don't try to find file length if streaming
[libav.git] / libavcodec / vp3.c
CommitLineData
d86053a4 1/*
67335dbc 2 * Copyright (C) 2003-2004 the ffmpeg project
d86053a4
MM
3 *
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
8 *
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
13 *
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 *
d86053a4
MM
18 */
19
20/**
21 * @file vp3.c
22 * On2 VP3 Video Decoder
0ad72bdd
MM
23 *
24 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
25 * For more information about the VP3 coding process, visit:
26 * http://multimedia.cx/
27 *
28 * Theora decoder by Alex Beregszaszi
d86053a4
MM
29 */
30
31#include <stdio.h>
32#include <stdlib.h>
33#include <string.h>
34#include <unistd.h>
35
36#include "common.h"
37#include "avcodec.h"
38#include "dsputil.h"
39#include "mpegvideo.h"
d86053a4
MM
40
41#include "vp3data.h"
42
43#define FRAGMENT_PIXELS 8
44
45/*
46 * Debugging Variables
47 *
48 * Define one or more of the following compile-time variables to 1 to obtain
49 * elaborate information about certain aspects of the decoding process.
50 *
96a7e73b 51 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
d86053a4
MM
52 * DEBUG_VP3: high-level decoding flow
53 * DEBUG_INIT: initialization parameters
54 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
55 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
56 * DEBUG_MODES: unpacking the coding modes for individual fragments
57 * DEBUG_VECTORS: display the motion vectors
58 * DEBUG_TOKEN: display exhaustive information about each DCT token
59 * DEBUG_VLC: display the VLCs as they are extracted from the stream
60 * DEBUG_DC_PRED: display the process of reversing DC prediction
61 * DEBUG_IDCT: show every detail of the IDCT process
62 */
63
96a7e73b
MM
64#define KEYFRAMES_ONLY 0
65
d86053a4
MM
66#define DEBUG_VP3 0
67#define DEBUG_INIT 0
68#define DEBUG_DEQUANTIZERS 0
69#define DEBUG_BLOCK_CODING 0
70#define DEBUG_MODES 0
71#define DEBUG_VECTORS 0
72#define DEBUG_TOKEN 0
73#define DEBUG_VLC 0
74#define DEBUG_DC_PRED 0
75#define DEBUG_IDCT 0
76
77#if DEBUG_VP3
78#define debug_vp3 printf
79#else
80static inline void debug_vp3(const char *format, ...) { }
81#endif
82
83#if DEBUG_INIT
84#define debug_init printf
85#else
86static inline void debug_init(const char *format, ...) { }
87#endif
88
89#if DEBUG_DEQUANTIZERS
90#define debug_dequantizers printf
91#else
92static inline void debug_dequantizers(const char *format, ...) { }
93#endif
94
95#if DEBUG_BLOCK_CODING
96#define debug_block_coding printf
97#else
98static inline void debug_block_coding(const char *format, ...) { }
99#endif
100
101#if DEBUG_MODES
102#define debug_modes printf
103#else
104static inline void debug_modes(const char *format, ...) { }
105#endif
106
107#if DEBUG_VECTORS
108#define debug_vectors printf
109#else
110static inline void debug_vectors(const char *format, ...) { }
111#endif
112
113#if DEBUG_TOKEN
114#define debug_token printf
115#else
116static inline void debug_token(const char *format, ...) { }
117#endif
118
119#if DEBUG_VLC
120#define debug_vlc printf
121#else
122static inline void debug_vlc(const char *format, ...) { }
123#endif
124
125#if DEBUG_DC_PRED
126#define debug_dc_pred printf
127#else
128static inline void debug_dc_pred(const char *format, ...) { }
129#endif
130
131#if DEBUG_IDCT
132#define debug_idct printf
133#else
134static inline void debug_idct(const char *format, ...) { }
135#endif
136
7beddb12
MN
137typedef struct Coeff {
138 struct Coeff *next;
139 DCTELEM coeff;
140 uint8_t index;
141} Coeff;
142
143//FIXME split things out into their own arrays
d86053a4 144typedef struct Vp3Fragment {
7beddb12 145 Coeff *next_coeff;
d86053a4
MM
146 /* address of first pixel taking into account which plane the fragment
147 * lives on as well as the plane stride */
148 int first_pixel;
149 /* this is the macroblock that the fragment belongs to */
288774bb
MN
150 uint16_t macroblock;
151 uint8_t coding_method;
152 uint8_t coeff_count;
288774bb
MN
153 int8_t motion_x;
154 int8_t motion_y;
d86053a4
MM
155} Vp3Fragment;
156
157#define SB_NOT_CODED 0
158#define SB_PARTIALLY_CODED 1
159#define SB_FULLY_CODED 2
160
161#define MODE_INTER_NO_MV 0
162#define MODE_INTRA 1
163#define MODE_INTER_PLUS_MV 2
164#define MODE_INTER_LAST_MV 3
165#define MODE_INTER_PRIOR_LAST 4
166#define MODE_USING_GOLDEN 5
167#define MODE_GOLDEN_MV 6
168#define MODE_INTER_FOURMV 7
169#define CODING_MODE_COUNT 8
170
171/* special internal mode */
172#define MODE_COPY 8
173
174/* There are 6 preset schemes, plus a free-form scheme */
175static int ModeAlphabet[7][CODING_MODE_COUNT] =
176{
177 /* this is the custom scheme */
178 { 0, 0, 0, 0, 0, 0, 0, 0 },
179
180 /* scheme 1: Last motion vector dominates */
181 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
182 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
183 MODE_INTRA, MODE_USING_GOLDEN,
184 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
185
186 /* scheme 2 */
187 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
188 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
189 MODE_INTRA, MODE_USING_GOLDEN,
190 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
191
192 /* scheme 3 */
193 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
194 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
195 MODE_INTRA, MODE_USING_GOLDEN,
196 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
197
198 /* scheme 4 */
199 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
200 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
201 MODE_INTRA, MODE_USING_GOLDEN,
202 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
203
204 /* scheme 5: No motion vector dominates */
205 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
206 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
207 MODE_INTRA, MODE_USING_GOLDEN,
208 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
209
210 /* scheme 6 */
211 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
212 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
213 MODE_INTER_PLUS_MV, MODE_INTRA,
214 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
215
216};
217
218#define MIN_DEQUANT_VAL 2
219
220typedef struct Vp3DecodeContext {
221 AVCodecContext *avctx;
f44ee2c3 222 int theora, theora_tables;
3c3f113e 223 int version;
d86053a4 224 int width, height;
d86053a4
MM
225 AVFrame golden_frame;
226 AVFrame last_frame;
227 AVFrame current_frame;
228 int keyframe;
229 DSPContext dsp;
9a7ad925 230 int flipped_image;
d86053a4
MM
231
232 int quality_index;
233 int last_quality_index;
234
235 int superblock_count;
236 int superblock_width;
237 int superblock_height;
892fc83e
MM
238 int y_superblock_width;
239 int y_superblock_height;
240 int c_superblock_width;
241 int c_superblock_height;
d86053a4
MM
242 int u_superblock_start;
243 int v_superblock_start;
244 unsigned char *superblock_coding;
245
246 int macroblock_count;
247 int macroblock_width;
248 int macroblock_height;
249
250 int fragment_count;
251 int fragment_width;
252 int fragment_height;
253
254 Vp3Fragment *all_fragments;
7beddb12
MN
255 Coeff *coeffs;
256 Coeff *next_coeff;
d86053a4
MM
257 int u_fragment_start;
258 int v_fragment_start;
f44ee2c3 259
36af0c95
MN
260 ScanTable scantable;
261
f44ee2c3
AB
262 /* tables */
263 uint16_t coded_dc_scale_factor[64];
67335dbc 264 uint32_t coded_ac_scale_factor[64];
f44ee2c3
AB
265 uint16_t coded_intra_y_dequant[64];
266 uint16_t coded_intra_c_dequant[64];
267 uint16_t coded_inter_dequant[64];
d86053a4
MM
268
269 /* this is a list of indices into the all_fragments array indicating
270 * which of the fragments are coded */
271 int *coded_fragment_list;
272 int coded_fragment_list_index;
273 int pixel_addresses_inited;
274
275 VLC dc_vlc[16];
276 VLC ac_vlc_1[16];
277 VLC ac_vlc_2[16];
278 VLC ac_vlc_3[16];
279 VLC ac_vlc_4[16];
280
0ad72bdd
MM
281 VLC superblock_run_length_vlc;
282 VLC fragment_run_length_vlc;
283 VLC mode_code_vlc;
284 VLC motion_vector_vlc;
285
38acbc3c
MM
286 /* these arrays need to be on 16-byte boundaries since SSE2 operations
287 * index into them */
288 int16_t __align16 intra_y_dequant[64];
289 int16_t __align16 intra_c_dequant[64];
290 int16_t __align16 inter_dequant[64];
d86053a4
MM
291
292 /* This table contains superblock_count * 16 entries. Each set of 16
293 * numbers corresponds to the fragment indices 0..15 of the superblock.
294 * An entry will be -1 to indicate that no entry corresponds to that
295 * index. */
296 int *superblock_fragments;
297
298 /* This table contains superblock_count * 4 entries. Each set of 4
299 * numbers corresponds to the macroblock indices 0..3 of the superblock.
300 * An entry will be -1 to indicate that no entry corresponds to that
301 * index. */
302 int *superblock_macroblocks;
303
304 /* This table contains macroblock_count * 6 entries. Each set of 6
305 * numbers corresponds to the fragment indices 0..5 which comprise
306 * the macroblock (4 Y fragments and 2 C fragments). */
307 int *macroblock_fragments;
74c0ac12
MM
308 /* This is an array that indicates how a particular macroblock
309 * is coded. */
96a7e73b 310 unsigned char *macroblock_coding;
d86053a4 311
04331882
MM
312 int first_coded_y_fragment;
313 int first_coded_c_fragment;
314 int last_coded_y_fragment;
315 int last_coded_c_fragment;
316
a2f11b3c 317 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
b928ec64 318 uint8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
d86053a4
MM
319} Vp3DecodeContext;
320
3c3f113e
AB
321static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb);
322static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb);
323
d86053a4
MM
324/************************************************************************
325 * VP3 specific functions
326 ************************************************************************/
327
328/*
329 * This function sets up all of the various blocks mappings:
330 * superblocks <-> fragments, macroblocks <-> fragments,
331 * superblocks <-> macroblocks
892fc83e
MM
332 *
333 * Returns 0 is successful; returns 1 if *anything* went wrong.
d86053a4 334 */
892fc83e 335static int init_block_mapping(Vp3DecodeContext *s)
d86053a4
MM
336{
337 int i, j;
338 signed int hilbert_walk_y[16];
339 signed int hilbert_walk_c[16];
340 signed int hilbert_walk_mb[4];
341
342 int current_fragment = 0;
343 int current_width = 0;
344 int current_height = 0;
345 int right_edge = 0;
346 int bottom_edge = 0;
347 int superblock_row_inc = 0;
348 int *hilbert = NULL;
349 int mapping_index = 0;
350
351 int current_macroblock;
352 int c_fragment;
353
354 signed char travel_width[16] = {
355 1, 1, 0, -1,
356 0, 0, 1, 0,
357 1, 0, 1, 0,
358 0, -1, 0, 1
359 };
360
361 signed char travel_height[16] = {
362 0, 0, 1, 0,
363 1, 1, 0, -1,
364 0, 1, 0, -1,
365 -1, 0, -1, 0
366 };
367
368 signed char travel_width_mb[4] = {
369 1, 0, 1, 0
370 };
371
372 signed char travel_height_mb[4] = {
373 0, 1, 0, -1
374 };
375
376 debug_vp3(" vp3: initialize block mapping tables\n");
377
378 /* figure out hilbert pattern per these frame dimensions */
379 hilbert_walk_y[0] = 1;
380 hilbert_walk_y[1] = 1;
381 hilbert_walk_y[2] = s->fragment_width;
382 hilbert_walk_y[3] = -1;
383 hilbert_walk_y[4] = s->fragment_width;
384 hilbert_walk_y[5] = s->fragment_width;
385 hilbert_walk_y[6] = 1;
386 hilbert_walk_y[7] = -s->fragment_width;
387 hilbert_walk_y[8] = 1;
388 hilbert_walk_y[9] = s->fragment_width;
389 hilbert_walk_y[10] = 1;
390 hilbert_walk_y[11] = -s->fragment_width;
391 hilbert_walk_y[12] = -s->fragment_width;
392 hilbert_walk_y[13] = -1;
393 hilbert_walk_y[14] = -s->fragment_width;
394 hilbert_walk_y[15] = 1;
395
396 hilbert_walk_c[0] = 1;
397 hilbert_walk_c[1] = 1;
398 hilbert_walk_c[2] = s->fragment_width / 2;
399 hilbert_walk_c[3] = -1;
400 hilbert_walk_c[4] = s->fragment_width / 2;
401 hilbert_walk_c[5] = s->fragment_width / 2;
402 hilbert_walk_c[6] = 1;
403 hilbert_walk_c[7] = -s->fragment_width / 2;
404 hilbert_walk_c[8] = 1;
405 hilbert_walk_c[9] = s->fragment_width / 2;
406 hilbert_walk_c[10] = 1;
407 hilbert_walk_c[11] = -s->fragment_width / 2;
408 hilbert_walk_c[12] = -s->fragment_width / 2;
409 hilbert_walk_c[13] = -1;
410 hilbert_walk_c[14] = -s->fragment_width / 2;
411 hilbert_walk_c[15] = 1;
412
413 hilbert_walk_mb[0] = 1;
414 hilbert_walk_mb[1] = s->macroblock_width;
415 hilbert_walk_mb[2] = 1;
416 hilbert_walk_mb[3] = -s->macroblock_width;
417
418 /* iterate through each superblock (all planes) and map the fragments */
419 for (i = 0; i < s->superblock_count; i++) {
420 debug_init(" superblock %d (u starts @ %d, v starts @ %d)\n",
421 i, s->u_superblock_start, s->v_superblock_start);
422
423 /* time to re-assign the limits? */
424 if (i == 0) {
425
426 /* start of Y superblocks */
427 right_edge = s->fragment_width;
428 bottom_edge = s->fragment_height;
892fc83e 429 current_width = -1;
d86053a4 430 current_height = 0;
892fc83e
MM
431 superblock_row_inc = 3 * s->fragment_width -
432 (s->y_superblock_width * 4 - s->fragment_width);
d86053a4
MM
433 hilbert = hilbert_walk_y;
434
435 /* the first operation for this variable is to advance by 1 */
436 current_fragment = -1;
437
438 } else if (i == s->u_superblock_start) {
439
440 /* start of U superblocks */
441 right_edge = s->fragment_width / 2;
442 bottom_edge = s->fragment_height / 2;
892fc83e 443 current_width = -1;
d86053a4 444 current_height = 0;
892fc83e
MM
445 superblock_row_inc = 3 * (s->fragment_width / 2) -
446 (s->c_superblock_width * 4 - s->fragment_width / 2);
d86053a4
MM
447 hilbert = hilbert_walk_c;
448
449 /* the first operation for this variable is to advance by 1 */
450 current_fragment = s->u_fragment_start - 1;
451
452 } else if (i == s->v_superblock_start) {
453
454 /* start of V superblocks */
455 right_edge = s->fragment_width / 2;
456 bottom_edge = s->fragment_height / 2;
892fc83e 457 current_width = -1;
d86053a4 458 current_height = 0;
892fc83e
MM
459 superblock_row_inc = 3 * (s->fragment_width / 2) -
460 (s->c_superblock_width * 4 - s->fragment_width / 2);
d86053a4
MM
461 hilbert = hilbert_walk_c;
462
463 /* the first operation for this variable is to advance by 1 */
464 current_fragment = s->v_fragment_start - 1;
465
466 }
467
892fc83e 468 if (current_width >= right_edge - 1) {
d86053a4 469 /* reset width and move to next superblock row */
892fc83e 470 current_width = -1;
d86053a4
MM
471 current_height += 4;
472
473 /* fragment is now at the start of a new superblock row */
474 current_fragment += superblock_row_inc;
475 }
476
477 /* iterate through all 16 fragments in a superblock */
478 for (j = 0; j < 16; j++) {
479 current_fragment += hilbert[j];
892fc83e 480 current_width += travel_width[j];
d86053a4
MM
481 current_height += travel_height[j];
482
483 /* check if the fragment is in bounds */
892fc83e 484 if ((current_width < right_edge) &&
d86053a4
MM
485 (current_height < bottom_edge)) {
486 s->superblock_fragments[mapping_index] = current_fragment;
892fc83e
MM
487 debug_init(" mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
488 s->superblock_fragments[mapping_index], i, j,
489 current_width, right_edge, current_height, bottom_edge);
d86053a4
MM
490 } else {
491 s->superblock_fragments[mapping_index] = -1;
892fc83e
MM
492 debug_init(" superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
493 i, j,
494 current_width, right_edge, current_height, bottom_edge);
d86053a4
MM
495 }
496
d86053a4
MM
497 mapping_index++;
498 }
499 }
500
501 /* initialize the superblock <-> macroblock mapping; iterate through
502 * all of the Y plane superblocks to build this mapping */
503 right_edge = s->macroblock_width;
504 bottom_edge = s->macroblock_height;
892fc83e 505 current_width = -1;
d86053a4 506 current_height = 0;
892fc83e
MM
507 superblock_row_inc = s->macroblock_width -
508 (s->y_superblock_width * 2 - s->macroblock_width);;
d86053a4
MM
509 hilbert = hilbert_walk_mb;
510 mapping_index = 0;
511 current_macroblock = -1;
512 for (i = 0; i < s->u_superblock_start; i++) {
513
96a7e73b 514 if (current_width >= right_edge - 1) {
d86053a4 515 /* reset width and move to next superblock row */
96a7e73b 516 current_width = -1;
d86053a4
MM
517 current_height += 2;
518
519 /* macroblock is now at the start of a new superblock row */
520 current_macroblock += superblock_row_inc;
521 }
522
523 /* iterate through each potential macroblock in the superblock */
524 for (j = 0; j < 4; j++) {
525 current_macroblock += hilbert_walk_mb[j];
892fc83e 526 current_width += travel_width_mb[j];
d86053a4
MM
527 current_height += travel_height_mb[j];
528
529 /* check if the macroblock is in bounds */
892fc83e 530 if ((current_width < right_edge) &&
d86053a4
MM
531 (current_height < bottom_edge)) {
532 s->superblock_macroblocks[mapping_index] = current_macroblock;
96a7e73b
MM
533 debug_init(" mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
534 s->superblock_macroblocks[mapping_index], i, j,
535 current_width, right_edge, current_height, bottom_edge);
d86053a4
MM
536 } else {
537 s->superblock_macroblocks[mapping_index] = -1;
96a7e73b
MM
538 debug_init(" superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
539 i, j,
540 current_width, right_edge, current_height, bottom_edge);
d86053a4
MM
541 }
542
d86053a4
MM
543 mapping_index++;
544 }
545 }
546
547 /* initialize the macroblock <-> fragment mapping */
548 current_fragment = 0;
549 current_macroblock = 0;
550 mapping_index = 0;
551 for (i = 0; i < s->fragment_height; i += 2) {
552
553 for (j = 0; j < s->fragment_width; j += 2) {
554
555 debug_init(" macroblock %d contains fragments: ", current_macroblock);
556 s->all_fragments[current_fragment].macroblock = current_macroblock;
557 s->macroblock_fragments[mapping_index++] = current_fragment;
558 debug_init("%d ", current_fragment);
559
560 if (j + 1 < s->fragment_width) {
561 s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
562 s->macroblock_fragments[mapping_index++] = current_fragment + 1;
563 debug_init("%d ", current_fragment + 1);
564 } else
565 s->macroblock_fragments[mapping_index++] = -1;
566
567 if (i + 1 < s->fragment_height) {
568 s->all_fragments[current_fragment + s->fragment_width].macroblock =
569 current_macroblock;
570 s->macroblock_fragments[mapping_index++] =
571 current_fragment + s->fragment_width;
572 debug_init("%d ", current_fragment + s->fragment_width);
573 } else
574 s->macroblock_fragments[mapping_index++] = -1;
575
576 if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
577 s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
578 current_macroblock;
579 s->macroblock_fragments[mapping_index++] =
580 current_fragment + s->fragment_width + 1;
581 debug_init("%d ", current_fragment + s->fragment_width + 1);
582 } else
583 s->macroblock_fragments[mapping_index++] = -1;
584
585 /* C planes */
586 c_fragment = s->u_fragment_start +
587 (i * s->fragment_width / 4) + (j / 2);
892fc83e 588 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
d86053a4
MM
589 s->macroblock_fragments[mapping_index++] = c_fragment;
590 debug_init("%d ", c_fragment);
591
592 c_fragment = s->v_fragment_start +
593 (i * s->fragment_width / 4) + (j / 2);
892fc83e 594 s->all_fragments[c_fragment].macroblock = s->macroblock_count;
d86053a4
MM
595 s->macroblock_fragments[mapping_index++] = c_fragment;
596 debug_init("%d ", c_fragment);
597
598 debug_init("\n");
599
600 if (j + 2 <= s->fragment_width)
601 current_fragment += 2;
602 else
603 current_fragment++;
604 current_macroblock++;
605 }
606
607 current_fragment += s->fragment_width;
608 }
892fc83e
MM
609
610 return 0; /* successful path out */
d86053a4
MM
611}
612
613/*
614 * This function unpacks a single token (which should be in the range 0..31)
615 * and returns a zero run (number of zero coefficients in current DCT matrix
616 * before next non-zero coefficient), the next DCT coefficient, and the
617 * number of consecutive, non-EOB'd DCT blocks to EOB.
618 */
619static void unpack_token(GetBitContext *gb, int token, int *zero_run,
620 DCTELEM *coeff, int *eob_run)
621{
622 int sign;
623
624 *zero_run = 0;
625 *eob_run = 0;
626 *coeff = 0;
627
628 debug_token(" vp3 token %d: ", token);
629 switch (token) {
630
631 case 0:
632 debug_token("DCT_EOB_TOKEN, EOB next block\n");
633 *eob_run = 1;
634 break;
635
636 case 1:
637 debug_token("DCT_EOB_PAIR_TOKEN, EOB next 2 blocks\n");
638 *eob_run = 2;
639 break;
640
641 case 2:
642 debug_token("DCT_EOB_TRIPLE_TOKEN, EOB next 3 blocks\n");
643 *eob_run = 3;
644 break;
645
646 case 3:
647 debug_token("DCT_REPEAT_RUN_TOKEN, ");
648 *eob_run = get_bits(gb, 2) + 4;
649 debug_token("EOB the next %d blocks\n", *eob_run);
650 break;
651
652 case 4:
653 debug_token("DCT_REPEAT_RUN2_TOKEN, ");
654 *eob_run = get_bits(gb, 3) + 8;
655 debug_token("EOB the next %d blocks\n", *eob_run);
656 break;
657
658 case 5:
659 debug_token("DCT_REPEAT_RUN3_TOKEN, ");
660 *eob_run = get_bits(gb, 4) + 16;
661 debug_token("EOB the next %d blocks\n", *eob_run);
662 break;
663
664 case 6:
665 debug_token("DCT_REPEAT_RUN4_TOKEN, ");
666 *eob_run = get_bits(gb, 12);
667 debug_token("EOB the next %d blocks\n", *eob_run);
668 break;
669
670 case 7:
671 debug_token("DCT_SHORT_ZRL_TOKEN, ");
672 /* note that this token actually indicates that (3 extra bits) + 1 0s
673 * should be output; this case specifies a run of (3 EBs) 0s and a
674 * coefficient of 0. */
675 *zero_run = get_bits(gb, 3);
676 *coeff = 0;
677 debug_token("skip the next %d positions in output matrix\n", *zero_run + 1);
678 break;
679
680 case 8:
681 debug_token("DCT_ZRL_TOKEN, ");
682 /* note that this token actually indicates that (6 extra bits) + 1 0s
683 * should be output; this case specifies a run of (6 EBs) 0s and a
684 * coefficient of 0. */
685 *zero_run = get_bits(gb, 6);
686 *coeff = 0;
687 debug_token("skip the next %d positions in output matrix\n", *zero_run + 1);
688 break;
689
690 case 9:
691 debug_token("ONE_TOKEN, output 1\n");
692 *coeff = 1;
693 break;
694
695 case 10:
696 debug_token("MINUS_ONE_TOKEN, output -1\n");
697 *coeff = -1;
698 break;
699
700 case 11:
701 debug_token("TWO_TOKEN, output 2\n");
702 *coeff = 2;
703 break;
704
705 case 12:
706 debug_token("MINUS_TWO_TOKEN, output -2\n");
707 *coeff = -2;
708 break;
709
710 case 13:
711 case 14:
712 case 15:
713 case 16:
714 debug_token("LOW_VAL_TOKENS, ");
715 if (get_bits(gb, 1))
716 *coeff = -(3 + (token - 13));
717 else
718 *coeff = 3 + (token - 13);
719 debug_token("output %d\n", *coeff);
720 break;
721
722 case 17:
723 debug_token("DCT_VAL_CATEGORY3, ");
724 sign = get_bits(gb, 1);
725 *coeff = 7 + get_bits(gb, 1);
726 if (sign)
727 *coeff = -(*coeff);
728 debug_token("output %d\n", *coeff);
729 break;
730
731 case 18:
732 debug_token("DCT_VAL_CATEGORY4, ");
733 sign = get_bits(gb, 1);
734 *coeff = 9 + get_bits(gb, 2);
735 if (sign)
736 *coeff = -(*coeff);
737 debug_token("output %d\n", *coeff);
738 break;
739
740 case 19:
741 debug_token("DCT_VAL_CATEGORY5, ");
742 sign = get_bits(gb, 1);
743 *coeff = 13 + get_bits(gb, 3);
744 if (sign)
745 *coeff = -(*coeff);
746 debug_token("output %d\n", *coeff);
747 break;
748
749 case 20:
750 debug_token("DCT_VAL_CATEGORY6, ");
751 sign = get_bits(gb, 1);
752 *coeff = 21 + get_bits(gb, 4);
753 if (sign)
754 *coeff = -(*coeff);
755 debug_token("output %d\n", *coeff);
756 break;
757
758 case 21:
759 debug_token("DCT_VAL_CATEGORY7, ");
760 sign = get_bits(gb, 1);
761 *coeff = 37 + get_bits(gb, 5);
762 if (sign)
763 *coeff = -(*coeff);
764 debug_token("output %d\n", *coeff);
765 break;
766
767 case 22:
768 debug_token("DCT_VAL_CATEGORY8, ");
769 sign = get_bits(gb, 1);
770 *coeff = 69 + get_bits(gb, 9);
771 if (sign)
772 *coeff = -(*coeff);
773 debug_token("output %d\n", *coeff);
774 break;
775
776 case 23:
777 case 24:
778 case 25:
779 case 26:
780 case 27:
781 debug_token("DCT_RUN_CATEGORY1, ");
782 *zero_run = token - 22;
783 if (get_bits(gb, 1))
784 *coeff = -1;
785 else
786 *coeff = 1;
787 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
788 break;
789
790 case 28:
791 debug_token("DCT_RUN_CATEGORY1B, ");
792 if (get_bits(gb, 1))
793 *coeff = -1;
794 else
795 *coeff = 1;
796 *zero_run = 6 + get_bits(gb, 2);
797 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
798 break;
799
800 case 29:
801 debug_token("DCT_RUN_CATEGORY1C, ");
802 if (get_bits(gb, 1))
803 *coeff = -1;
804 else
805 *coeff = 1;
806 *zero_run = 10 + get_bits(gb, 3);
807 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
808 break;
809
810 case 30:
811 debug_token("DCT_RUN_CATEGORY2, ");
812 sign = get_bits(gb, 1);
813 *coeff = 2 + get_bits(gb, 1);
814 if (sign)
815 *coeff = -(*coeff);
816 *zero_run = 1;
817 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
818 break;
819
820 case 31:
821 debug_token("DCT_RUN_CATEGORY2, ");
822 sign = get_bits(gb, 1);
823 *coeff = 2 + get_bits(gb, 1);
824 if (sign)
825 *coeff = -(*coeff);
826 *zero_run = 2 + get_bits(gb, 1);
827 debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
828 break;
829
830 default:
9b879566 831 av_log(NULL, AV_LOG_ERROR, " vp3: help! Got a bad token: %d > 31\n", token);
d86053a4
MM
832 break;
833
834 }
835}
836
837/*
838 * This function wipes out all of the fragment data.
839 */
840static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
841{
842 int i;
843
844 /* zero out all of the fragment information */
845 s->coded_fragment_list_index = 0;
846 for (i = 0; i < s->fragment_count; i++) {
d86053a4 847 s->all_fragments[i].coeff_count = 0;
642d7e84
MM
848s->all_fragments[i].motion_x = 0xbeef;
849s->all_fragments[i].motion_y = 0xbeef;
7beddb12
MN
850s->all_fragments[i].next_coeff= NULL;
851 s->coeffs[i].index=
852 s->coeffs[i].coeff=0;
853 s->coeffs[i].next= NULL;
d86053a4
MM
854 }
855}
856
857/*
858 * This function sets of the dequantization tables used for a particular
859 * frame.
860 */
861static void init_dequantizer(Vp3DecodeContext *s)
862{
863
67335dbc 864 int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
f44ee2c3 865 int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
d86053a4
MM
866 int i, j;
867
868 debug_vp3(" vp3: initializing dequantization tables\n");
869
870 /*
871 * Scale dequantizers:
872 *
873 * quantizer * sf
874 * --------------
875 * 100
876 *
877 * where sf = dc_scale_factor for DC quantizer
67335dbc 878 * or ac_scale_factor for AC quantizer
d86053a4
MM
879 *
880 * Then, saturate the result to a lower limit of MIN_DEQUANT_VAL.
881 */
3d32b429 882#define SCALER 4
d86053a4
MM
883
884 /* scale DC quantizers */
f44ee2c3 885 s->intra_y_dequant[0] = s->coded_intra_y_dequant[0] * dc_scale_factor / 100;
d86053a4
MM
886 if (s->intra_y_dequant[0] < MIN_DEQUANT_VAL * 2)
887 s->intra_y_dequant[0] = MIN_DEQUANT_VAL * 2;
888 s->intra_y_dequant[0] *= SCALER;
889
f44ee2c3 890 s->intra_c_dequant[0] = s->coded_intra_c_dequant[0] * dc_scale_factor / 100;
d86053a4
MM
891 if (s->intra_c_dequant[0] < MIN_DEQUANT_VAL * 2)
892 s->intra_c_dequant[0] = MIN_DEQUANT_VAL * 2;
893 s->intra_c_dequant[0] *= SCALER;
894
f44ee2c3 895 s->inter_dequant[0] = s->coded_inter_dequant[0] * dc_scale_factor / 100;
d86053a4
MM
896 if (s->inter_dequant[0] < MIN_DEQUANT_VAL * 4)
897 s->inter_dequant[0] = MIN_DEQUANT_VAL * 4;
898 s->inter_dequant[0] *= SCALER;
899
900 /* scale AC quantizers, zigzag at the same time in preparation for
901 * the dequantization phase */
902 for (i = 1; i < 64; i++) {
36af0c95
MN
903 int k= s->scantable.scantable[i];
904 j = s->scantable.permutated[i];
d86053a4 905
36af0c95 906 s->intra_y_dequant[j] = s->coded_intra_y_dequant[k] * ac_scale_factor / 100;
d86053a4
MM
907 if (s->intra_y_dequant[j] < MIN_DEQUANT_VAL)
908 s->intra_y_dequant[j] = MIN_DEQUANT_VAL;
909 s->intra_y_dequant[j] *= SCALER;
910
36af0c95 911 s->intra_c_dequant[j] = s->coded_intra_c_dequant[k] * ac_scale_factor / 100;
d86053a4
MM
912 if (s->intra_c_dequant[j] < MIN_DEQUANT_VAL)
913 s->intra_c_dequant[j] = MIN_DEQUANT_VAL;
914 s->intra_c_dequant[j] *= SCALER;
915
36af0c95 916 s->inter_dequant[j] = s->coded_inter_dequant[k] * ac_scale_factor / 100;
d86053a4
MM
917 if (s->inter_dequant[j] < MIN_DEQUANT_VAL * 2)
918 s->inter_dequant[j] = MIN_DEQUANT_VAL * 2;
919 s->inter_dequant[j] *= SCALER;
920 }
b928ec64
MN
921
922 memset(s->qscale_table, (FFMAX(s->intra_y_dequant[1], s->intra_c_dequant[1])+8)/16, 512); //FIXME finetune
d86053a4
MM
923
924 /* print debug information as requested */
925 debug_dequantizers("intra Y dequantizers:\n");
926 for (i = 0; i < 8; i++) {
927 for (j = i * 8; j < i * 8 + 8; j++) {
928 debug_dequantizers(" %4d,", s->intra_y_dequant[j]);
929 }
930 debug_dequantizers("\n");
931 }
932 debug_dequantizers("\n");
933
934 debug_dequantizers("intra C dequantizers:\n");
935 for (i = 0; i < 8; i++) {
936 for (j = i * 8; j < i * 8 + 8; j++) {
937 debug_dequantizers(" %4d,", s->intra_c_dequant[j]);
938 }
939 debug_dequantizers("\n");
940 }
941 debug_dequantizers("\n");
942
943 debug_dequantizers("interframe dequantizers:\n");
944 for (i = 0; i < 8; i++) {
945 for (j = i * 8; j < i * 8 + 8; j++) {
946 debug_dequantizers(" %4d,", s->inter_dequant[j]);
947 }
948 debug_dequantizers("\n");
949 }
950 debug_dequantizers("\n");
951}
952
953/*
954 * This function is used to fetch runs of 1s or 0s from the bitstream for
955 * use in determining which superblocks are fully and partially coded.
956 *
957 * Codeword RunLength
958 * 0 1
959 * 10x 2-3
960 * 110x 4-5
961 * 1110xx 6-9
962 * 11110xxx 10-17
963 * 111110xxxx 18-33
964 * 111111xxxxxxxxxxxx 34-4129
965 */
966static int get_superblock_run_length(GetBitContext *gb)
967{
968
969 if (get_bits(gb, 1) == 0)
970 return 1;
971
972 else if (get_bits(gb, 1) == 0)
973 return (2 + get_bits(gb, 1));
974
975 else if (get_bits(gb, 1) == 0)
976 return (4 + get_bits(gb, 1));
977
978 else if (get_bits(gb, 1) == 0)
979 return (6 + get_bits(gb, 2));
980
981 else if (get_bits(gb, 1) == 0)
982 return (10 + get_bits(gb, 3));
983
984 else if (get_bits(gb, 1) == 0)
985 return (18 + get_bits(gb, 4));
986
987 else
988 return (34 + get_bits(gb, 12));
989
990}
991
992/*
993 * This function is used to fetch runs of 1s or 0s from the bitstream for
994 * use in determining which particular fragments are coded.
995 *
996 * Codeword RunLength
997 * 0x 1-2
998 * 10x 3-4
999 * 110x 5-6
1000 * 1110xx 7-10
1001 * 11110xx 11-14
1002 * 11111xxxx 15-30
1003 */
1004static int get_fragment_run_length(GetBitContext *gb)
1005{
1006
1007 if (get_bits(gb, 1) == 0)
1008 return (1 + get_bits(gb, 1));
1009
1010 else if (get_bits(gb, 1) == 0)
1011 return (3 + get_bits(gb, 1));
1012
1013 else if (get_bits(gb, 1) == 0)
1014 return (5 + get_bits(gb, 1));
1015
1016 else if (get_bits(gb, 1) == 0)
1017 return (7 + get_bits(gb, 2));
1018
1019 else if (get_bits(gb, 1) == 0)
1020 return (11 + get_bits(gb, 2));
1021
1022 else
1023 return (15 + get_bits(gb, 4));
1024
1025}
1026
1027/*
1028 * This function decodes a VLC from the bitstream and returns a number
1029 * that ranges from 0..7. The number indicates which of the 8 coding
1030 * modes to use.
1031 *
1032 * VLC Number
1033 * 0 0
1034 * 10 1
1035 * 110 2
1036 * 1110 3
1037 * 11110 4
1038 * 111110 5
1039 * 1111110 6
1040 * 1111111 7
1041 *
1042 */
1043static int get_mode_code(GetBitContext *gb)
1044{
1045
1046 if (get_bits(gb, 1) == 0)
1047 return 0;
1048
1049 else if (get_bits(gb, 1) == 0)
1050 return 1;
1051
1052 else if (get_bits(gb, 1) == 0)
1053 return 2;
1054
1055 else if (get_bits(gb, 1) == 0)
1056 return 3;
1057
1058 else if (get_bits(gb, 1) == 0)
1059 return 4;
1060
1061 else if (get_bits(gb, 1) == 0)
1062 return 5;
1063
1064 else if (get_bits(gb, 1) == 0)
1065 return 6;
1066
1067 else
1068 return 7;
1069
1070}
1071
1072/*
1073 * This function extracts a motion vector from the bitstream using a VLC
1074 * scheme. 3 bits are fetched from the bitstream and 1 of 8 actions is
1075 * taken depending on the value on those 3 bits:
1076 *
1077 * 0: return 0
1078 * 1: return 1
1079 * 2: return -1
1080 * 3: if (next bit is 1) return -2, else return 2
1081 * 4: if (next bit is 1) return -3, else return 3
1082 * 5: return 4 + (next 2 bits), next bit is sign
1083 * 6: return 8 + (next 3 bits), next bit is sign
1084 * 7: return 16 + (next 4 bits), next bit is sign
1085 */
1086static int get_motion_vector_vlc(GetBitContext *gb)
1087{
1088 int bits;
1089
1090 bits = get_bits(gb, 3);
1091
1092 switch(bits) {
1093
1094 case 0:
1095 bits = 0;
1096 break;
1097
1098 case 1:
1099 bits = 1;
1100 break;
1101
1102 case 2:
1103 bits = -1;
1104 break;
1105
1106 case 3:
1107 if (get_bits(gb, 1) == 0)
1108 bits = 2;
1109 else
1110 bits = -2;
1111 break;
1112
1113 case 4:
1114 if (get_bits(gb, 1) == 0)
1115 bits = 3;
1116 else
1117 bits = -3;
1118 break;
1119
1120 case 5:
1121 bits = 4 + get_bits(gb, 2);
1122 if (get_bits(gb, 1) == 1)
1123 bits = -bits;
1124 break;
1125
1126 case 6:
1127 bits = 8 + get_bits(gb, 3);
1128 if (get_bits(gb, 1) == 1)
1129 bits = -bits;
1130 break;
1131
1132 case 7:
1133 bits = 16 + get_bits(gb, 4);
1134 if (get_bits(gb, 1) == 1)
1135 bits = -bits;
1136 break;
1137
1138 }
1139
1140 return bits;
1141}
1142
1143/*
1144 * This function fetches a 5-bit number from the stream followed by
1145 * a sign and calls it a motion vector.
1146 */
1147static int get_motion_vector_fixed(GetBitContext *gb)
1148{
1149
1150 int bits;
1151
1152 bits = get_bits(gb, 5);
1153
1154 if (get_bits(gb, 1) == 1)
1155 bits = -bits;
1156
1157 return bits;
1158}
1159
1160/*
1161 * This function unpacks all of the superblock/macroblock/fragment coding
1162 * information from the bitstream.
1163 */
892fc83e 1164static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
1165{
1166 int bit = 0;
1167 int current_superblock = 0;
1168 int current_run = 0;
1169 int decode_fully_flags = 0;
1170 int decode_partial_blocks = 0;
22493ab9 1171 int first_c_fragment_seen;
d86053a4
MM
1172
1173 int i, j;
1174 int current_fragment;
1175
1176 debug_vp3(" vp3: unpacking superblock coding\n");
1177
1178 if (s->keyframe) {
1179
1180 debug_vp3(" keyframe-- all superblocks are fully coded\n");
1181 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
1182
1183 } else {
1184
1185 /* unpack the list of partially-coded superblocks */
1186 bit = get_bits(gb, 1);
1187 /* toggle the bit because as soon as the first run length is
1188 * fetched the bit will be toggled again */
1189 bit ^= 1;
1190 while (current_superblock < s->superblock_count) {
b5da3635 1191 if (current_run-- == 0) {
d86053a4 1192 bit ^= 1;
d8278bab
MM
1193#if 1
1194 current_run = get_vlc2(gb,
b5da3635
MN
1195 s->superblock_run_length_vlc.table, 6, 2);
1196 if (current_run == 33)
d8278bab
MM
1197 current_run += get_bits(gb, 12);
1198#else
d86053a4 1199 current_run = get_superblock_run_length(gb);
d8278bab 1200#endif
d86053a4
MM
1201 debug_block_coding(" setting superblocks %d..%d to %s\n",
1202 current_superblock,
1203 current_superblock + current_run - 1,
1204 (bit) ? "partially coded" : "not coded");
1205
1206 /* if any of the superblocks are not partially coded, flag
1207 * a boolean to decode the list of fully-coded superblocks */
642d7e84 1208 if (bit == 0) {
d86053a4 1209 decode_fully_flags = 1;
642d7e84 1210 } else {
d86053a4 1211
642d7e84
MM
1212 /* make a note of the fact that there are partially coded
1213 * superblocks */
1214 decode_partial_blocks = 1;
1215 }
d86053a4 1216 }
b5da3635 1217 s->superblock_coding[current_superblock++] = bit;
d86053a4
MM
1218 }
1219
1220 /* unpack the list of fully coded superblocks if any of the blocks were
1221 * not marked as partially coded in the previous step */
1222 if (decode_fully_flags) {
1223
1224 current_superblock = 0;
1225 current_run = 0;
1226 bit = get_bits(gb, 1);
1227 /* toggle the bit because as soon as the first run length is
1228 * fetched the bit will be toggled again */
1229 bit ^= 1;
1230 while (current_superblock < s->superblock_count) {
1231
1232 /* skip any superblocks already marked as partially coded */
1233 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
1234
b5da3635 1235 if (current_run-- == 0) {
d86053a4 1236 bit ^= 1;
d8278bab
MM
1237#if 1
1238 current_run = get_vlc2(gb,
b5da3635
MN
1239 s->superblock_run_length_vlc.table, 6, 2);
1240 if (current_run == 33)
d8278bab
MM
1241 current_run += get_bits(gb, 12);
1242#else
d86053a4 1243 current_run = get_superblock_run_length(gb);
d8278bab 1244#endif
d86053a4
MM
1245 }
1246
1247 debug_block_coding(" setting superblock %d to %s\n",
1248 current_superblock,
1249 (bit) ? "fully coded" : "not coded");
b5da3635 1250 s->superblock_coding[current_superblock] = 2*bit;
d86053a4
MM
1251 }
1252 current_superblock++;
1253 }
1254 }
1255
1256 /* if there were partial blocks, initialize bitstream for
1257 * unpacking fragment codings */
1258 if (decode_partial_blocks) {
1259
1260 current_run = 0;
1261 bit = get_bits(gb, 1);
1262 /* toggle the bit because as soon as the first run length is
1263 * fetched the bit will be toggled again */
1264 bit ^= 1;
1265 }
1266 }
1267
1268 /* figure out which fragments are coded; iterate through each
1269 * superblock (all planes) */
1270 s->coded_fragment_list_index = 0;
7beddb12 1271 s->next_coeff= s->coeffs + s->fragment_count;
04331882
MM
1272 s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
1273 s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
22493ab9 1274 first_c_fragment_seen = 0;
96a7e73b 1275 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
d86053a4
MM
1276 for (i = 0; i < s->superblock_count; i++) {
1277
1278 /* iterate through all 16 fragments in a superblock */
1279 for (j = 0; j < 16; j++) {
1280
1281 /* if the fragment is in bounds, check its coding status */
1282 current_fragment = s->superblock_fragments[i * 16 + j];
892fc83e 1283 if (current_fragment >= s->fragment_count) {
9b879566 1284 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
892fc83e
MM
1285 current_fragment, s->fragment_count);
1286 return 1;
1287 }
d86053a4
MM
1288 if (current_fragment != -1) {
1289 if (s->superblock_coding[i] == SB_NOT_CODED) {
1290
1291 /* copy all the fragments from the prior frame */
1292 s->all_fragments[current_fragment].coding_method =
1293 MODE_COPY;
1294
1295 } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
1296
1297 /* fragment may or may not be coded; this is the case
1298 * that cares about the fragment coding runs */
b5da3635 1299 if (current_run-- == 0) {
d86053a4 1300 bit ^= 1;
0ad72bdd
MM
1301#if 1
1302 current_run = get_vlc2(gb,
b5da3635 1303 s->fragment_run_length_vlc.table, 5, 2);
0ad72bdd 1304#else
d86053a4 1305 current_run = get_fragment_run_length(gb);
0ad72bdd 1306#endif
d86053a4
MM
1307 }
1308
1309 if (bit) {
22493ab9
MM
1310 /* default mode; actual mode will be decoded in
1311 * the next phase */
d86053a4
MM
1312 s->all_fragments[current_fragment].coding_method =
1313 MODE_INTER_NO_MV;
7beddb12 1314 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
04331882 1315 s->coded_fragment_list[s->coded_fragment_list_index] =
d86053a4 1316 current_fragment;
04331882 1317 if ((current_fragment >= s->u_fragment_start) &&
22493ab9
MM
1318 (s->last_coded_y_fragment == -1) &&
1319 (!first_c_fragment_seen)) {
04331882
MM
1320 s->first_coded_c_fragment = s->coded_fragment_list_index;
1321 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
22493ab9 1322 first_c_fragment_seen = 1;
04331882
MM
1323 }
1324 s->coded_fragment_list_index++;
96a7e73b 1325 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
d86053a4
MM
1326 debug_block_coding(" superblock %d is partially coded, fragment %d is coded\n",
1327 i, current_fragment);
1328 } else {
1329 /* not coded; copy this fragment from the prior frame */
1330 s->all_fragments[current_fragment].coding_method =
1331 MODE_COPY;
1332 debug_block_coding(" superblock %d is partially coded, fragment %d is not coded\n",
1333 i, current_fragment);
1334 }
1335
d86053a4
MM
1336 } else {
1337
1338 /* fragments are fully coded in this superblock; actual
1339 * coding will be determined in next step */
1340 s->all_fragments[current_fragment].coding_method =
1341 MODE_INTER_NO_MV;
7beddb12 1342 s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
04331882 1343 s->coded_fragment_list[s->coded_fragment_list_index] =
d86053a4 1344 current_fragment;
04331882 1345 if ((current_fragment >= s->u_fragment_start) &&
22493ab9
MM
1346 (s->last_coded_y_fragment == -1) &&
1347 (!first_c_fragment_seen)) {
04331882
MM
1348 s->first_coded_c_fragment = s->coded_fragment_list_index;
1349 s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
22493ab9 1350 first_c_fragment_seen = 1;
04331882
MM
1351 }
1352 s->coded_fragment_list_index++;
96a7e73b 1353 s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
d86053a4
MM
1354 debug_block_coding(" superblock %d is fully coded, fragment %d is coded\n",
1355 i, current_fragment);
1356 }
1357 }
1358 }
1359 }
04331882 1360
22493ab9
MM
1361 if (!first_c_fragment_seen)
1362 /* only Y fragments coded in this frame */
04331882 1363 s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
22493ab9 1364 else
642d7e84 1365 /* end the list of coded C fragments */
04331882 1366 s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
22493ab9 1367
04331882
MM
1368 debug_block_coding(" %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
1369 s->coded_fragment_list_index,
1370 s->first_coded_y_fragment,
1371 s->last_coded_y_fragment,
1372 s->first_coded_c_fragment,
1373 s->last_coded_c_fragment);
892fc83e
MM
1374
1375 return 0;
d86053a4
MM
1376}
1377
1378/*
1379 * This function unpacks all the coding mode data for individual macroblocks
1380 * from the bitstream.
1381 */
892fc83e 1382static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
1383{
1384 int i, j, k;
1385 int scheme;
1386 int current_macroblock;
1387 int current_fragment;
1388 int coding_mode;
1389
1390 debug_vp3(" vp3: unpacking encoding modes\n");
1391
1392 if (s->keyframe) {
1393 debug_vp3(" keyframe-- all blocks are coded as INTRA\n");
1394
1395 for (i = 0; i < s->fragment_count; i++)
1396 s->all_fragments[i].coding_method = MODE_INTRA;
1397
1398 } else {
1399
1400 /* fetch the mode coding scheme for this frame */
1401 scheme = get_bits(gb, 3);
1402 debug_modes(" using mode alphabet %d\n", scheme);
1403
1404 /* is it a custom coding scheme? */
1405 if (scheme == 0) {
1406 debug_modes(" custom mode alphabet ahead:\n");
1407 for (i = 0; i < 8; i++)
a466e345 1408 ModeAlphabet[scheme][get_bits(gb, 3)] = i;
d86053a4
MM
1409 }
1410
1411 for (i = 0; i < 8; i++)
1412 debug_modes(" mode[%d][%d] = %d\n", scheme, i,
1413 ModeAlphabet[scheme][i]);
1414
1415 /* iterate through all of the macroblocks that contain 1 or more
1416 * coded fragments */
1417 for (i = 0; i < s->u_superblock_start; i++) {
1418
1419 for (j = 0; j < 4; j++) {
1420 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1421 if ((current_macroblock == -1) ||
96a7e73b 1422 (s->macroblock_coding[current_macroblock] == MODE_COPY))
d86053a4 1423 continue;
892fc83e 1424 if (current_macroblock >= s->macroblock_count) {
9b879566 1425 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
892fc83e
MM
1426 current_macroblock, s->macroblock_count);
1427 return 1;
1428 }
d86053a4
MM
1429
1430 /* mode 7 means get 3 bits for each coding mode */
1431 if (scheme == 7)
1432 coding_mode = get_bits(gb, 3);
1433 else
0ad72bdd
MM
1434{
1435#if 1
1436 coding_mode = ModeAlphabet[scheme]
1437 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
1438#else
d86053a4 1439 coding_mode = ModeAlphabet[scheme][get_mode_code(gb)];
0ad72bdd
MM
1440#endif
1441}
d86053a4 1442
96a7e73b 1443 s->macroblock_coding[current_macroblock] = coding_mode;
d86053a4
MM
1444 for (k = 0; k < 6; k++) {
1445 current_fragment =
1446 s->macroblock_fragments[current_macroblock * 6 + k];
892fc83e
MM
1447 if (current_fragment == -1)
1448 continue;
1449 if (current_fragment >= s->fragment_count) {
9b879566 1450 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
892fc83e
MM
1451 current_fragment, s->fragment_count);
1452 return 1;
1453 }
d86053a4
MM
1454 if (s->all_fragments[current_fragment].coding_method !=
1455 MODE_COPY)
1456 s->all_fragments[current_fragment].coding_method =
1457 coding_mode;
1458 }
1459
1460 debug_modes(" coding method for macroblock starting @ fragment %d = %d\n",
1461 s->macroblock_fragments[current_macroblock * 6], coding_mode);
1462 }
1463 }
1464 }
892fc83e
MM
1465
1466 return 0;
44ae98dd
MM
1467}
1468
1469/*
d86053a4
MM
1470 * This function unpacks all the motion vectors for the individual
1471 * macroblocks from the bitstream.
1472 */
892fc83e 1473static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
1474{
1475 int i, j, k;
1476 int coding_mode;
1477 int motion_x[6];
1478 int motion_y[6];
1479 int last_motion_x = 0;
1480 int last_motion_y = 0;
1481 int prior_last_motion_x = 0;
1482 int prior_last_motion_y = 0;
1483 int current_macroblock;
1484 int current_fragment;
1485
1486 debug_vp3(" vp3: unpacking motion vectors\n");
d86053a4
MM
1487 if (s->keyframe) {
1488
1489 debug_vp3(" keyframe-- there are no motion vectors\n");
1490
1491 } else {
1492
1493 memset(motion_x, 0, 6 * sizeof(int));
1494 memset(motion_y, 0, 6 * sizeof(int));
1495
1496 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
1497 coding_mode = get_bits(gb, 1);
1498 debug_vectors(" using %s scheme for unpacking motion vectors\n",
1499 (coding_mode == 0) ? "VLC" : "fixed-length");
1500
1501 /* iterate through all of the macroblocks that contain 1 or more
1502 * coded fragments */
1503 for (i = 0; i < s->u_superblock_start; i++) {
1504
1505 for (j = 0; j < 4; j++) {
1506 current_macroblock = s->superblock_macroblocks[i * 4 + j];
1507 if ((current_macroblock == -1) ||
96a7e73b 1508 (s->macroblock_coding[current_macroblock] == MODE_COPY))
d86053a4 1509 continue;
892fc83e 1510 if (current_macroblock >= s->macroblock_count) {
9b879566 1511 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
892fc83e
MM
1512 current_macroblock, s->macroblock_count);
1513 return 1;
1514 }
d86053a4
MM
1515
1516 current_fragment = s->macroblock_fragments[current_macroblock * 6];
892fc83e 1517 if (current_fragment >= s->fragment_count) {
9b879566 1518 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
892fc83e
MM
1519 current_fragment, s->fragment_count);
1520 return 1;
1521 }
96a7e73b 1522 switch (s->macroblock_coding[current_macroblock]) {
d86053a4
MM
1523
1524 case MODE_INTER_PLUS_MV:
1525 case MODE_GOLDEN_MV:
1526 /* all 6 fragments use the same motion vector */
1527 if (coding_mode == 0) {
f4daf33e 1528#if 1
0ad72bdd
MM
1529 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1530 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1531#else
d86053a4
MM
1532 motion_x[0] = get_motion_vector_vlc(gb);
1533 motion_y[0] = get_motion_vector_vlc(gb);
0ad72bdd 1534#endif
d86053a4
MM
1535 } else {
1536 motion_x[0] = get_motion_vector_fixed(gb);
1537 motion_y[0] = get_motion_vector_fixed(gb);
1538 }
1539 for (k = 1; k < 6; k++) {
1540 motion_x[k] = motion_x[0];
1541 motion_y[k] = motion_y[0];
1542 }
1543
1544 /* vector maintenance, only on MODE_INTER_PLUS_MV */
642d7e84 1545 if (s->macroblock_coding[current_macroblock] ==
d86053a4
MM
1546 MODE_INTER_PLUS_MV) {
1547 prior_last_motion_x = last_motion_x;
1548 prior_last_motion_y = last_motion_y;
1549 last_motion_x = motion_x[0];
1550 last_motion_y = motion_y[0];
1551 }
1552 break;
1553
1554 case MODE_INTER_FOURMV:
1555 /* fetch 4 vectors from the bitstream, one for each
1556 * Y fragment, then average for the C fragment vectors */
1557 motion_x[4] = motion_y[4] = 0;
1558 for (k = 0; k < 4; k++) {
1559 if (coding_mode == 0) {
f4daf33e
MM
1560#if 1
1561 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1562 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
0ad72bdd 1563#else
d86053a4
MM
1564 motion_x[k] = get_motion_vector_vlc(gb);
1565 motion_y[k] = get_motion_vector_vlc(gb);
0ad72bdd 1566#endif
d86053a4
MM
1567 } else {
1568 motion_x[k] = get_motion_vector_fixed(gb);
1569 motion_y[k] = get_motion_vector_fixed(gb);
1570 }
1571 motion_x[4] += motion_x[k];
1572 motion_y[4] += motion_y[k];
1573 }
1574
1575 if (motion_x[4] >= 0)
1576 motion_x[4] = (motion_x[4] + 2) / 4;
1577 else
1578 motion_x[4] = (motion_x[4] - 2) / 4;
1579 motion_x[5] = motion_x[4];
1580
1581 if (motion_y[4] >= 0)
1582 motion_y[4] = (motion_y[4] + 2) / 4;
1583 else
1584 motion_y[4] = (motion_y[4] - 2) / 4;
1585 motion_y[5] = motion_y[4];
1586
1587 /* vector maintenance; vector[3] is treated as the
1588 * last vector in this case */
1589 prior_last_motion_x = last_motion_x;
1590 prior_last_motion_y = last_motion_y;
1591 last_motion_x = motion_x[3];
1592 last_motion_y = motion_y[3];
1593 break;
1594
1595 case MODE_INTER_LAST_MV:
1596 /* all 6 fragments use the last motion vector */
1597 motion_x[0] = last_motion_x;
1598 motion_y[0] = last_motion_y;
1599 for (k = 1; k < 6; k++) {
1600 motion_x[k] = motion_x[0];
1601 motion_y[k] = motion_y[0];
1602 }
1603
1604 /* no vector maintenance (last vector remains the
1605 * last vector) */
1606 break;
1607
1608 case MODE_INTER_PRIOR_LAST:
1609 /* all 6 fragments use the motion vector prior to the
1610 * last motion vector */
1611 motion_x[0] = prior_last_motion_x;
1612 motion_y[0] = prior_last_motion_y;
1613 for (k = 1; k < 6; k++) {
1614 motion_x[k] = motion_x[0];
1615 motion_y[k] = motion_y[0];
1616 }
1617
1618 /* vector maintenance */
1619 prior_last_motion_x = last_motion_x;
1620 prior_last_motion_y = last_motion_y;
1621 last_motion_x = motion_x[0];
1622 last_motion_y = motion_y[0];
1623 break;
44ae98dd
MM
1624
1625 default:
1626 /* covers intra, inter without MV, golden without MV */
1627 memset(motion_x, 0, 6 * sizeof(int));
1628 memset(motion_y, 0, 6 * sizeof(int));
1629
1630 /* no vector maintenance */
1631 break;
d86053a4
MM
1632 }
1633
1634 /* assign the motion vectors to the correct fragments */
1635 debug_vectors(" vectors for macroblock starting @ fragment %d (coding method %d):\n",
1636 current_fragment,
642d7e84 1637 s->macroblock_coding[current_macroblock]);
d86053a4
MM
1638 for (k = 0; k < 6; k++) {
1639 current_fragment =
1640 s->macroblock_fragments[current_macroblock * 6 + k];
892fc83e
MM
1641 if (current_fragment == -1)
1642 continue;
1643 if (current_fragment >= s->fragment_count) {
9b879566 1644 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
892fc83e
MM
1645 current_fragment, s->fragment_count);
1646 return 1;
1647 }
d86053a4 1648 s->all_fragments[current_fragment].motion_x = motion_x[k];
44ae98dd 1649 s->all_fragments[current_fragment].motion_y = motion_y[k];
642d7e84
MM
1650 debug_vectors(" vector %d: fragment %d = (%d, %d)\n",
1651 k, current_fragment, motion_x[k], motion_y[k]);
d86053a4
MM
1652 }
1653 }
1654 }
1655 }
892fc83e
MM
1656
1657 return 0;
d86053a4
MM
1658}
1659
1660/*
1661 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1662 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1663 * data. This function unpacks all the VLCs for either the Y plane or both
1664 * C planes, and is called for DC coefficients or different AC coefficient
1665 * levels (since different coefficient types require different VLC tables.
1666 *
1667 * This function returns a residual eob run. E.g, if a particular token gave
1668 * instructions to EOB the next 5 fragments and there were only 2 fragments
1669 * left in the current fragment range, 3 would be returned so that it could
1670 * be passed into the next call to this same function.
1671 */
1672static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1673 VLC *table, int coeff_index,
1674 int first_fragment, int last_fragment,
1675 int eob_run)
1676{
1677 int i;
1678 int token;
d3076955
MM
1679 int zero_run = 0;
1680 DCTELEM coeff = 0;
d86053a4 1681 Vp3Fragment *fragment;
36af0c95 1682 uint8_t *perm= s->scantable.permutated;
d3076955 1683 int bits_to_get;
d86053a4 1684
22493ab9 1685 if ((first_fragment >= s->fragment_count) ||
74c0ac12
MM
1686 (last_fragment >= s->fragment_count)) {
1687
9b879566 1688 av_log(s->avctx, AV_LOG_ERROR, " vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
74c0ac12 1689 first_fragment, last_fragment);
22493ab9 1690 return 0;
74c0ac12
MM
1691 }
1692
04331882 1693 for (i = first_fragment; i <= last_fragment; i++) {
d86053a4
MM
1694
1695 fragment = &s->all_fragments[s->coded_fragment_list[i]];
1696 if (fragment->coeff_count > coeff_index)
1697 continue;
1698
1699 if (!eob_run) {
1700 /* decode a VLC into a token */
1701 token = get_vlc2(gb, table->table, 5, 3);
1702 debug_vlc(" token = %2d, ", token);
1703 /* use the token to get a zero run, a coefficient, and an eob run */
d3076955
MM
1704#if 1
1705 if (token <= 6) {
1706 eob_run = eob_run_base[token];
1707 if (eob_run_get_bits[token])
1708 eob_run += get_bits(gb, eob_run_get_bits[token]);
1709 coeff = zero_run = 0;
1710 } else {
1711 bits_to_get = coeff_get_bits[token];
1712 if (!bits_to_get)
1713 coeff = coeff_tables[token][0];
1714 else
1715 coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1716
1717 zero_run = zero_run_base[token];
1718 if (zero_run_get_bits[token])
1719 zero_run += get_bits(gb, zero_run_get_bits[token]);
1720 }
1721#else
d86053a4 1722 unpack_token(gb, token, &zero_run, &coeff, &eob_run);
d3076955 1723#endif
d86053a4
MM
1724 }
1725
1726 if (!eob_run) {
1727 fragment->coeff_count += zero_run;
7beddb12
MN
1728 if (fragment->coeff_count < 64){
1729 fragment->next_coeff->coeff= coeff;
1730 fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1731 fragment->next_coeff->next= s->next_coeff;
1732 s->next_coeff->next=NULL;
1733 fragment->next_coeff= s->next_coeff++;
1734 }
d86053a4 1735 debug_vlc(" fragment %d coeff = %d\n",
7beddb12 1736 s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
d86053a4 1737 } else {
7beddb12 1738 fragment->coeff_count |= 128;
d86053a4 1739 debug_vlc(" fragment %d eob with %d coefficients\n",
7beddb12 1740 s->coded_fragment_list[i], fragment->coeff_count&127);
d86053a4
MM
1741 eob_run--;
1742 }
1743 }
1744
1745 return eob_run;
1746}
1747
1748/*
1749 * This function unpacks all of the DCT coefficient data from the
1750 * bitstream.
1751 */
892fc83e 1752static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
1753{
1754 int i;
1755 int dc_y_table;
1756 int dc_c_table;
1757 int ac_y_table;
1758 int ac_c_table;
1759 int residual_eob_run = 0;
1760
d86053a4
MM
1761 /* fetch the DC table indices */
1762 dc_y_table = get_bits(gb, 4);
1763 dc_c_table = get_bits(gb, 4);
1764
1765 /* unpack the Y plane DC coefficients */
1766 debug_vp3(" vp3: unpacking Y plane DC coefficients using table %d\n",
1767 dc_y_table);
1768 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
04331882 1769 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
d86053a4
MM
1770
1771 /* unpack the C plane DC coefficients */
1772 debug_vp3(" vp3: unpacking C plane DC coefficients using table %d\n",
1773 dc_c_table);
1774 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
04331882 1775 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
d86053a4 1776
a466e345 1777 /* fetch the AC table indices */
d86053a4
MM
1778 ac_y_table = get_bits(gb, 4);
1779 ac_c_table = get_bits(gb, 4);
1780
a466e345 1781 /* unpack the group 1 AC coefficients (coeffs 1-5) */
d86053a4
MM
1782 for (i = 1; i <= 5; i++) {
1783
1784 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1785 i, ac_y_table);
1786 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
04331882 1787 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
d86053a4
MM
1788
1789 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1790 i, ac_c_table);
1791 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
04331882 1792 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
d86053a4
MM
1793 }
1794
a466e345 1795 /* unpack the group 2 AC coefficients (coeffs 6-14) */
d86053a4
MM
1796 for (i = 6; i <= 14; i++) {
1797
1798 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1799 i, ac_y_table);
1800 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
04331882 1801 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
d86053a4
MM
1802
1803 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1804 i, ac_c_table);
1805 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
04331882 1806 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
d86053a4
MM
1807 }
1808
a466e345 1809 /* unpack the group 3 AC coefficients (coeffs 15-27) */
d86053a4
MM
1810 for (i = 15; i <= 27; i++) {
1811
1812 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1813 i, ac_y_table);
1814 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
04331882 1815 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
d86053a4
MM
1816
1817 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1818 i, ac_c_table);
1819 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
04331882 1820 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
d86053a4
MM
1821 }
1822
a466e345 1823 /* unpack the group 4 AC coefficients (coeffs 28-63) */
d86053a4
MM
1824 for (i = 28; i <= 63; i++) {
1825
1826 debug_vp3(" vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1827 i, ac_y_table);
1828 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
04331882 1829 s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
d86053a4
MM
1830
1831 debug_vp3(" vp3: unpacking level %d C plane AC coefficients using table %d\n",
1832 i, ac_c_table);
1833 residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
04331882 1834 s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
d86053a4 1835 }
892fc83e
MM
1836
1837 return 0;
d86053a4
MM
1838}
1839
1840/*
1841 * This function reverses the DC prediction for each coded fragment in
1842 * the frame. Much of this function is adapted directly from the original
1843 * VP3 source code.
1844 */
1845#define COMPATIBLE_FRAME(x) \
1846 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1847#define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
7beddb12 1848#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
d86053a4
MM
1849static inline int iabs (int x) { return ((x < 0) ? -x : x); }
1850
1851static void reverse_dc_prediction(Vp3DecodeContext *s,
1852 int first_fragment,
1853 int fragment_width,
1854 int fragment_height)
1855{
1856
1857#define PUL 8
1858#define PU 4
1859#define PUR 2
1860#define PL 1
1861
1862 int x, y;
1863 int i = first_fragment;
1864
1865 /*
1866 * Fragment prediction groups:
1867 *
1868 * 32222222226
1869 * 10000000004
1870 * 10000000004
1871 * 10000000004
1872 * 10000000004
1873 *
1874 * Note: Groups 5 and 7 do not exist as it would mean that the
1875 * fragment's x coordinate is both 0 and (width - 1) at the same time.
1876 */
1877 int predictor_group;
1878 short predicted_dc;
1879
1880 /* validity flags for the left, up-left, up, and up-right fragments */
1881 int fl, ful, fu, fur;
1882
1883 /* DC values for the left, up-left, up, and up-right fragments */
1884 int vl, vul, vu, vur;
1885
1886 /* indices for the left, up-left, up, and up-right fragments */
1887 int l, ul, u, ur;
1888
1889 /*
1890 * The 6 fields mean:
1891 * 0: up-left multiplier
1892 * 1: up multiplier
1893 * 2: up-right multiplier
1894 * 3: left multiplier
1895 * 4: mask
1896 * 5: right bit shift divisor (e.g., 7 means >>=7, a.k.a. div by 128)
1897 */
1898 int predictor_transform[16][6] = {
1899 { 0, 0, 0, 0, 0, 0 },
1900 { 0, 0, 0, 1, 0, 0 }, // PL
1901 { 0, 0, 1, 0, 0, 0 }, // PUR
1902 { 0, 0, 53, 75, 127, 7 }, // PUR|PL
1903 { 0, 1, 0, 0, 0, 0 }, // PU
1904 { 0, 1, 0, 1, 1, 1 }, // PU|PL
1905 { 0, 1, 0, 0, 0, 0 }, // PU|PUR
1906 { 0, 0, 53, 75, 127, 7 }, // PU|PUR|PL
1907 { 1, 0, 0, 0, 0, 0 }, // PUL
1908 { 0, 0, 0, 1, 0, 0 }, // PUL|PL
1909 { 1, 0, 1, 0, 1, 1 }, // PUL|PUR
1910 { 0, 0, 53, 75, 127, 7 }, // PUL|PUR|PL
1911 { 0, 1, 0, 0, 0, 0 }, // PUL|PU
1912 {-26, 29, 0, 29, 31, 5 }, // PUL|PU|PL
1913 { 3, 10, 3, 0, 15, 4 }, // PUL|PU|PUR
1914 {-26, 29, 0, 29, 31, 5 } // PUL|PU|PUR|PL
1915 };
1916
1917 /* This table shows which types of blocks can use other blocks for
1918 * prediction. For example, INTRA is the only mode in this table to
1919 * have a frame number of 0. That means INTRA blocks can only predict
1920 * from other INTRA blocks. There are 2 golden frame coding types;
1921 * blocks encoding in these modes can only predict from other blocks
1922 * that were encoded with these 1 of these 2 modes. */
1923 unsigned char compatible_frame[8] = {
1924 1, /* MODE_INTER_NO_MV */
1925 0, /* MODE_INTRA */
1926 1, /* MODE_INTER_PLUS_MV */
1927 1, /* MODE_INTER_LAST_MV */
1928 1, /* MODE_INTER_PRIOR_MV */
1929 2, /* MODE_USING_GOLDEN */
1930 2, /* MODE_GOLDEN_MV */
1931 1 /* MODE_INTER_FOUR_MV */
1932 };
1933 int current_frame_type;
1934
1935 /* there is a last DC predictor for each of the 3 frame types */
1936 short last_dc[3];
1937
1938 int transform = 0;
1939
1940 debug_vp3(" vp3: reversing DC prediction\n");
1941
1942 vul = vu = vur = vl = 0;
1943 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1944
1945 /* for each fragment row... */
1946 for (y = 0; y < fragment_height; y++) {
1947
1948 /* for each fragment in a row... */
1949 for (x = 0; x < fragment_width; x++, i++) {
1950
1951 /* reverse prediction if this block was coded */
1952 if (s->all_fragments[i].coding_method != MODE_COPY) {
1953
1954 current_frame_type =
1955 compatible_frame[s->all_fragments[i].coding_method];
1956 predictor_group = (x == 0) + ((y == 0) << 1) +
1957 ((x + 1 == fragment_width) << 2);
1958 debug_dc_pred(" frag %d: group %d, orig DC = %d, ",
7beddb12 1959 i, predictor_group, DC_COEFF(i));
d86053a4
MM
1960
1961 switch (predictor_group) {
1962
1963 case 0:
1964 /* main body of fragments; consider all 4 possible
1965 * fragments for prediction */
1966
1967 /* calculate the indices of the predicting fragments */
1968 ul = i - fragment_width - 1;
1969 u = i - fragment_width;
1970 ur = i - fragment_width + 1;
1971 l = i - 1;
1972
1973 /* fetch the DC values for the predicting fragments */
7beddb12
MN
1974 vul = DC_COEFF(ul);
1975 vu = DC_COEFF(u);
1976 vur = DC_COEFF(ur);
1977 vl = DC_COEFF(l);
d86053a4
MM
1978
1979 /* figure out which fragments are valid */
1980 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1981 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1982 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1983 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1984
1985 /* decide which predictor transform to use */
1986 transform = (fl*PL) | (fu*PU) | (ful*PUL) | (fur*PUR);
1987
1988 break;
1989
1990 case 1:
1991 /* left column of fragments, not including top corner;
1992 * only consider up and up-right fragments */
1993
1994 /* calculate the indices of the predicting fragments */
1995 u = i - fragment_width;
1996 ur = i - fragment_width + 1;
1997
1998 /* fetch the DC values for the predicting fragments */
7beddb12
MN
1999 vu = DC_COEFF(u);
2000 vur = DC_COEFF(ur);
d86053a4
MM
2001
2002 /* figure out which fragments are valid */
2003 fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
2004 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
2005
2006 /* decide which predictor transform to use */
2007 transform = (fu*PU) | (fur*PUR);
2008
2009 break;
2010
2011 case 2:
2012 case 6:
2013 /* top row of fragments, not including top-left frag;
2014 * only consider the left fragment for prediction */
2015
2016 /* calculate the indices of the predicting fragments */
2017 l = i - 1;
2018
2019 /* fetch the DC values for the predicting fragments */
7beddb12 2020 vl = DC_COEFF(l);
d86053a4
MM
2021
2022 /* figure out which fragments are valid */
2023 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
2024
2025 /* decide which predictor transform to use */
2026 transform = (fl*PL);
2027
2028 break;
2029
2030 case 3:
2031 /* top-left fragment */
2032
2033 /* nothing to predict from in this case */
2034 transform = 0;
2035
2036 break;
2037
2038 case 4:
2039 /* right column of fragments, not including top corner;
2040 * consider up-left, up, and left fragments for
2041 * prediction */
2042
2043 /* calculate the indices of the predicting fragments */
2044 ul = i - fragment_width - 1;
2045 u = i - fragment_width;
2046 l = i - 1;
2047
2048 /* fetch the DC values for the predicting fragments */
7beddb12
MN
2049 vul = DC_COEFF(ul);
2050 vu = DC_COEFF(u);
2051 vl = DC_COEFF(l);
d86053a4
MM
2052
2053 /* figure out which fragments are valid */
2054 ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
2055 fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
2056 fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
2057
2058 /* decide which predictor transform to use */
2059 transform = (fl*PL) | (fu*PU) | (ful*PUL);
2060
2061 break;
2062
2063 }
2064
2065 debug_dc_pred("transform = %d, ", transform);
2066
2067 if (transform == 0) {
2068
2069 /* if there were no fragments to predict from, use last
2070 * DC saved */
7beddb12 2071 predicted_dc = last_dc[current_frame_type];
d86053a4 2072 debug_dc_pred("from last DC (%d) = %d\n",
7beddb12 2073 current_frame_type, DC_COEFF(i));
d86053a4
MM
2074
2075 } else {
2076
2077 /* apply the appropriate predictor transform */
2078 predicted_dc =
2079 (predictor_transform[transform][0] * vul) +
2080 (predictor_transform[transform][1] * vu) +
2081 (predictor_transform[transform][2] * vur) +
2082 (predictor_transform[transform][3] * vl);
2083
2084 /* if there is a shift value in the transform, add
2085 * the sign bit before the shift */
2086 if (predictor_transform[transform][5] != 0) {
2087 predicted_dc += ((predicted_dc >> 15) &
2088 predictor_transform[transform][4]);
2089 predicted_dc >>= predictor_transform[transform][5];
2090 }
2091
2092 /* check for outranging on the [ul u l] and
2093 * [ul u ur l] predictors */
2094 if ((transform == 13) || (transform == 15)) {
2095 if (iabs(predicted_dc - vu) > 128)
2096 predicted_dc = vu;
2097 else if (iabs(predicted_dc - vl) > 128)
2098 predicted_dc = vl;
2099 else if (iabs(predicted_dc - vul) > 128)
2100 predicted_dc = vul;
2101 }
2102
d86053a4 2103 debug_dc_pred("from pred DC = %d\n",
7beddb12 2104 DC_COEFF(i));
d86053a4
MM
2105 }
2106
7beddb12
MN
2107 /* at long last, apply the predictor */
2108 if(s->coeffs[i].index){
2109 *s->next_coeff= s->coeffs[i];
2110 s->coeffs[i].index=0;
2111 s->coeffs[i].coeff=0;
2112 s->coeffs[i].next= s->next_coeff++;
2113 }
2114 s->coeffs[i].coeff += predicted_dc;
d86053a4 2115 /* save the DC */
7beddb12
MN
2116 last_dc[current_frame_type] = DC_COEFF(i);
2117 if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
2118 s->all_fragments[i].coeff_count= 129;
2119// s->all_fragments[i].next_coeff= s->next_coeff;
2120 s->coeffs[i].next= s->next_coeff;
2121 (s->next_coeff++)->next=NULL;
2122 }
d86053a4
MM
2123 }
2124 }
2125 }
2126}
2127
2128/*
2129 * This function performs the final rendering of each fragment's data
2130 * onto the output frame.
2131 */
2132static void render_fragments(Vp3DecodeContext *s,
2133 int first_fragment,
44ae98dd
MM
2134 int width,
2135 int height,
d86053a4
MM
2136 int plane /* 0 = Y, 1 = U, 2 = V */)
2137{
8b6103da 2138 int x, y, j;
d86053a4
MM
2139 int m, n;
2140 int i = first_fragment;
d86053a4 2141 int16_t *dequantizer;
7beddb12 2142 DCTELEM __align16 block[64];
d86053a4
MM
2143 unsigned char *output_plane;
2144 unsigned char *last_plane;
2145 unsigned char *golden_plane;
2146 int stride;
4df8ca9d 2147 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
a466e345 2148 int upper_motion_limit, lower_motion_limit;
44ae98dd 2149 int motion_halfpel_index;
a2f11b3c 2150 uint8_t *motion_source;
d86053a4
MM
2151
2152 debug_vp3(" vp3: rendering final fragments for %s\n",
2153 (plane == 0) ? "Y plane" : (plane == 1) ? "U plane" : "V plane");
2154
2155 /* set up plane-specific parameters */
2156 if (plane == 0) {
d86053a4 2157 output_plane = s->current_frame.data[0];
61873c4a
MM
2158 last_plane = s->last_frame.data[0];
2159 golden_plane = s->golden_frame.data[0];
9a7ad925
AB
2160 stride = s->current_frame.linesize[0];
2161 if (!s->flipped_image) stride = -stride;
a466e345
MM
2162 upper_motion_limit = 7 * s->current_frame.linesize[0];
2163 lower_motion_limit = height * s->current_frame.linesize[0] + width - 8;
d86053a4 2164 } else if (plane == 1) {
d86053a4 2165 output_plane = s->current_frame.data[1];
61873c4a
MM
2166 last_plane = s->last_frame.data[1];
2167 golden_plane = s->golden_frame.data[1];
9a7ad925
AB
2168 stride = s->current_frame.linesize[1];
2169 if (!s->flipped_image) stride = -stride;
a466e345
MM
2170 upper_motion_limit = 7 * s->current_frame.linesize[1];
2171 lower_motion_limit = height * s->current_frame.linesize[1] + width - 8;
d86053a4 2172 } else {
d86053a4 2173 output_plane = s->current_frame.data[2];
61873c4a
MM
2174 last_plane = s->last_frame.data[2];
2175 golden_plane = s->golden_frame.data[2];
9a7ad925
AB
2176 stride = s->current_frame.linesize[2];
2177 if (!s->flipped_image) stride = -stride;
a466e345
MM
2178 upper_motion_limit = 7 * s->current_frame.linesize[2];
2179 lower_motion_limit = height * s->current_frame.linesize[2] + width - 8;
d86053a4 2180 }
0ecca7a4 2181
edaf11e8 2182 if(ABS(stride) > 2048)
0ecca7a4 2183 return; //various tables are fixed size
d86053a4
MM
2184
2185 /* for each fragment row... */
44ae98dd 2186 for (y = 0; y < height; y += 8) {
d86053a4
MM
2187
2188 /* for each fragment in a row... */
44ae98dd 2189 for (x = 0; x < width; x += 8, i++) {
d86053a4 2190
74c0ac12 2191 if ((i < 0) || (i >= s->fragment_count)) {
9b879566 2192 av_log(s->avctx, AV_LOG_ERROR, " vp3:render_fragments(): bad fragment number (%d)\n", i);
74c0ac12
MM
2193 return;
2194 }
2195
d86053a4 2196 /* transform if this block was coded */
f8830383
AB
2197 if ((s->all_fragments[i].coding_method != MODE_COPY) &&
2198 !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
44ae98dd 2199
a2f11b3c
MN
2200 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
2201 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
2202 motion_source= golden_plane;
2203 else
2204 motion_source= last_plane;
2205
2206 motion_source += s->all_fragments[i].first_pixel;
642d7e84
MM
2207 motion_halfpel_index = 0;
2208
2209 /* sort out the motion vector if this fragment is coded
2210 * using a motion vector method */
2211 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
2212 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
a2f11b3c 2213 int src_x, src_y;
642d7e84
MM
2214 motion_x = s->all_fragments[i].motion_x;
2215 motion_y = s->all_fragments[i].motion_y;
b928ec64
MN
2216 if(plane){
2217 motion_x= (motion_x>>1) | (motion_x&1);
2218 motion_y= (motion_y>>1) | (motion_y&1);
2219 }
2220
a2f11b3c
MN
2221 src_x= (motion_x>>1) + x;
2222 src_y= (motion_y>>1) + y;
642d7e84 2223if ((motion_x == 0xbeef) || (motion_y == 0xbeef))
9b879566 2224av_log(s->avctx, AV_LOG_ERROR, " help! got beefy vector! (%X, %X)\n", motion_x, motion_y);
642d7e84 2225
b928ec64
MN
2226 motion_halfpel_index = motion_x & 0x01;
2227 motion_source += (motion_x >> 1);
642d7e84
MM
2228
2229// motion_y = -motion_y;
b928ec64
MN
2230 motion_halfpel_index |= (motion_y & 0x01) << 1;
2231 motion_source += ((motion_y >> 1) * stride);
642d7e84 2232
a2f11b3c
MN
2233 if(src_x<0 || src_y<0 || src_x + 9 >= width || src_y + 9 >= height){
2234 uint8_t *temp= s->edge_emu_buffer;
2235 if(stride<0) temp -= 9*stride;
9a7ad925 2236 else temp += 9*stride;
a2f11b3c
MN
2237
2238 ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, width, height);
2239 motion_source= temp;
642d7e84 2240 }
a466e345 2241 }
c0a0170c 2242
a466e345 2243
44ae98dd
MM
2244 /* first, take care of copying a block from either the
2245 * previous or the golden frame */
44ae98dd 2246 if (s->all_fragments[i].coding_method != MODE_INTRA) {
c0a0170c
MN
2247 //Note, it is possible to implement all MC cases with put_no_rnd_pixels_l2 which would look more like the VP3 source but this would be slower as put_no_rnd_pixels_tab is better optimzed
2248 if(motion_halfpel_index != 3){
2249 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
2250 output_plane + s->all_fragments[i].first_pixel,
2251 motion_source, stride, 8);
2252 }else{
2253 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
2254 s->dsp.put_no_rnd_pixels_l2[1](
2255 output_plane + s->all_fragments[i].first_pixel,
2256 motion_source - d,
2257 motion_source + stride + 1 + d,
2258 stride, 8);
2259 }
7bad42cf
MN
2260 dequantizer = s->inter_dequant;
2261 }else{
2262 if (plane == 0)
2263 dequantizer = s->intra_y_dequant;
2264 else
2265 dequantizer = s->intra_c_dequant;
44ae98dd
MM
2266 }
2267
d86053a4 2268 /* dequantize the DCT coefficients */
44ae98dd
MM
2269 debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
2270 i, s->all_fragments[i].coding_method,
7beddb12 2271 DC_COEFF(i), dequantizer[0]);
d86053a4 2272
8b6103da 2273 if(s->avctx->idct_algo==FF_IDCT_VP3){
7beddb12
MN
2274 Coeff *coeff= s->coeffs + i;
2275 memset(block, 0, sizeof(block));
2276 while(coeff->next){
2277 block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
2278 coeff= coeff->next;
8b6103da
MN
2279 }
2280 }else{
7beddb12
MN
2281 Coeff *coeff= s->coeffs + i;
2282 memset(block, 0, sizeof(block));
2283 while(coeff->next){
2284 block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
2285 coeff= coeff->next;
8b6103da
MN
2286 }
2287 }
2288
4e80eb21 2289 /* invert DCT and place (or add) in final output */
8b6103da 2290
4e80eb21 2291 if (s->all_fragments[i].coding_method == MODE_INTRA) {
8b6103da 2292 if(s->avctx->idct_algo!=FF_IDCT_VP3)
7beddb12 2293 block[0] += 128<<3;
8b6103da 2294 s->dsp.idct_put(
f9ed9d85 2295 output_plane + s->all_fragments[i].first_pixel,
8b6103da 2296 stride,
7beddb12 2297 block);
4e80eb21 2298 } else {
8b6103da 2299 s->dsp.idct_add(
44ae98dd 2300 output_plane + s->all_fragments[i].first_pixel,
8b6103da 2301 stride,
7beddb12 2302 block);
4e80eb21 2303 }
44ae98dd
MM
2304
2305 debug_idct("block after idct_%s():\n",
2306 (s->all_fragments[i].coding_method == MODE_INTRA)?
2307 "put" : "add");
d86053a4
MM
2308 for (m = 0; m < 8; m++) {
2309 for (n = 0; n < 8; n++) {
44ae98dd
MM
2310 debug_idct(" %3d", *(output_plane +
2311 s->all_fragments[i].first_pixel + (m * stride + n)));
d86053a4
MM
2312 }
2313 debug_idct("\n");
2314 }
2315 debug_idct("\n");
d86053a4
MM
2316
2317 } else {
2318
44ae98dd
MM
2319 /* copy directly from the previous frame */
2320 s->dsp.put_pixels_tab[1][0](
2321 output_plane + s->all_fragments[i].first_pixel,
2322 last_plane + s->all_fragments[i].first_pixel,
2323 stride, 8);
d86053a4
MM
2324
2325 }
2326 }
2327 }
2328
2329 emms_c();
a54ea19a
MM
2330}
2331
a54ea19a
MM
2332static void horizontal_filter(unsigned char *first_pixel, int stride,
2333 int *bounding_values)
2334{
6f7aa589 2335 unsigned char *end;
a54ea19a
MM
2336 int filter_value;
2337
6f7aa589 2338 for (end= first_pixel + 8*stride; first_pixel < end; first_pixel += stride) {
a54ea19a 2339 filter_value =
6f7aa589
MN
2340 (first_pixel[-2] - first_pixel[ 1])
2341 +3*(first_pixel[ 0] - first_pixel[-1]);
a54ea19a 2342 filter_value = bounding_values[(filter_value + 4) >> 3];
ee408ead
MN
2343 first_pixel[-1] = clip_uint8(first_pixel[-1] + filter_value);
2344 first_pixel[ 0] = clip_uint8(first_pixel[ 0] - filter_value);
a54ea19a
MM
2345 }
2346}
2347
2348static void vertical_filter(unsigned char *first_pixel, int stride,
2349 int *bounding_values)
2350{
6f7aa589 2351 unsigned char *end;
a54ea19a 2352 int filter_value;
6f7aa589 2353 const int nstride= -stride;
a54ea19a 2354
6f7aa589 2355 for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
a54ea19a 2356 filter_value =
6f7aa589
MN
2357 (first_pixel[2 * nstride] - first_pixel[ stride])
2358 +3*(first_pixel[0 ] - first_pixel[nstride]);
a54ea19a 2359 filter_value = bounding_values[(filter_value + 4) >> 3];
6f7aa589 2360 first_pixel[nstride] = clip_uint8(first_pixel[nstride] + filter_value);
ee408ead 2361 first_pixel[0] = clip_uint8(first_pixel[0] - filter_value);
a54ea19a
MM
2362 }
2363}
2364
2365static void apply_loop_filter(Vp3DecodeContext *s)
2366{
2367 int x, y, plane;
2368 int width, height;
2369 int fragment;
2370 int stride;
2371 unsigned char *plane_data;
d5e18835
MN
2372
2373 int bounding_values_array[256];
2374 int *bounding_values= bounding_values_array+127;
a54ea19a
MM
2375 int filter_limit;
2376
2377 /* find the right loop limit value */
2378 for (x = 63; x >= 0; x--) {
2379 if (vp31_ac_scale_factor[x] >= s->quality_index)
2380 break;
2381 }
d5e18835 2382 filter_limit = vp31_filter_limit_values[s->quality_index];
a54ea19a
MM
2383
2384 /* set up the bounding values */
d5e18835 2385 memset(bounding_values_array, 0, 256 * sizeof(int));
a54ea19a
MM
2386 for (x = 0; x < filter_limit; x++) {
2387 bounding_values[-x - filter_limit] = -filter_limit + x;
2388 bounding_values[-x] = -x;
2389 bounding_values[x] = x;
2390 bounding_values[x + filter_limit] = filter_limit - x;
2391 }
2392
2393 for (plane = 0; plane < 3; plane++) {
2394
2395 if (plane == 0) {
2396 /* Y plane parameters */
2397 fragment = 0;
2398 width = s->fragment_width;
2399 height = s->fragment_height;
2400 stride = s->current_frame.linesize[0];
2401 plane_data = s->current_frame.data[0];
2402 } else if (plane == 1) {
2403 /* U plane parameters */
2404 fragment = s->u_fragment_start;
2405 width = s->fragment_width / 2;
2406 height = s->fragment_height / 2;
2407 stride = s->current_frame.linesize[1];
2408 plane_data = s->current_frame.data[1];
2409 } else {
2410 /* V plane parameters */
2411 fragment = s->v_fragment_start;
2412 width = s->fragment_width / 2;
2413 height = s->fragment_height / 2;
2414 stride = s->current_frame.linesize[2];
2415 plane_data = s->current_frame.data[2];
2416 }
2417
2418 for (y = 0; y < height; y++) {
d86053a4 2419
a54ea19a 2420 for (x = 0; x < width; x++) {
220a6f40 2421START_TIMER
a54ea19a
MM
2422 /* do not perform left edge filter for left columns frags */
2423 if ((x > 0) &&
2424 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2425 horizontal_filter(
d5e18835 2426 plane_data + s->all_fragments[fragment].first_pixel - 7*stride,
a54ea19a
MM
2427 stride, bounding_values);
2428 }
2429
2430 /* do not perform top edge filter for top row fragments */
2431 if ((y > 0) &&
2432 (s->all_fragments[fragment].coding_method != MODE_COPY)) {
2433 vertical_filter(
d5e18835 2434 plane_data + s->all_fragments[fragment].first_pixel + stride,
a54ea19a
MM
2435 stride, bounding_values);
2436 }
2437
2438 /* do not perform right edge filter for right column
2439 * fragments or if right fragment neighbor is also coded
2440 * in this frame (it will be filtered in next iteration) */
2441 if ((x < width - 1) &&
2442 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2443 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
2444 horizontal_filter(
d5e18835 2445 plane_data + s->all_fragments[fragment + 1].first_pixel - 7*stride,
a54ea19a
MM
2446 stride, bounding_values);
2447 }
2448
2449 /* do not perform bottom edge filter for bottom row
2450 * fragments or if bottom fragment neighbor is also coded
2451 * in this frame (it will be filtered in the next row) */
2452 if ((y < height - 1) &&
2453 (s->all_fragments[fragment].coding_method != MODE_COPY) &&
2454 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
2455 vertical_filter(
d5e18835 2456 plane_data + s->all_fragments[fragment + width].first_pixel + stride,
a54ea19a
MM
2457 stride, bounding_values);
2458 }
2459
2460 fragment++;
220a6f40 2461STOP_TIMER("loop filter")
a54ea19a
MM
2462 }
2463 }
2464 }
d86053a4
MM
2465}
2466
2467/*
2468 * This function computes the first pixel addresses for each fragment.
2469 * This function needs to be invoked after the first frame is allocated
2470 * so that it has access to the plane strides.
2471 */
2472static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
2473{
2474
2475 int i, x, y;
2476
2477 /* figure out the first pixel addresses for each of the fragments */
2478 /* Y plane */
2479 i = 0;
2480 for (y = s->fragment_height; y > 0; y--) {
2481 for (x = 0; x < s->fragment_width; x++) {
2482 s->all_fragments[i++].first_pixel =
2483 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2484 s->golden_frame.linesize[0] +
2485 x * FRAGMENT_PIXELS;
2486 debug_init(" fragment %d, first pixel @ %d\n",
2487 i-1, s->all_fragments[i-1].first_pixel);
2488 }
2489 }
2490
2491 /* U plane */
2492 i = s->u_fragment_start;
2493 for (y = s->fragment_height / 2; y > 0; y--) {
2494 for (x = 0; x < s->fragment_width / 2; x++) {
2495 s->all_fragments[i++].first_pixel =
2496 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2497 s->golden_frame.linesize[1] +
2498 x * FRAGMENT_PIXELS;
2499 debug_init(" fragment %d, first pixel @ %d\n",
2500 i-1, s->all_fragments[i-1].first_pixel);
2501 }
2502 }
2503
2504 /* V plane */
2505 i = s->v_fragment_start;
2506 for (y = s->fragment_height / 2; y > 0; y--) {
2507 for (x = 0; x < s->fragment_width / 2; x++) {
2508 s->all_fragments[i++].first_pixel =
2509 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2510 s->golden_frame.linesize[2] +
2511 x * FRAGMENT_PIXELS;
2512 debug_init(" fragment %d, first pixel @ %d\n",
2513 i-1, s->all_fragments[i-1].first_pixel);
2514 }
2515 }
2516}
2517
9a7ad925
AB
2518/* FIXME: this should be merged with the above! */
2519static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
2520{
2521
2522 int i, x, y;
2523
2524 /* figure out the first pixel addresses for each of the fragments */
2525 /* Y plane */
2526 i = 0;
2527 for (y = 1; y <= s->fragment_height; y++) {
2528 for (x = 0; x < s->fragment_width; x++) {
2529 s->all_fragments[i++].first_pixel =
2530 s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2531 s->golden_frame.linesize[0] +
2532 x * FRAGMENT_PIXELS;
2533 debug_init(" fragment %d, first pixel @ %d\n",
2534 i-1, s->all_fragments[i-1].first_pixel);
2535 }
2536 }
2537
2538 /* U plane */
2539 i = s->u_fragment_start;
2540 for (y = 1; y <= s->fragment_height / 2; y++) {
2541 for (x = 0; x < s->fragment_width / 2; x++) {
2542 s->all_fragments[i++].first_pixel =
2543 s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2544 s->golden_frame.linesize[1] +
2545 x * FRAGMENT_PIXELS;
2546 debug_init(" fragment %d, first pixel @ %d\n",
2547 i-1, s->all_fragments[i-1].first_pixel);
2548 }
2549 }
2550
2551 /* V plane */
2552 i = s->v_fragment_start;
2553 for (y = 1; y <= s->fragment_height / 2; y++) {
2554 for (x = 0; x < s->fragment_width / 2; x++) {
2555 s->all_fragments[i++].first_pixel =
2556 s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2557 s->golden_frame.linesize[2] +
2558 x * FRAGMENT_PIXELS;
2559 debug_init(" fragment %d, first pixel @ %d\n",
2560 i-1, s->all_fragments[i-1].first_pixel);
2561 }
2562 }
2563}
2564
d86053a4
MM
2565/*
2566 * This is the ffmpeg/libavcodec API init function.
2567 */
2568static int vp3_decode_init(AVCodecContext *avctx)
2569{
2570 Vp3DecodeContext *s = avctx->priv_data;
2571 int i;
892fc83e
MM
2572 int c_width;
2573 int c_height;
2574 int y_superblock_count;
2575 int c_superblock_count;
d86053a4 2576
3c3f113e
AB
2577 if (avctx->codec_tag == MKTAG('V','P','3','0'))
2578 s->version = 0;
2579 else
2580 s->version = 1;
2581
d86053a4 2582 s->avctx = avctx;
642d7e84
MM
2583 s->width = (avctx->width + 15) & 0xFFFFFFF0;
2584 s->height = (avctx->height + 15) & 0xFFFFFFF0;
d86053a4
MM
2585 avctx->pix_fmt = PIX_FMT_YUV420P;
2586 avctx->has_b_frames = 0;
8b6103da
MN
2587 if(avctx->idct_algo==FF_IDCT_AUTO)
2588 avctx->idct_algo=FF_IDCT_VP3;
d86053a4 2589 dsputil_init(&s->dsp, avctx);
36af0c95
MN
2590
2591 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
d86053a4
MM
2592
2593 /* initialize to an impossible value which will force a recalculation
2594 * in the first frame decode */
2595 s->quality_index = -1;
2596
892fc83e
MM
2597 s->y_superblock_width = (s->width + 31) / 32;
2598 s->y_superblock_height = (s->height + 31) / 32;
2599 y_superblock_count = s->y_superblock_width * s->y_superblock_height;
2600
2601 /* work out the dimensions for the C planes */
2602 c_width = s->width / 2;
2603 c_height = s->height / 2;
2604 s->c_superblock_width = (c_width + 31) / 32;
2605 s->c_superblock_height = (c_height + 31) / 32;
2606 c_superblock_count = s->c_superblock_width * s->c_superblock_height;
2607
2608 s->superblock_count = y_superblock_count + (c_superblock_count * 2);
2609 s->u_superblock_start = y_superblock_count;
2610 s->v_superblock_start = s->u_superblock_start + c_superblock_count;
d86053a4
MM
2611 s->superblock_coding = av_malloc(s->superblock_count);
2612
2613 s->macroblock_width = (s->width + 15) / 16;
2614 s->macroblock_height = (s->height + 15) / 16;
2615 s->macroblock_count = s->macroblock_width * s->macroblock_height;
2616
2617 s->fragment_width = s->width / FRAGMENT_PIXELS;
2618 s->fragment_height = s->height / FRAGMENT_PIXELS;
2619
2620 /* fragment count covers all 8x8 blocks for all 3 planes */
2621 s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
2622 s->u_fragment_start = s->fragment_width * s->fragment_height;
2623 s->v_fragment_start = s->fragment_width * s->fragment_height * 5 / 4;
2624
892fc83e
MM
2625 debug_init(" Y plane: %d x %d\n", s->width, s->height);
2626 debug_init(" C plane: %d x %d\n", c_width, c_height);
2627 debug_init(" Y superblocks: %d x %d, %d total\n",
2628 s->y_superblock_width, s->y_superblock_height, y_superblock_count);
2629 debug_init(" C superblocks: %d x %d, %d total\n",
2630 s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2631 debug_init(" total superblocks = %d, U starts @ %d, V starts @ %d\n",
2632 s->superblock_count, s->u_superblock_start, s->v_superblock_start);
d86053a4
MM
2633 debug_init(" macroblocks: %d x %d, %d total\n",
2634 s->macroblock_width, s->macroblock_height, s->macroblock_count);
2635 debug_init(" %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2636 s->fragment_count,
2637 s->fragment_width,
2638 s->fragment_height,
2639 s->u_fragment_start,
2640 s->v_fragment_start);
2641
2642 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
7beddb12 2643 s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
d86053a4
MM
2644 s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2645 s->pixel_addresses_inited = 0;
2646
f44ee2c3
AB
2647 if (!s->theora_tables)
2648 {
2649 for (i = 0; i < 64; i++)
2650 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2651 for (i = 0; i < 64; i++)
67335dbc 2652 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
f44ee2c3
AB
2653 for (i = 0; i < 64; i++)
2654 s->coded_intra_y_dequant[i] = vp31_intra_y_dequant[i];
2655 for (i = 0; i < 64; i++)
2656 s->coded_intra_c_dequant[i] = vp31_intra_c_dequant[i];
2657 for (i = 0; i < 64; i++)
2658 s->coded_inter_dequant[i] = vp31_inter_dequant[i];
2659 }
2660
d86053a4
MM
2661 /* init VLC tables */
2662 for (i = 0; i < 16; i++) {
2663
96a7e73b 2664 /* DC histograms */
d86053a4
MM
2665 init_vlc(&s->dc_vlc[i], 5, 32,
2666 &dc_bias[i][0][1], 4, 2,
073c2593 2667 &dc_bias[i][0][0], 4, 2, 0);
d86053a4 2668
a466e345 2669 /* group 1 AC histograms */
d86053a4
MM
2670 init_vlc(&s->ac_vlc_1[i], 5, 32,
2671 &ac_bias_0[i][0][1], 4, 2,
073c2593 2672 &ac_bias_0[i][0][0], 4, 2, 0);
d86053a4 2673
a466e345 2674 /* group 2 AC histograms */
d86053a4
MM
2675 init_vlc(&s->ac_vlc_2[i], 5, 32,
2676 &ac_bias_1[i][0][1], 4, 2,
073c2593 2677 &ac_bias_1[i][0][0], 4, 2, 0);
d86053a4 2678
a466e345 2679 /* group 3 AC histograms */
d86053a4
MM
2680 init_vlc(&s->ac_vlc_3[i], 5, 32,
2681 &ac_bias_2[i][0][1], 4, 2,
073c2593 2682 &ac_bias_2[i][0][0], 4, 2, 0);
d86053a4 2683
a466e345 2684 /* group 4 AC histograms */
d86053a4
MM
2685 init_vlc(&s->ac_vlc_4[i], 5, 32,
2686 &ac_bias_3[i][0][1], 4, 2,
073c2593 2687 &ac_bias_3[i][0][0], 4, 2, 0);
d86053a4
MM
2688 }
2689
d8278bab
MM
2690 init_vlc(&s->superblock_run_length_vlc, 6, 34,
2691 &superblock_run_length_vlc_table[0][1], 4, 2,
2692 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2693
0ad72bdd
MM
2694 init_vlc(&s->fragment_run_length_vlc, 5, 31,
2695 &fragment_run_length_vlc_table[0][1], 4, 2,
2696 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2697
2698 init_vlc(&s->mode_code_vlc, 3, 8,
2699 &mode_code_vlc_table[0][1], 2, 1,
2700 &mode_code_vlc_table[0][0], 2, 1, 0);
2701
2702 init_vlc(&s->motion_vector_vlc, 6, 63,
2703 &motion_vector_vlc_table[0][1], 2, 1,
2704 &motion_vector_vlc_table[0][0], 2, 1, 0);
2705
d86053a4
MM
2706 /* work out the block mapping tables */
2707 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2708 s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2709 s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
96a7e73b 2710 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
d86053a4
MM
2711 init_block_mapping(s);
2712
44ae98dd
MM
2713 for (i = 0; i < 3; i++) {
2714 s->current_frame.data[i] = NULL;
2715 s->last_frame.data[i] = NULL;
2716 s->golden_frame.data[i] = NULL;
61873c4a
MM
2717 }
2718
d86053a4
MM
2719 return 0;
2720}
2721
2722/*
2723 * This is the ffmpeg/libavcodec API frame decode function.
2724 */
2725static int vp3_decode_frame(AVCodecContext *avctx,
2726 void *data, int *data_size,
2727 uint8_t *buf, int buf_size)
2728{
2729 Vp3DecodeContext *s = avctx->priv_data;
2730 GetBitContext gb;
2731 static int counter = 0;
2732
d86053a4 2733 init_get_bits(&gb, buf, buf_size * 8);
f44ee2c3
AB
2734
2735 if (s->theora && get_bits1(&gb))
2736 {
3c3f113e 2737 int ptype = get_bits(&gb, 7);
d86053a4 2738
3c3f113e
AB
2739 skip_bits(&gb, 6*8); /* "theora" */
2740
2741 switch(ptype)
9a7ad925 2742 {
3c3f113e
AB
2743 case 1:
2744 theora_decode_comments(avctx, gb);
2745 break;
2746 case 2:
2747 theora_decode_tables(avctx, gb);
2748 init_dequantizer(s);
2749 break;
2750 default:
2751 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
9a7ad925 2752 }
3c3f113e 2753 return buf_size;
f44ee2c3 2754 }
3c3f113e
AB
2755
2756 s->keyframe = !get_bits1(&gb);
2757 if (!s->theora)
f44ee2c3 2758 skip_bits(&gb, 1);
3c3f113e
AB
2759 s->last_quality_index = s->quality_index;
2760 s->quality_index = get_bits(&gb, 6);
ba7ee4a4 2761 if (s->theora >= 0x030200)
3c3f113e 2762 skip_bits1(&gb);
d86053a4 2763
f8830383
AB
2764 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2765 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2766 s->keyframe?"key":"", counter, s->quality_index);
d86053a4
MM
2767 counter++;
2768
642d7e84
MM
2769 if (s->quality_index != s->last_quality_index)
2770 init_dequantizer(s);
2771
d86053a4 2772 if (s->keyframe) {
3c3f113e
AB
2773 if (!s->theora)
2774 {
2775 skip_bits(&gb, 4); /* width code */
2776 skip_bits(&gb, 4); /* height code */
2777 if (s->version)
2778 {
2779 s->version = get_bits(&gb, 5);
2780 if (counter == 1)
2781 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2782 }
2783 }
2784 if (s->version || s->theora)
2785 {
2786 if (get_bits1(&gb))
2787 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2788 skip_bits(&gb, 2); /* reserved? */
2789 }
2790
74c0ac12
MM
2791 if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2792 if (s->golden_frame.data[0])
2793 avctx->release_buffer(avctx, &s->golden_frame);
8e39d4a7 2794 s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
74c0ac12
MM
2795 } else {
2796 if (s->golden_frame.data[0])
2797 avctx->release_buffer(avctx, &s->golden_frame);
2798 if (s->last_frame.data[0])
2799 avctx->release_buffer(avctx, &s->last_frame);
2800 }
d86053a4 2801
8e39d4a7 2802 s->golden_frame.reference = 3;
d86053a4 2803 if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
9b879566 2804 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
d86053a4
MM
2805 return -1;
2806 }
2807
d86053a4 2808 /* golden frame is also the current frame */
61873c4a 2809 memcpy(&s->current_frame, &s->golden_frame, sizeof(AVFrame));
d86053a4
MM
2810
2811 /* time to figure out pixel addresses? */
2812 if (!s->pixel_addresses_inited)
9a7ad925
AB
2813 {
2814 if (!s->flipped_image)
2815 vp3_calculate_pixel_addresses(s);
2816 else
2817 theora_calculate_pixel_addresses(s);
2818 }
d86053a4 2819 } else {
d86053a4 2820 /* allocate a new current frame */
8e39d4a7 2821 s->current_frame.reference = 3;
d86053a4 2822 if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
9b879566 2823 av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
d86053a4
MM
2824 return -1;
2825 }
d86053a4
MM
2826 }
2827
b928ec64
MN
2828 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2829 s->current_frame.qstride= 0;
2830
220a6f40 2831 {START_TIMER
d86053a4 2832 init_frame(s, &gb);
220a6f40 2833 STOP_TIMER("init_frame")}
d86053a4 2834
892fc83e
MM
2835#if KEYFRAMES_ONLY
2836if (!s->keyframe) {
2837
2838 memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2839 s->current_frame.linesize[0] * s->height);
2840 memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2841 s->current_frame.linesize[1] * s->height / 2);
2842 memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2843 s->current_frame.linesize[2] * s->height / 2);
2844
2845} else {
2846#endif
2847
220a6f40
MN
2848 {START_TIMER
2849 if (unpack_superblocks(s, &gb)){
2850 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2851 return -1;
2852 }
2853 STOP_TIMER("unpack_superblocks")}
2854 {START_TIMER
2855 if (unpack_modes(s, &gb)){
2856 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2857 return -1;
2858 }
2859 STOP_TIMER("unpack_modes")}
2860 {START_TIMER
2861 if (unpack_vectors(s, &gb)){
2862 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2863 return -1;
2864 }
2865 STOP_TIMER("unpack_vectors")}
2866 {START_TIMER
2867 if (unpack_dct_coeffs(s, &gb)){
2868 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
892fc83e
MM
2869 return -1;
2870 }
220a6f40
MN
2871 STOP_TIMER("unpack_dct_coeffs")}
2872 {START_TIMER
d86053a4
MM
2873
2874 reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
220a6f40
MN
2875 STOP_TIMER("reverse_dc_prediction")}
2876 {START_TIMER
44ae98dd 2877 render_fragments(s, 0, s->width, s->height, 0);
220a6f40 2878 STOP_TIMER("render_fragments")}
3d32b429
MM
2879
2880 if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2881 reverse_dc_prediction(s, s->u_fragment_start,
2882 s->fragment_width / 2, s->fragment_height / 2);
2883 reverse_dc_prediction(s, s->v_fragment_start,
2884 s->fragment_width / 2, s->fragment_height / 2);
2885 render_fragments(s, s->u_fragment_start, s->width / 2, s->height / 2, 1);
2886 render_fragments(s, s->v_fragment_start, s->width / 2, s->height / 2, 2);
2887 } else {
2888 memset(s->current_frame.data[1], 0x80, s->width * s->height / 4);
2889 memset(s->current_frame.data[2], 0x80, s->width * s->height / 4);
2890 }
d86053a4 2891
220a6f40 2892 {START_TIMER
d5e18835 2893 apply_loop_filter(s);
220a6f40 2894 STOP_TIMER("apply_loop_filter")}
892fc83e
MM
2895#if KEYFRAMES_ONLY
2896}
2897#endif
2898
d86053a4
MM
2899 *data_size=sizeof(AVFrame);
2900 *(AVFrame*)data= s->current_frame;
2901
44ae98dd
MM
2902 /* release the last frame, if it is allocated and if it is not the
2903 * golden frame */
2904 if ((s->last_frame.data[0]) &&
2905 (s->last_frame.data[0] != s->golden_frame.data[0]))
2906 avctx->release_buffer(avctx, &s->last_frame);
d86053a4 2907
61873c4a
MM
2908 /* shuffle frames (last = current) */
2909 memcpy(&s->last_frame, &s->current_frame, sizeof(AVFrame));
8e39d4a7 2910 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
d86053a4
MM
2911
2912 return buf_size;
2913}
2914
2915/*
2916 * This is the ffmpeg/libavcodec API module cleanup function.
2917 */
2918static int vp3_decode_end(AVCodecContext *avctx)
2919{
2920 Vp3DecodeContext *s = avctx->priv_data;
2921
2922 av_free(s->all_fragments);
a2df5a50 2923 av_free(s->coeffs);
d86053a4
MM
2924 av_free(s->coded_fragment_list);
2925 av_free(s->superblock_fragments);
2926 av_free(s->superblock_macroblocks);
2927 av_free(s->macroblock_fragments);
96a7e73b 2928 av_free(s->macroblock_coding);
8e39d4a7 2929
d86053a4 2930 /* release all frames */
8e39d4a7 2931 if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
892fc83e
MM
2932 avctx->release_buffer(avctx, &s->golden_frame);
2933 if (s->last_frame.data[0])
2934 avctx->release_buffer(avctx, &s->last_frame);
2935 /* no need to release the current_frame since it will always be pointing
2936 * to the same frame as either the golden or last frame */
d86053a4
MM
2937
2938 return 0;
2939}
2940
f44ee2c3
AB
2941static int theora_decode_header(AVCodecContext *avctx, GetBitContext gb)
2942{
2943 Vp3DecodeContext *s = avctx->priv_data;
9a7ad925
AB
2944 int major, minor, micro;
2945
2946 major = get_bits(&gb, 8); /* version major */
2947 minor = get_bits(&gb, 8); /* version minor */
2948 micro = get_bits(&gb, 8); /* version micro */
2949 av_log(avctx, AV_LOG_INFO, "Theora bitstream version %d.%d.%d\n",
2950 major, minor, micro);
2951
105c3d25
AB
2952 /* FIXME: endianess? */
2953 s->theora = (major << 16) | (minor << 8) | micro;
2954
ba7ee4a4 2955 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
9a7ad925 2956 /* but previous versions have the image flipped relative to vp3 */
ba7ee4a4 2957 if (s->theora < 0x030200)
9a7ad925
AB
2958 {
2959 s->flipped_image = 1;
2960 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2961 }
f44ee2c3 2962
f44ee2c3
AB
2963 s->width = get_bits(&gb, 16) << 4;
2964 s->height = get_bits(&gb, 16) << 4;
2965
0ecca7a4
MN
2966 if(avcodec_check_dimensions(avctx, s->width, s->height)){
2967 s->width= s->height= 0;
2968 return -1;
2969 }
2970
f44ee2c3
AB
2971 skip_bits(&gb, 24); /* frame width */
2972 skip_bits(&gb, 24); /* frame height */
2973
2974 skip_bits(&gb, 8); /* offset x */
2975 skip_bits(&gb, 8); /* offset y */
2976
2977 skip_bits(&gb, 32); /* fps numerator */
2978 skip_bits(&gb, 32); /* fps denumerator */
2979 skip_bits(&gb, 24); /* aspect numerator */
2980 skip_bits(&gb, 24); /* aspect denumerator */
2981
ba7ee4a4 2982 if (s->theora < 0x030200)
105c3d25 2983 skip_bits(&gb, 5); /* keyframe frequency force */
f44ee2c3
AB
2984 skip_bits(&gb, 8); /* colorspace */
2985 skip_bits(&gb, 24); /* bitrate */
2986
2987 skip_bits(&gb, 6); /* last(?) quality index */
2988
ba7ee4a4 2989 if (s->theora >= 0x030200)
105c3d25
AB
2990 {
2991 skip_bits(&gb, 5); /* keyframe frequency force */
2992 skip_bits(&gb, 5); /* spare bits */
2993 }
2994
f44ee2c3
AB
2995// align_get_bits(&gb);
2996
2997 avctx->width = s->width;
2998 avctx->height = s->height;
2999
f44ee2c3
AB
3000 return 0;
3001}
3002
91381201
AB
3003static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb)
3004{
3005 int nb_comments, i, tmp;
3006
8a515d2d 3007 tmp = get_bits_long(&gb, 32);
3c3f113e
AB
3008 tmp = be2me_32(tmp);
3009 while(tmp--)
3010 skip_bits(&gb, 8);
91381201 3011
8a515d2d 3012 nb_comments = get_bits_long(&gb, 32);
3c3f113e 3013 nb_comments = be2me_32(nb_comments);
91381201
AB
3014 for (i = 0; i < nb_comments; i++)
3015 {
8a515d2d 3016 tmp = get_bits_long(&gb, 32);
3c3f113e
AB
3017 tmp = be2me_32(tmp);
3018 while(tmp--)
91381201
AB
3019 skip_bits(&gb, 8);
3020 }
3021
3022 return 0;
3023}
3024
f44ee2c3
AB
3025static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb)
3026{
3027 Vp3DecodeContext *s = avctx->priv_data;
ba7ee4a4
MC
3028 int i, n;
3029
3030 if (s->theora >= 0x030200) {
3031 n = get_bits(&gb, 3);
3032 /* loop filter table */
3033 for (i = 0; i < 64; i++)
3034 skip_bits(&gb, n);
3035 }
f44ee2c3 3036
ba7ee4a4
MC
3037 if (s->theora >= 0x030200)
3038 n = get_bits(&gb, 4) + 1;
3039 else
3040 n = 16;
f44ee2c3
AB
3041 /* quality threshold table */
3042 for (i = 0; i < 64; i++)
ba7ee4a4 3043 s->coded_ac_scale_factor[i] = get_bits(&gb, n);
f44ee2c3 3044
ba7ee4a4
MC
3045 if (s->theora >= 0x030200)
3046 n = get_bits(&gb, 4) + 1;
3047 else
3048 n = 16;
f44ee2c3
AB
3049 /* dc scale factor table */
3050 for (i = 0; i < 64; i++)
ba7ee4a4 3051 s->coded_dc_scale_factor[i] = get_bits(&gb, n);
f44ee2c3 3052
ba7ee4a4
MC
3053 if (s->theora >= 0x030200)
3054 n = get_bits(&gb, 9) + 1;
3055 else
3056 n = 3;
3057 if (n != 3) {
3058 av_log(NULL,AV_LOG_ERROR, "unsupported nbms : %d\n", n);
3059 return -1;
3060 }
f44ee2c3
AB
3061 /* y coeffs */
3062 for (i = 0; i < 64; i++)
3063 s->coded_intra_y_dequant[i] = get_bits(&gb, 8);
3064
3065 /* uv coeffs */
3066 for (i = 0; i < 64; i++)
3067 s->coded_intra_c_dequant[i] = get_bits(&gb, 8);
3068
3069 /* inter coeffs */
3070 for (i = 0; i < 64; i++)
3071 s->coded_inter_dequant[i] = get_bits(&gb, 8);
3c3f113e
AB
3072
3073 /* FIXME: read huffmann tree.. */
f44ee2c3
AB
3074
3075 s->theora_tables = 1;
3076
3077 return 0;
3078}
3079
3080static int theora_decode_init(AVCodecContext *avctx)
3081{
3082 Vp3DecodeContext *s = avctx->priv_data;
3083 GetBitContext gb;
3084 int ptype;
ee89b2b9
MN
3085 uint8_t *p= avctx->extradata;
3086 int op_bytes, i;
f44ee2c3
AB
3087
3088 s->theora = 1;
3089
3090 if (!avctx->extradata_size)
3091 return -1;
3092
ee89b2b9
MN
3093 for(i=0;i<3;i++) {
3094 op_bytes = *(p++)<<8;
3095 op_bytes += *(p++);
3096
3097 init_get_bits(&gb, p, op_bytes);
3098 p += op_bytes;
f44ee2c3
AB
3099
3100 ptype = get_bits(&gb, 8);
3101 debug_vp3("Theora headerpacket type: %x\n", ptype);
3102
3103 if (!(ptype & 0x80))
3104 return -1;
3105
3106 skip_bits(&gb, 6*8); /* "theora" */
3107
3108 switch(ptype)
3109 {
3110 case 0x80:
3111 theora_decode_header(avctx, gb);
f44ee2c3
AB
3112 break;
3113 case 0x81:
91381201 3114 theora_decode_comments(avctx, gb);
f44ee2c3
AB
3115 break;
3116 case 0x82:
3117 theora_decode_tables(avctx, gb);
3118 break;
3119 }
ee89b2b9 3120 }
f44ee2c3 3121
ba7ee4a4 3122 vp3_decode_init(avctx);
f44ee2c3
AB
3123 return 0;
3124}
3125
d86053a4
MM
3126AVCodec vp3_decoder = {
3127 "vp3",
3128 CODEC_TYPE_VIDEO,
3129 CODEC_ID_VP3,
3130 sizeof(Vp3DecodeContext),
3131 vp3_decode_init,
3132 NULL,
3133 vp3_decode_end,
3134 vp3_decode_frame,
3135 0,
3136 NULL
3137};
f44ee2c3 3138
aac064b5 3139#ifndef CONFIG_LIBTHEORA
f44ee2c3
AB
3140AVCodec theora_decoder = {
3141 "theora",
3142 CODEC_TYPE_VIDEO,
3143 CODEC_ID_THEORA,
3144 sizeof(Vp3DecodeContext),
3145 theora_decode_init,
3146 NULL,
3147 vp3_decode_end,
3148 vp3_decode_frame,
3149 0,
3150 NULL
3151};
aac064b5 3152#endif