2264fa5
DC
1951
1952 for (i = 0; i < s->nqps; i++)
1953 // reinit all dequantizers if the first one changed, because
1954 // the DC of the first quantizer must be used for all matrices
1955 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1956 init_dequantizer(s, i);
642d7e84 1957
068e82ba
DC
1958 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1959 return buf_size;
1960
d86053a4 1961 if (s->keyframe) {
bb270c08
DB
1962 if (!s->theora)
1963 {
1964 skip_bits(&gb, 4); /* width code */
1965 skip_bits(&gb, 4); /* height code */
1966 if (s->version)
1967 {
1968 s->version = get_bits(&gb, 5);
1969 if (counter == 1)
1970 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1971 }
1972 }
1973 if (s->version || s->theora)
1974 {
1975 if (get_bits1(&gb))
1976 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1977 skip_bits(&gb, 2); /* reserved? */
1978 }
3c3f113e 1979
74c0ac12
MM
1980 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1981 if (s->golden_frame.data[0])
1982 avctx->release_buffer(avctx, &s->golden_frame);
8e39d4a7 1983 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
74c0ac12
MM
1984 } else {
1985 if (s->golden_frame.data[0])
1986 avctx->release_buffer(avctx, &s->golden_frame);
1987 if (s->last_frame.data[0])
1988 avctx->release_buffer(avctx, &s->last_frame);
1989 }
d86053a4 1990
8e39d4a7 1991 s->golden_frame.reference = 3;
d86053a4 1992 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
9b879566 1993 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
d86053a4
MM
1994 return -1;
1995 }
1996
d86053a4 1997 /* golden frame is also the current frame */
4501e678 1998 s->current_frame= s->golden_frame;
d86053a4
MM
1999
2000 /* time to figure out pixel addresses? */
5e534865 2001 if (!s->pixel_addresses_initialized)
bb270c08 2002 {
ba31df31 2003 vp3_calculate_pixel_addresses(s);
5e534865 2004 s->pixel_addresses_initialized = 1;
bb270c08 2005 }
d86053a4 2006 } else {
d86053a4 2007 /* allocate a new current frame */
8e39d4a7 2008 s->current_frame.reference = 3;
5e534865 2009 if (!s->pixel_addresses_initialized) {
bc185f72
RD
2010 av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2011 return -1;
2012 }
d86053a4 2013 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
9b879566 2014 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
d86053a4
MM
2015 return -1;
2016 }
d86053a4
MM
2017 }
2018
b928ec64
MN
2019 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2020 s->current_frame.qstride= 0;
2021
d86053a4
MM
2022 init_frame(s, &gb);
2023
220a6f40
MN
2024 if (unpack_superblocks(s, &gb)){
2025 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2026 return -1;
2027 }
220a6f40
MN
2028 if (unpack_modes(s, &gb)){
2029 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2030 return -1;
2031 }
220a6f40
MN
2032 if (unpack_vectors(s, &gb)){
2033 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2034 return -1;
2035 }
f2264fa5
DC
2036 if (unpack_block_qpis(s, &gb)){
2037 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2038 return -1;
2039 }
220a6f40
MN
2040 if (unpack_dct_coeffs(s, &gb)){
2041 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
892fc83e
MM
2042 return -1;
2043 }
d86053a4 2044
a8de3901 2045 s->last_slice_end = 0;
dc4b78d9
MM
2046 for (i = 0; i < s->macroblock_height; i++)
2047 render_slice(s, i);
d86053a4 2048
256c0662
DC
2049 // filter the last row
2050 for (i = 0; i < 3; i++) {
2051 int row = (s->height >> (3+!!i)) - 1;
2052 apply_loop_filter(s, i, row, row+1);
2053 }
a8de3901 2054 vp3_draw_horiz_band(s, s->height);
892fc83e 2055
d86053a4
MM
2056 *data_size=sizeof(AVFrame);
2057 *(AVFrame*)data= s->current_frame;
2058
44ae98dd
MM
2059 /* release the last frame, if it is allocated and if it is not the
2060 * golden frame */
2061 if ((s->last_frame.data[0]) &&
2062 (s->last_frame.data[0] != s->golden_frame.data[0]))
2063 avctx->release_buffer(avctx, &s->last_frame);
d86053a4 2064
61873c4a 2065 /* shuffle frames (last = current) */
4501e678 2066 s->last_frame= s->current_frame;
8e39d4a7 2067 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
d86053a4
MM
2068
2069 return buf_size;
2070}
2071
2072/*
2073 * This is the ffmpeg/libavcodec API module cleanup function.
2074 */
98a6fff9 2075static av_cold int vp3_decode_end(AVCodecContext *avctx)
d86053a4
MM
2076{
2077 Vp3DecodeContext *s = avctx->priv_data;
6f4e2b5a 2078 int i;
d86053a4 2079
6f4e2b5a 2080 av_free(s->superblock_coding);
d86053a4 2081 av_free(s->all_fragments);
36e16253 2082 av_free(s->coeff_counts);
a2df5a50 2083 av_free(s->coeffs);
d86053a4 2084 av_free(s->coded_fragment_list);
098523eb 2085 av_free(s->fast_fragment_list);
d86053a4
MM
2086 av_free(s->superblock_fragments);
2087 av_free(s->superblock_macroblocks);
2088 av_free(s->macroblock_fragments);
96a7e73b 2089 av_free(s->macroblock_coding);
115329f1 2090
6f4e2b5a
MR
2091 for (i = 0; i < 16; i++) {
2092 free_vlc(&s->dc_vlc[i]);
2093 free_vlc(&s->ac_vlc_1[i]);
2094 free_vlc(&s->ac_vlc_2[i]);
2095 free_vlc(&s->ac_vlc_3[i]);
2096 free_vlc(&s->ac_vlc_4[i]);
2097 }
2098
2099 free_vlc(&s->superblock_run_length_vlc);
2100 free_vlc(&s->fragment_run_length_vlc);
2101 free_vlc(&s->mode_code_vlc);
2102 free_vlc(&s->motion_vector_vlc);
2103
d86053a4 2104 /* release all frames */
8e39d4a7 2105 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
892fc83e
MM
2106 avctx->release_buffer(avctx, &s->golden_frame);
2107 if (s->last_frame.data[0])
2108 avctx->release_buffer(avctx, &s->last_frame);
2109 /* no need to release the current_frame since it will always be pointing
2110 * to the same frame as either the golden or last frame */
d86053a4
MM
2111
2112 return 0;
2113}
2114
39922395
MM
2115static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2116{
2117 Vp3DecodeContext *s = avctx->priv_data;
2118
5fc32c27 2119 if (get_bits1(gb)) {
39922395
MM
2120 int token;
2121 if (s->entries >= 32) { /* overflow */
2122 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2123 return -1;
2124 }
2125 token = get_bits(gb, 5);
2126 //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);
2127 s->huffman_table[s->hti][token][0] = s->hbits;
2128 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2129 s->entries++;
2130 }
2131 else {
2132 if (s->huff_code_size >= 32) {/* overflow */
2133 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2134 return -1;
2135 }
2136 s->huff_code_size++;
2137 s->hbits <<= 1;
00bbe276
AC
2138 if (read_huffman_tree(avctx, gb))
2139 return -1;
39922395 2140 s->hbits |= 1;
00bbe276
AC
2141 if (read_huffman_tree(avctx, gb))
2142 return -1;
39922395
MM
2143 s->hbits >>= 1;
2144 s->huff_code_size--;
2145 }
2146 return 0;
2147}
2148
b250f9c6 2149#if CONFIG_THEORA_DECODER
e278056f 2150static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
2151{
2152 Vp3DecodeContext *s = avctx->priv_data;
ea3c2d53 2153 int visible_width, visible_height, colorspace;
9a7ad925 2154
e278056f 2155 s->theora = get_bits_long(gb, 24);
356306ac 2156 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
105c3d25 2157
ba7ee4a4 2158 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
9a7ad925 2159 /* but previous versions have the image flipped relative to vp3 */
ba7ee4a4 2160 if (s->theora < 0x030200)
9a7ad925 2161 {
bb270c08 2162 s->flipped_image = 1;
9a7ad925
AB
2163 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2164 }
f44ee2c3 2165
277e3e53
DC
2166 visible_width = s->width = get_bits(gb, 16) << 4;
2167 visible_height = s->height = get_bits(gb, 16) << 4;
115329f1 2168
0ecca7a4 2169 if(avcodec_check_dimensions(avctx, s->width, s->height)){
7146d2c2 2170 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
0ecca7a4
MN
2171 s->width= s->height= 0;
2172 return -1;
2173 }
7146d2c2 2174
277e3e53 2175 if (s->theora >= 0x030200) {
a0ce2d1b
DC
2176 visible_width = get_bits_long(gb, 24);
2177 visible_height = get_bits_long(gb, 24);
c0f716b8 2178
ba4816a0
AJ
2179 skip_bits(gb, 8); /* offset x */
2180 skip_bits(gb, 8); /* offset y */
2181 }
f44ee2c3 2182
e278056f
MN
2183 skip_bits(gb, 32); /* fps numerator */
2184 skip_bits(gb, 32); /* fps denumerator */
2185 skip_bits(gb, 24); /* aspect numerator */
2186 skip_bits(gb, 24); /* aspect denumerator */
115329f1 2187
ba7ee4a4 2188 if (s->theora < 0x030200)
e278056f 2189 skip_bits(gb, 5); /* keyframe frequency force */
ea3c2d53 2190 colorspace = get_bits(gb, 8);
e278056f 2191 skip_bits(gb, 24); /* bitrate */
f44ee2c3 2192
e278056f 2193 skip_bits(gb, 6); /* quality hint */
115329f1 2194
ba7ee4a4 2195 if (s->theora >= 0x030200)
105c3d25 2196 {
e278056f 2197 skip_bits(gb, 5); /* keyframe frequency force */
337f5c6e
DC
2198 skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2199 skip_bits(gb, 3); /* reserved */
105c3d25 2200 }
115329f1 2201
e278056f 2202// align_get_bits(gb);
115329f1 2203
c0f716b8
AJ
2204 if ( visible_width <= s->width && visible_width > s->width-16
2205 && visible_height <= s->height && visible_height > s->height-16)
2206 avcodec_set_dimensions(avctx, visible_width, visible_height);
2207 else
2208 avcodec_set_dimensions(avctx, s->width, s->height);
f44ee2c3 2209
ea3c2d53
DC
2210 if (colorspace == 1) {
2211 avctx->color_primaries = AVCOL_PRI_BT470M;
2212 } else if (colorspace == 2) {
2213 avctx->color_primaries = AVCOL_PRI_BT470BG;
2214 }
2215 if (colorspace == 1 || colorspace == 2) {
2216 avctx->colorspace = AVCOL_SPC_BT470BG;
2217 avctx->color_trc = AVCOL_TRC_BT709;
2218 }
2219
f44ee2c3
AB
2220 return 0;
2221}
2222
e278056f 2223static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
2224{
2225 Vp3DecodeContext *s = avctx->priv_data;
ae1dd8e1 2226 int i, n, matrices, inter, plane;
ba7ee4a4
MC
2227
2228 if (s->theora >= 0x030200) {
e278056f 2229 n = get_bits(gb, 3);
9c7154c7 2230 /* loop filter limit values table */
e13cca4b 2231 for (i = 0; i < 64; i++) {
e278056f 2232 s->filter_limit_values[i] = get_bits(gb, n);
e13cca4b
RD
2233 if (s->filter_limit_values[i] > 127) {
2234 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2235 s->filter_limit_values[i] = 127;
2236 }
2237 }
ba7ee4a4 2238 }
115329f1 2239
ba7ee4a4 2240 if (s->theora >= 0x030200)
e278056f 2241 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2242 else
2243 n = 16;
f44ee2c3
AB
2244 /* quality threshold table */
2245 for (i = 0; i < 64; i++)
e278056f 2246 s->coded_ac_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2247
ba7ee4a4 2248 if (s->theora >= 0x030200)
e278056f 2249 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2250 else
2251 n = 16;
f44ee2c3
AB
2252 /* dc scale factor table */
2253 for (i = 0; i < 64; i++)
e278056f 2254 s->coded_dc_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2255
ba7ee4a4 2256 if (s->theora >= 0x030200)
e278056f 2257 matrices = get_bits(gb, 9) + 1;
ba7ee4a4 2258 else
2da2ba03 2259 matrices = 3;
f44ee2c3 2260
ae1dd8e1
MN
2261 if(matrices > 384){
2262 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2263 return -1;
2264 }
3c3f113e 2265
ae1dd8e1 2266 for(n=0; n<matrices; n++){
bb270c08 2267 for (i = 0; i < 64; i++)
ae1dd8e1
MN
2268 s->base_matrix[n][i]= get_bits(gb, 8);
2269 }
2da2ba03 2270
ae1dd8e1
MN
2271 for (inter = 0; inter <= 1; inter++) {
2272 for (plane = 0; plane <= 2; plane++) {
2273 int newqr= 1;
2274 if (inter || plane > 0)
5fc32c27 2275 newqr = get_bits1(gb);
39922395 2276 if (!newqr) {
ae1dd8e1 2277 int qtj, plj;
5fc32c27 2278 if(inter && get_bits1(gb)){
ae1dd8e1
MN
2279 qtj = 0;
2280 plj = plane;
2281 }else{
2282 qtj= (3*inter + plane - 1) / 3;
2283 plj= (plane + 2) % 3;
2284 }
2285 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2286 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2287 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2288 } else {
2289 int qri= 0;
39922395 2290 int qi = 0;
ae1dd8e1
MN
2291
2292 for(;;){
2293 i= get_bits(gb, av_log2(matrices-1)+1);
2294 if(i>= matrices){
2295 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2296 return -1;
2297 }
2298 s->qr_base[inter][plane][qri]= i;
2299 if(qi >= 63)
2300 break;
2301 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2302 s->qr_size[inter][plane][qri++]= i;
2303 qi += i;
39922395 2304 }
ae1dd8e1 2305
2da2ba03 2306 if (qi > 63) {
7146d2c2 2307 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
bb270c08
DB
2308 return -1;
2309 }
ae1dd8e1 2310 s->qr_count[inter][plane]= qri;
39922395
MM
2311 }
2312 }
2313 }
2314
2da2ba03 2315 /* Huffman tables */
39922395
MM
2316 for (s->hti = 0; s->hti < 80; s->hti++) {
2317 s->entries = 0;
2318 s->huff_code_size = 1;
5fc32c27 2319 if (!get_bits1(gb)) {
39922395 2320 s->hbits = 0;
00bbe276
AC
2321 if(read_huffman_tree(avctx, gb))
2322 return -1;
39922395 2323 s->hbits = 1;
00bbe276
AC
2324 if(read_huffman_tree(avctx, gb))
2325 return -1;
39922395
MM
2326 }
2327 }
115329f1 2328
f44ee2c3 2329 s->theora_tables = 1;
115329f1 2330
f44ee2c3
AB
2331 return 0;
2332}
2333
5ef251e5 2334static av_cold int theora_decode_init(AVCodecContext *avctx)
f44ee2c3
AB
2335{
2336 Vp3DecodeContext *s = avctx->priv_data;
2337 GetBitContext gb;
2338 int ptype;
da91ed59
AJ
2339 uint8_t *header_start[3];
2340 int header_len[3];
2341 int i;
115329f1 2342
f44ee2c3
AB
2343 s->theora = 1;
2344
2345 if (!avctx->extradata_size)
7146d2c2
AB
2346 {
2347 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
bb270c08 2348 return -1;
7146d2c2 2349 }
f44ee2c3 2350
da91ed59
AJ
2351 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2352 42, header_start, header_len) < 0) {
2353 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2354 return -1;
2355 }
ee89b2b9 2356
da91ed59 2357 for(i=0;i<3;i++) {
fa6f2751 2358 init_get_bits(&gb, header_start[i], header_len[i] * 8);
f44ee2c3
AB
2359
2360 ptype = get_bits(&gb, 8);
115329f1 2361
7146d2c2
AB
2362 if (!(ptype & 0x80))
2363 {
2364 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
e278056f 2365// return -1;
115329f1 2366 }
7146d2c2 2367
3700dab4 2368 // FIXME: Check for this as well.
0a8dedc9 2369 skip_bits_long(&gb, 6*8); /* "theora" */
115329f1 2370
f44ee2c3
AB
2371 switch(ptype)
2372 {
2373 case 0x80:
e278056f 2374 theora_decode_header(avctx, &gb);
bb270c08
DB
2375 break;
2376 case 0x81:
2da2ba03 2377// FIXME: is this needed? it breaks sometimes
bb270c08
DB
2378// theora_decode_comments(avctx, gb);
2379 break;
2380 case 0x82:
00bbe276
AC
2381 if (theora_decode_tables(avctx, &gb))
2382 return -1;
bb270c08
DB
2383 break;
2384 default:
2385 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2386 break;
f44ee2c3 2387 }
12ce1f3f
DC
2388 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2389 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
116d866c
MC
2390 if (s->theora < 0x030200)
2391 break;
ee89b2b9 2392 }
f44ee2c3 2393
c79c960a 2394 return vp3_decode_init(avctx);
f44ee2c3
AB
2395}
2396
6f6a3e2a
DB
2397AVCodec theora_decoder = {
2398 "theora",
d86053a4 2399 CODEC_TYPE_VIDEO,
6f6a3e2a 2400 CODEC_ID_THEORA,
d86053a4 2401 sizeof(Vp3DecodeContext),
6f6a3e2a 2402 theora_decode_init,
d86053a4
MM
2403 NULL,
2404 vp3_decode_end,
2405 vp3_decode_frame,
a8de3901 2406 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
d5202e4f 2407 NULL,
fe4bf374 2408 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
d86053a4 2409};
6f6a3e2a 2410#endif
f44ee2c3 2411
6f6a3e2a
DB
2412AVCodec vp3_decoder = {
2413 "vp3",
f44ee2c3 2414 CODEC_TYPE_VIDEO,
6f6a3e2a 2415 CODEC_ID_VP3,
f44ee2c3 2416 sizeof(Vp3DecodeContext),
6f6a3e2a 2417 vp3_decode_init,
f44ee2c3
AB
2418 NULL,
2419 vp3_decode_end,
2420 vp3_decode_frame,
a8de3901 2421 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
d5202e4f 2422 NULL,
fe4bf374 2423 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
f44ee2c3 2424};