Add missing git rev hash.
[libav.git] / libavcodec / vp3.c
CommitLineData
d86053a4 1/*
67335dbc 2 * Copyright (C) 2003-2004 the ffmpeg project
d86053a4 3 *
b78e7197
DB
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
d86053a4
MM
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
b78e7197 9 * version 2.1 of the License, or (at your option) any later version.
d86053a4 10 *
b78e7197 11 * FFmpeg is distributed in the hope that it will be useful,
d86053a4
MM
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
b78e7197 17 * License along with FFmpeg; if not, write to the Free Software
5509bffa 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
d86053a4
MM
19 */
20
21/**
ba87f080 22 * @file
d86053a4 23 * On2 VP3 Video Decoder
0ad72bdd
MM
24 *
25 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26 * For more information about the VP3 coding process, visit:
9db5bdfa 27 * http://wiki.multimedia.cx/index.php?title=On2_VP3
0ad72bdd
MM
28 *
29 * Theora decoder by Alex Beregszaszi
d86053a4
MM
30 */
31
32#include <stdio.h>
33#include <stdlib.h>
34#include <string.h>
d86053a4 35
6ce9b431 36#include "libavcore/imgutils.h"
d86053a4
MM
37#include "avcodec.h"
38#include "dsputil.h"
9106a698 39#include "get_bits.h"
d86053a4
MM
40
41#include "vp3data.h"
da91ed59 42#include "xiph.h"
d86053a4
MM
43
44#define FRAGMENT_PIXELS 8
45
0efbd068
MM
46static av_cold int vp3_decode_end(AVCodecContext *avctx);
47
7beddb12 48//FIXME split things out into their own arrays
d86053a4 49typedef struct Vp3Fragment {
c72625f2 50 int16_t dc;
288774bb 51 uint8_t coding_method;
f2264fa5 52 uint8_t qpi;
d86053a4
MM
53} Vp3Fragment;
54
55#define SB_NOT_CODED 0
56#define SB_PARTIALLY_CODED 1
57#define SB_FULLY_CODED 2
58
ecb51b25
DC
59// This is the maximum length of a single long bit run that can be encoded
60// for superblock coding or block qps. Theora special-cases this to read a
61// bit instead of flipping the current bit to allow for runs longer than 4129.
62#define MAXIMUM_LONG_BIT_RUN 4129
63
d86053a4
MM
64#define MODE_INTER_NO_MV 0
65#define MODE_INTRA 1
66#define MODE_INTER_PLUS_MV 2
67#define MODE_INTER_LAST_MV 3
68#define MODE_INTER_PRIOR_LAST 4
69#define MODE_USING_GOLDEN 5
70#define MODE_GOLDEN_MV 6
71#define MODE_INTER_FOURMV 7
72#define CODING_MODE_COUNT 8
73
74/* special internal mode */
75#define MODE_COPY 8
76
77/* There are 6 preset schemes, plus a free-form scheme */
e8e47435 78static const int ModeAlphabet[6][CODING_MODE_COUNT] =
d86053a4 79{
d86053a4 80 /* scheme 1: Last motion vector dominates */
115329f1 81 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
d86053a4 82 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
115329f1 83 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
84 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
85
86 /* scheme 2 */
115329f1 87 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
d86053a4 88 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
115329f1 89 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
90 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
91
92 /* scheme 3 */
115329f1 93 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
d86053a4 94 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
115329f1 95 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
96 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
97
98 /* scheme 4 */
115329f1 99 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
d86053a4 100 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
115329f1 101 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
102 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
103
104 /* scheme 5: No motion vector dominates */
115329f1 105 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
d86053a4 106 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
115329f1 107 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
108 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
109
110 /* scheme 6 */
115329f1 111 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
d86053a4 112 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
115329f1 113 MODE_INTER_PLUS_MV, MODE_INTRA,
d86053a4
MM
114 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
115
116};
117
7a095ea6
DC
118static const uint8_t hilbert_offset[16][2] = {
119 {0,0}, {1,0}, {1,1}, {0,1},
120 {0,2}, {0,3}, {1,3}, {1,2},
121 {2,2}, {2,3}, {3,3}, {3,2},
122 {3,1}, {2,1}, {2,0}, {3,0}
123};
124
d86053a4
MM
125#define MIN_DEQUANT_VAL 2
126
127typedef struct Vp3DecodeContext {
128 AVCodecContext *avctx;
f44ee2c3 129 int theora, theora_tables;
3c3f113e 130 int version;
d86053a4 131 int width, height;
1e76a1da 132 int chroma_x_shift, chroma_y_shift;
d86053a4
MM
133 AVFrame golden_frame;
134 AVFrame last_frame;
135 AVFrame current_frame;
136 int keyframe;
137 DSPContext dsp;
9a7ad925 138 int flipped_image;
a8de3901 139 int last_slice_end;
a4501a45 140 int skip_loop_filter;
d86053a4 141
f2264fa5
DC
142 int qps[3];
143 int nqps;
144 int last_qps[3];
d86053a4
MM
145
146 int superblock_count;
892fc83e
MM
147 int y_superblock_width;
148 int y_superblock_height;
35c28d23 149 int y_superblock_count;
892fc83e
MM
150 int c_superblock_width;
151 int c_superblock_height;
35c28d23 152 int c_superblock_count;
d86053a4
MM
153 int u_superblock_start;
154 int v_superblock_start;
155 unsigned char *superblock_coding;
156
157 int macroblock_count;
158 int macroblock_width;
159 int macroblock_height;
160
161 int fragment_count;
57783884
DC
162 int fragment_width[2];
163 int fragment_height[2];
d86053a4
MM
164
165 Vp3Fragment *all_fragments;
1abbf64e 166 int fragment_start[3];
735acf56 167 int data_offset[3];
115329f1 168
14268254
DC
169 int8_t (*motion_val[2])[2];
170
36af0c95 171 ScanTable scantable;
115329f1 172
f44ee2c3
AB
173 /* tables */
174 uint16_t coded_dc_scale_factor[64];
67335dbc 175 uint32_t coded_ac_scale_factor[64];
ae1dd8e1
MN
176 uint8_t base_matrix[384][64];
177 uint8_t qr_count[2][3];
178 uint8_t qr_size [2][3][64];
179 uint16_t qr_base[2][3][64];
d86053a4 180
c72625f2
DC
181 /**
182 * This is a list of all tokens in bitstream order. Reordering takes place
183 * by pulling from each level during IDCT. As a consequence, IDCT must be
184 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
185 * otherwise. The 32 different tokens with up to 12 bits of extradata are
186 * collapsed into 3 types, packed as follows:
187 * (from the low to high bits)
188 *
189 * 2 bits: type (0,1,2)
190 * 0: EOB run, 14 bits for run length (12 needed)
191 * 1: zero run, 7 bits for run length
192 * 7 bits for the next coefficient (3 needed)
193 * 2: coefficient, 14 bits (11 needed)
194 *
195 * Coefficients are signed, so are packed in the highest bits for automatic
196 * sign extension.
197 */
198 int16_t *dct_tokens[3][64];
199 int16_t *dct_tokens_base;
200#define TOKEN_EOB(eob_run) ((eob_run) << 2)
201#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
202#define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
203
204 /**
205 * number of blocks that contain DCT coefficients at the given level or higher
206 */
207 int num_coded_frags[3][64];
208 int total_num_coded_frags;
209
f4433de9 210 /* this is a list of indexes into the all_fragments array indicating
d86053a4 211 * which of the fragments are coded */
c72625f2 212 int *coded_fragment_list[3];
098523eb 213
d86053a4
MM
214 VLC dc_vlc[16];
215 VLC ac_vlc_1[16];
216 VLC ac_vlc_2[16];
217 VLC ac_vlc_3[16];
218 VLC ac_vlc_4[16];
219
0ad72bdd
MM
220 VLC superblock_run_length_vlc;
221 VLC fragment_run_length_vlc;
222 VLC mode_code_vlc;
223 VLC motion_vector_vlc;
224
38acbc3c
MM
225 /* these arrays need to be on 16-byte boundaries since SSE2 operations
226 * index into them */
84dc2d8a 227 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; //<qmat[qpi][is_inter][plane]
d86053a4
MM
228
229 /* This table contains superblock_count * 16 entries. Each set of 16
f4433de9 230 * numbers corresponds to the fragment indexes 0..15 of the superblock.
d86053a4
MM
231 * An entry will be -1 to indicate that no entry corresponds to that
232 * index. */
233 int *superblock_fragments;
234
115329f1 235 /* This is an array that indicates how a particular macroblock
74c0ac12 236 * is coded. */
96a7e73b 237 unsigned char *macroblock_coding;
d86053a4 238
a2f11b3c 239 uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
191e8ca7 240 int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
39922395 241
f44b08a5
MM
242 /* Huffman decode */
243 int hti;
244 unsigned int hbits;
245 int entries;
246 int huff_code_size;
8e6daa4a 247 uint32_t huffman_table[80][32][2];
f44b08a5 248
51ace577 249 uint8_t filter_limit_values[64];
84dc2d8a 250 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
d86053a4
MM
251} Vp3DecodeContext;
252
253/************************************************************************
254 * VP3 specific functions
255 ************************************************************************/
256
257/*
258 * This function sets up all of the various blocks mappings:
259 * superblocks <-> fragments, macroblocks <-> fragments,
260 * superblocks <-> macroblocks
892fc83e 261 *
49bd8e4b 262 * @return 0 is successful; returns 1 if *anything* went wrong.
d86053a4 263 */
115329f1 264static int init_block_mapping(Vp3DecodeContext *s)
d86053a4 265{
a16389c1
DC
266 int sb_x, sb_y, plane;
267 int x, y, i, j = 0;
268
269 for (plane = 0; plane < 3; plane++) {
270 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
271 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
57783884
DC
272 int frag_width = s->fragment_width[!!plane];
273 int frag_height = s->fragment_height[!!plane];
a16389c1
DC
274
275 for (sb_y = 0; sb_y < sb_height; sb_y++)
276 for (sb_x = 0; sb_x < sb_width; sb_x++)
277 for (i = 0; i < 16; i++) {
278 x = 4*sb_x + hilbert_offset[i][0];
279 y = 4*sb_y + hilbert_offset[i][1];
280
281 if (x < frag_width && y < frag_height)
282 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
283 else
284 s->superblock_fragments[j++] = -1;
285 }
d86053a4
MM
286 }
287
892fc83e 288 return 0; /* successful path out */
d86053a4
MM
289}
290
291/*
f44b08a5 292 * This function sets up the dequantization tables used for a particular
d86053a4
MM
293 * frame.
294 */
f2264fa5 295static void init_dequantizer(Vp3DecodeContext *s, int qpi)
d86053a4 296{
f2264fa5
DC
297 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
298 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
36c32bdd 299 int i, plane, inter, qri, bmi, bmj, qistart;
d86053a4 300
ae1dd8e1
MN
301 for(inter=0; inter<2; inter++){
302 for(plane=0; plane<3; plane++){
303 int sum=0;
304 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
305 sum+= s->qr_size[inter][plane][qri];
f2264fa5 306 if(s->qps[qpi] <= sum)
ae1dd8e1
MN
307 break;
308 }
309 qistart= sum - s->qr_size[inter][plane][qri];
310 bmi= s->qr_base[inter][plane][qri ];
311 bmj= s->qr_base[inter][plane][qri+1];
312 for(i=0; i<64; i++){
f2264fa5
DC
313 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
314 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
ae1dd8e1
MN
315 + s->qr_size[inter][plane][qri])
316 / (2*s->qr_size[inter][plane][qri]);
317
a14ab4e4 318 int qmin= 8<<(inter + !i);
ae1dd8e1
MN
319 int qscale= i ? ac_scale_factor : dc_scale_factor;
320
f2264fa5 321 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
ae1dd8e1 322 }
f2264fa5
DC
323 // all DC coefficients use the same quant so as not to interfere with DC prediction
324 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
ae1dd8e1 325 }
d86053a4 326 }
115329f1 327
f2264fa5 328 memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
d86053a4
MM
329}
330
331/*
f44b08a5
MM
332 * This function initializes the loop filter boundary limits if the frame's
333 * quality index is different from the previous frame's.
7fa5f999
RD
334 *
335 * The filter_limit_values may not be larger than 127.
f44b08a5
MM
336 */
337static void init_loop_filter(Vp3DecodeContext *s)
338{
339 int *bounding_values= s->bounding_values_array+127;
340 int filter_limit;
341 int x;
7fa5f999 342 int value;
f44b08a5 343
f2264fa5 344 filter_limit = s->filter_limit_values[s->qps[0]];
f44b08a5
MM
345
346 /* set up the bounding values */
347 memset(s->bounding_values_array, 0, 256 * sizeof(int));
348 for (x = 0; x < filter_limit; x++) {
f44b08a5
MM
349 bounding_values[-x] = -x;
350 bounding_values[x] = x;
f44b08a5 351 }
7fa5f999
RD
352 for (x = value = filter_limit; x < 128 && value; x++, value--) {
353 bounding_values[ x] = value;
354 bounding_values[-x] = -value;
355 }
356 if (value)
357 bounding_values[128] = value;
357f45d9 358 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
f44b08a5
MM
359}
360
361/*
115329f1 362 * This function unpacks all of the superblock/macroblock/fragment coding
d86053a4
MM
363 * information from the bitstream.
364 */
892fc83e 365static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 366{
e2720b6b 367 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
d86053a4
MM
368 int bit = 0;
369 int current_superblock = 0;
370 int current_run = 0;
855c720c 371 int num_partial_superblocks = 0;
d86053a4
MM
372
373 int i, j;
374 int current_fragment;
35c28d23 375 int plane;
d86053a4 376
d86053a4 377 if (s->keyframe) {
d86053a4
MM
378 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
379
380 } else {
381
382 /* unpack the list of partially-coded superblocks */
cc46005f
DC
383 bit = get_bits1(gb) ^ 1;
384 current_run = 0;
385
60867312 386 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
cc46005f
DC
387 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
388 bit = get_bits1(gb);
389 else
390 bit ^= 1;
391
115329f1 392 current_run = get_vlc2(gb,
33dbc1b7
DC
393 s->superblock_run_length_vlc.table, 6, 2) + 1;
394 if (current_run == 34)
d8278bab 395 current_run += get_bits(gb, 12);
d86053a4 396
33dbc1b7
DC
397 if (current_superblock + current_run > s->superblock_count) {
398 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
399 return -1;
400 }
401
402 memset(s->superblock_coding + current_superblock, bit, current_run);
403
404 current_superblock += current_run;
855c720c
DC
405 if (bit)
406 num_partial_superblocks += current_run;
d86053a4
MM
407 }
408
409 /* unpack the list of fully coded superblocks if any of the blocks were
410 * not marked as partially coded in the previous step */
855c720c
DC
411 if (num_partial_superblocks < s->superblock_count) {
412 int superblocks_decoded = 0;
d86053a4
MM
413
414 current_superblock = 0;
cc46005f
DC
415 bit = get_bits1(gb) ^ 1;
416 current_run = 0;
417
60867312
DC
418 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
419 && get_bits_left(gb) > 0) {
cc46005f
DC
420
421 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
422 bit = get_bits1(gb);
423 else
424 bit ^= 1;
425
115329f1 426 current_run = get_vlc2(gb,
855c720c
DC
427 s->superblock_run_length_vlc.table, 6, 2) + 1;
428 if (current_run == 34)
d8278bab 429 current_run += get_bits(gb, 12);
855c720c
DC
430
431 for (j = 0; j < current_run; current_superblock++) {
432 if (current_superblock >= s->superblock_count) {
433 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
434 return -1;
d86053a4 435 }
855c720c
DC
436
437 /* skip any superblocks already marked as partially coded */
438 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
b5da3635 439 s->superblock_coding[current_superblock] = 2*bit;
855c720c
DC
440 j++;
441 }
d86053a4 442 }
855c720c 443 superblocks_decoded += current_run;
d86053a4
MM
444 }
445 }
446
447 /* if there were partial blocks, initialize bitstream for
448 * unpacking fragment codings */
855c720c 449 if (num_partial_superblocks) {
d86053a4
MM
450
451 current_run = 0;
5fc32c27 452 bit = get_bits1(gb);
115329f1 453 /* toggle the bit because as soon as the first run length is
d86053a4
MM
454 * fetched the bit will be toggled again */
455 bit ^= 1;
456 }
457 }
458
459 /* figure out which fragments are coded; iterate through each
460 * superblock (all planes) */
c72625f2 461 s->total_num_coded_frags = 0;
96a7e73b 462 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
35c28d23
DC
463
464 for (plane = 0; plane < 3; plane++) {
e2720b6b 465 int sb_start = superblock_starts[plane];
35c28d23 466 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
c72625f2 467 int num_coded_frags = 0;
35c28d23 468
60867312 469 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
d86053a4
MM
470
471 /* iterate through all 16 fragments in a superblock */
472 for (j = 0; j < 16; j++) {
473
474 /* if the fragment is in bounds, check its coding status */
475 current_fragment = s->superblock_fragments[i * 16 + j];
476 if (current_fragment != -1) {
6cb35b45 477 int coded = s->superblock_coding[i];
d86053a4 478
6cb35b45 479 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
d86053a4
MM
480
481 /* fragment may or may not be coded; this is the case
482 * that cares about the fragment coding runs */
b5da3635 483 if (current_run-- == 0) {
d86053a4 484 bit ^= 1;
115329f1 485 current_run = get_vlc2(gb,
b5da3635 486 s->fragment_run_length_vlc.table, 5, 2);
d86053a4 487 }
6cb35b45
DC
488 coded = bit;
489 }
d86053a4 490
6cb35b45 491 if (coded) {
115329f1 492 /* default mode; actual mode will be decoded in
22493ab9 493 * the next phase */
115329f1 494 s->all_fragments[current_fragment].coding_method =
d86053a4 495 MODE_INTER_NO_MV;
c72625f2 496 s->coded_fragment_list[plane][num_coded_frags++] =
d86053a4 497 current_fragment;
d86053a4
MM
498 } else {
499 /* not coded; copy this fragment from the prior frame */
500 s->all_fragments[current_fragment].coding_method =
501 MODE_COPY;
d86053a4 502 }
d86053a4
MM
503 }
504 }
505 }
c72625f2
DC
506 s->total_num_coded_frags += num_coded_frags;
507 for (i = 0; i < 64; i++)
508 s->num_coded_frags[plane][i] = num_coded_frags;
509 if (plane < 2)
510 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
35c28d23 511 }
892fc83e 512 return 0;
d86053a4
MM
513}
514
515/*
516 * This function unpacks all the coding mode data for individual macroblocks
517 * from the bitstream.
518 */
892fc83e 519static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 520{
19cd517d 521 int i, j, k, sb_x, sb_y;
d86053a4
MM
522 int scheme;
523 int current_macroblock;
524 int current_fragment;
525 int coding_mode;
e8e47435 526 int custom_mode_alphabet[CODING_MODE_COUNT];
7c2e31d1 527 const int *alphabet;
1e76a1da 528 Vp3Fragment *frag;
d86053a4 529
d86053a4 530 if (s->keyframe) {
d86053a4
MM
531 for (i = 0; i < s->fragment_count; i++)
532 s->all_fragments[i].coding_method = MODE_INTRA;
533
534 } else {
535
536 /* fetch the mode coding scheme for this frame */
537 scheme = get_bits(gb, 3);
d86053a4
MM
538
539 /* is it a custom coding scheme? */
540 if (scheme == 0) {
d86053a4 541 for (i = 0; i < 8; i++)
2c823b3c
AC
542 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
543 for (i = 0; i < 8; i++)
e8e47435 544 custom_mode_alphabet[get_bits(gb, 3)] = i;
7c2e31d1
DC
545 alphabet = custom_mode_alphabet;
546 } else
547 alphabet = ModeAlphabet[scheme-1];
d86053a4 548
d86053a4
MM
549 /* iterate through all of the macroblocks that contain 1 or more
550 * coded fragments */
19cd517d
DC
551 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
552 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
60867312
DC
553 if (get_bits_left(gb) <= 0)
554 return -1;
d86053a4
MM
555
556 for (j = 0; j < 4; j++) {
19cd517d
DC
557 int mb_x = 2*sb_x + (j>>1);
558 int mb_y = 2*sb_y + (((j>>1)+j)&1);
559 current_macroblock = mb_y * s->macroblock_width + mb_x;
560
15675ce6 561 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
d86053a4
MM
562 continue;
563
ea676144
DC
564#define BLOCK_X (2*mb_x + (k&1))
565#define BLOCK_Y (2*mb_y + (k>>1))
15675ce6
DC
566 /* coding modes are only stored if the macroblock has at least one
567 * luma block coded, otherwise it must be INTER_NO_MV */
568 for (k = 0; k < 4; k++) {
57783884 569 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
15675ce6
DC
570 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
571 break;
572 }
573 if (k == 4) {
574 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
575 continue;
576 }
ea676144 577
d86053a4
MM
578 /* mode 7 means get 3 bits for each coding mode */
579 if (scheme == 7)
580 coding_mode = get_bits(gb, 3);
581 else
7c2e31d1 582 coding_mode = alphabet
0ad72bdd 583 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
d86053a4 584
96a7e73b 585 s->macroblock_coding[current_macroblock] = coding_mode;
ea676144 586 for (k = 0; k < 4; k++) {
1e76a1da
DC
587 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
588 if (frag->coding_method != MODE_COPY)
589 frag->coding_method = coding_mode;
ea676144 590 }
1e76a1da
DC
591
592#define SET_CHROMA_MODES \
593 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
594 frag[s->fragment_start[1]].coding_method = coding_mode;\
595 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
596 frag[s->fragment_start[2]].coding_method = coding_mode;
597
598 if (s->chroma_y_shift) {
599 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
600 SET_CHROMA_MODES
601 } else if (s->chroma_x_shift) {
602 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
603 for (k = 0; k < 2; k++) {
604 SET_CHROMA_MODES
605 frag += s->fragment_width[1];
606 }
607 } else {
608 for (k = 0; k < 4; k++) {
609 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
610 SET_CHROMA_MODES
611 }
d86053a4 612 }
d86053a4 613 }
19cd517d 614 }
d86053a4
MM
615 }
616 }
892fc83e
MM
617
618 return 0;
44ae98dd
MM
619}
620
621/*
d86053a4
MM
622 * This function unpacks all the motion vectors for the individual
623 * macroblocks from the bitstream.
624 */
892fc83e 625static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 626{
eb691ef2 627 int j, k, sb_x, sb_y;
d86053a4 628 int coding_mode;
1c183aa4
DC
629 int motion_x[4];
630 int motion_y[4];
d86053a4
MM
631 int last_motion_x = 0;
632 int last_motion_y = 0;
633 int prior_last_motion_x = 0;
634 int prior_last_motion_y = 0;
635 int current_macroblock;
636 int current_fragment;
14268254 637 int frag;
d86053a4 638
6599e2a7 639 if (s->keyframe)
6298f49f 640 return 0;
10f38380 641
1ae4518d
DC
642 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
643 coding_mode = get_bits1(gb);
d86053a4 644
1ae4518d
DC
645 /* iterate through all of the macroblocks that contain 1 or more
646 * coded fragments */
19cd517d
DC
647 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
648 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
60867312
DC
649 if (get_bits_left(gb) <= 0)
650 return -1;
d86053a4 651
1ae4518d 652 for (j = 0; j < 4; j++) {
19cd517d
DC
653 int mb_x = 2*sb_x + (j>>1);
654 int mb_y = 2*sb_y + (((j>>1)+j)&1);
655 current_macroblock = mb_y * s->macroblock_width + mb_x;
656
657 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
1ae4518d
DC
658 (s->macroblock_coding[current_macroblock] == MODE_COPY))
659 continue;
d86053a4 660
1ae4518d
DC
661 switch (s->macroblock_coding[current_macroblock]) {
662
663 case MODE_INTER_PLUS_MV:
664 case MODE_GOLDEN_MV:
665 /* all 6 fragments use the same motion vector */
666 if (coding_mode == 0) {
667 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
668 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
669 } else {
670 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
671 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
892fc83e 672 }
7f9926a4 673
1ae4518d
DC
674 /* vector maintenance, only on MODE_INTER_PLUS_MV */
675 if (s->macroblock_coding[current_macroblock] ==
676 MODE_INTER_PLUS_MV) {
e32e2d56
AJ
677 prior_last_motion_x = last_motion_x;
678 prior_last_motion_y = last_motion_y;
1ae4518d
DC
679 last_motion_x = motion_x[0];
680 last_motion_y = motion_y[0];
681 }
682 break;
683
684 case MODE_INTER_FOURMV:
685 /* vector maintenance */
686 prior_last_motion_x = last_motion_x;
687 prior_last_motion_y = last_motion_y;
688
689 /* fetch 4 vectors from the bitstream, one for each
690 * Y fragment, then average for the C fragment vectors */
1ae4518d 691 for (k = 0; k < 4; k++) {
57783884 692 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
eb691ef2 693 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
1ae4518d
DC
694 if (coding_mode == 0) {
695 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
696 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
feaf1a73 697 } else {
1ae4518d
DC
698 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
699 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
feaf1a73 700 }
1ae4518d
DC
701 last_motion_x = motion_x[k];
702 last_motion_y = motion_y[k];
703 } else {
704 motion_x[k] = 0;
705 motion_y[k] = 0;
d86053a4 706 }
1ae4518d 707 }
1ae4518d
DC
708 break;
709
710 case MODE_INTER_LAST_MV:
711 /* all 6 fragments use the last motion vector */
712 motion_x[0] = last_motion_x;
713 motion_y[0] = last_motion_y;
d86053a4 714
1ae4518d
DC
715 /* no vector maintenance (last vector remains the
716 * last vector) */
717 break;
718
719 case MODE_INTER_PRIOR_LAST:
720 /* all 6 fragments use the motion vector prior to the
721 * last motion vector */
722 motion_x[0] = prior_last_motion_x;
723 motion_y[0] = prior_last_motion_y;
d86053a4 724
1ae4518d
DC
725 /* vector maintenance */
726 prior_last_motion_x = last_motion_x;
727 prior_last_motion_y = last_motion_y;
728 last_motion_x = motion_x[0];
729 last_motion_y = motion_y[0];
730 break;
44ae98dd 731
1ae4518d
DC
732 default:
733 /* covers intra, inter without MV, golden without MV */
e6e32bdc
MM
734 motion_x[0] = 0;
735 motion_y[0] = 0;
44ae98dd 736
1ae4518d
DC
737 /* no vector maintenance */
738 break;
739 }
d86053a4 740
1ae4518d 741 /* assign the motion vectors to the correct fragments */
ea676144 742 for (k = 0; k < 4; k++) {
1ae4518d 743 current_fragment =
57783884 744 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
e6e32bdc 745 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
14268254
DC
746 s->motion_val[0][current_fragment][0] = motion_x[k];
747 s->motion_val[0][current_fragment][1] = motion_y[k];
e6e32bdc 748 } else {
14268254
DC
749 s->motion_val[0][current_fragment][0] = motion_x[0];
750 s->motion_val[0][current_fragment][1] = motion_y[0];
e6e32bdc 751 }
d86053a4 752 }
1e76a1da 753
1e76a1da 754 if (s->chroma_y_shift) {
1c183aa4
DC
755 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
756 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
757 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
758 }
1e76a1da
DC
759 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
760 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
14268254
DC
761 frag = mb_y*s->fragment_width[1] + mb_x;
762 s->motion_val[1][frag][0] = motion_x[0];
763 s->motion_val[1][frag][1] = motion_y[0];
1e76a1da
DC
764 } else if (s->chroma_x_shift) {
765 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
766 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
767 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
768 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
769 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
770 } else {
771 motion_x[1] = motion_x[0];
772 motion_y[1] = motion_y[0];
773 }
774 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
775 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
776
14268254 777 frag = 2*mb_y*s->fragment_width[1] + mb_x;
1e76a1da 778 for (k = 0; k < 2; k++) {
14268254
DC
779 s->motion_val[1][frag][0] = motion_x[k];
780 s->motion_val[1][frag][1] = motion_y[k];
1e76a1da
DC
781 frag += s->fragment_width[1];
782 }
783 } else {
784 for (k = 0; k < 4; k++) {
14268254 785 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
1e76a1da 786 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
14268254
DC
787 s->motion_val[1][frag][0] = motion_x[k];
788 s->motion_val[1][frag][1] = motion_y[k];
1e76a1da 789 } else {
14268254
DC
790 s->motion_val[1][frag][0] = motion_x[0];
791 s->motion_val[1][frag][1] = motion_y[0];
1e76a1da
DC
792 }
793 }
ea676144 794 }
d86053a4 795 }
19cd517d 796 }
1ae4518d 797 }
892fc83e
MM
798
799 return 0;
d86053a4
MM
800}
801
f2264fa5
DC
802static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
803{
804 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
c72625f2 805 int num_blocks = s->total_num_coded_frags;
f2264fa5
DC
806
807 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
808 i = blocks_decoded = num_blocks_at_qpi = 0;
809
cc46005f
DC
810 bit = get_bits1(gb) ^ 1;
811 run_length = 0;
f2264fa5
DC
812
813 do {
cc46005f
DC
814 if (run_length == MAXIMUM_LONG_BIT_RUN)
815 bit = get_bits1(gb);
816 else
817 bit ^= 1;
818
f2264fa5
DC
819 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
820 if (run_length == 34)
821 run_length += get_bits(gb, 12);
822 blocks_decoded += run_length;
823
824 if (!bit)
825 num_blocks_at_qpi += run_length;
826
827 for (j = 0; j < run_length; i++) {
c72625f2 828 if (i >= s->total_num_coded_frags)
f2264fa5
DC
829 return -1;
830
c72625f2
DC
831 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
832 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
f2264fa5
DC
833 j++;
834 }
835 }
60867312 836 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
f2264fa5
DC
837
838 num_blocks -= num_blocks_at_qpi;
839 }
840
841 return 0;
842}
843
115329f1 844/*
d86053a4
MM
845 * This function is called by unpack_dct_coeffs() to extract the VLCs from
846 * the bitstream. The VLCs encode tokens which are used to unpack DCT
847 * data. This function unpacks all the VLCs for either the Y plane or both
848 * C planes, and is called for DC coefficients or different AC coefficient
849 * levels (since different coefficient types require different VLC tables.
850 *
851 * This function returns a residual eob run. E.g, if a particular token gave
852 * instructions to EOB the next 5 fragments and there were only 2 fragments
853 * left in the current fragment range, 3 would be returned so that it could
854 * be passed into the next call to this same function.
855 */
856static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
857 VLC *table, int coeff_index,
c72625f2 858 int plane,
d86053a4
MM
859 int eob_run)
860{
c72625f2 861 int i, j = 0;
d86053a4 862 int token;
d3076955
MM
863 int zero_run = 0;
864 DCTELEM coeff = 0;
d3076955 865 int bits_to_get;
c72625f2
DC
866 int blocks_ended;
867 int coeff_i = 0;
868 int num_coeffs = s->num_coded_frags[plane][coeff_index];
869 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
d86053a4 870
ee3d7f58 871 /* local references to structure members to avoid repeated deferences */
c72625f2 872 int *coded_fragment_list = s->coded_fragment_list[plane];
ee3d7f58 873 Vp3Fragment *all_fragments = s->all_fragments;
ee3d7f58
MM
874 VLC_TYPE (*vlc_table)[2] = table->table;
875
c72625f2
DC
876 if (num_coeffs < 0)
877 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
878
879 if (eob_run > num_coeffs) {
880 coeff_i = blocks_ended = num_coeffs;
881 eob_run -= num_coeffs;
098523eb 882 } else {
c72625f2
DC
883 coeff_i = blocks_ended = eob_run;
884 eob_run = 0;
74c0ac12
MM
885 }
886
c72625f2
DC
887 // insert fake EOB token to cover the split between planes or zzi
888 if (blocks_ended)
889 dct_tokens[j++] = blocks_ended << 2;
d86053a4 890
f50dafa8 891 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
d86053a4 892 /* decode a VLC into a token */
8e6daa4a 893 token = get_vlc2(gb, vlc_table, 11, 3);
d86053a4 894 /* use the token to get a zero run, a coefficient, and an eob run */
d3076955
MM
895 if (token <= 6) {
896 eob_run = eob_run_base[token];
897 if (eob_run_get_bits[token])
898 eob_run += get_bits(gb, eob_run_get_bits[token]);
c72625f2
DC
899
900 // record only the number of blocks ended in this plane,
901 // any spill will be recorded in the next plane.
902 if (eob_run > num_coeffs - coeff_i) {
903 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
904 blocks_ended += num_coeffs - coeff_i;
905 eob_run -= num_coeffs - coeff_i;
906 coeff_i = num_coeffs;
907 } else {
908 dct_tokens[j++] = TOKEN_EOB(eob_run);
909 blocks_ended += eob_run;
910 coeff_i += eob_run;
911 eob_run = 0;
912 }
d3076955
MM
913 } else {
914 bits_to_get = coeff_get_bits[token];
428984b0
MM
915 if (bits_to_get)
916 bits_to_get = get_bits(gb, bits_to_get);
917 coeff = coeff_tables[token][bits_to_get];
d3076955
MM
918
919 zero_run = zero_run_base[token];
920 if (zero_run_get_bits[token])
921 zero_run += get_bits(gb, zero_run_get_bits[token]);
d86053a4 922
c72625f2
DC
923 if (zero_run) {
924 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
925 } else {
926 // Save DC into the fragment structure. DC prediction is
927 // done in raster order, so the actual DC can't be in with
928 // other tokens. We still need the token in dct_tokens[]
929 // however, or else the structure collapses on itself.
930 if (!coeff_index)
931 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
932
933 dct_tokens[j++] = TOKEN_COEFF(coeff);
934 }
935
936 if (coeff_index + zero_run > 64) {
01f9640b 937 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
c72625f2
DC
938 " %d coeffs left\n", zero_run, 64-coeff_index);
939 zero_run = 64 - coeff_index;
940 }
098523eb 941
c72625f2
DC
942 // zero runs code multiple coefficients,
943 // so don't try to decode coeffs for those higher levels
944 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
945 s->num_coded_frags[plane][i]--;
946 coeff_i++;
947 }
d86053a4
MM
948 }
949
c72625f2
DC
950 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
951 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
952
953 // decrement the number of blocks that have higher coeffecients for each
954 // EOB run at this level
955 if (blocks_ended)
956 for (i = coeff_index+1; i < 64; i++)
957 s->num_coded_frags[plane][i] -= blocks_ended;
958
959 // setup the next buffer
960 if (plane < 2)
961 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
962 else if (coeff_index < 63)
963 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
964
d86053a4
MM
965 return eob_run;
966}
967
138fe832
MM
968static void reverse_dc_prediction(Vp3DecodeContext *s,
969 int first_fragment,
970 int fragment_width,
971 int fragment_height);
d86053a4
MM
972/*
973 * This function unpacks all of the DCT coefficient data from the
974 * bitstream.
975 */
892fc83e 976static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
977{
978 int i;
979 int dc_y_table;
980 int dc_c_table;
981 int ac_y_table;
982 int ac_c_table;
983 int residual_eob_run = 0;
9d8bb031
MM
984 VLC *y_tables[64];
985 VLC *c_tables[64];
d86053a4 986
c72625f2
DC
987 s->dct_tokens[0][0] = s->dct_tokens_base;
988
f4433de9 989 /* fetch the DC table indexes */
d86053a4
MM
990 dc_y_table = get_bits(gb, 4);
991 dc_c_table = get_bits(gb, 4);
992
993 /* unpack the Y plane DC coefficients */
115329f1 994 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
c72625f2 995 0, residual_eob_run);
d86053a4 996
138fe832 997 /* reverse prediction of the Y-plane DC coefficients */
57783884 998 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
138fe832 999
d86053a4 1000 /* unpack the C plane DC coefficients */
d86053a4 1001 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
c72625f2
DC
1002 1, residual_eob_run);
1003 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1004 2, residual_eob_run);
d86053a4 1005
138fe832
MM
1006 /* reverse prediction of the C-plane DC coefficients */
1007 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1008 {
1009 reverse_dc_prediction(s, s->fragment_start[1],
57783884 1010 s->fragment_width[1], s->fragment_height[1]);
138fe832 1011 reverse_dc_prediction(s, s->fragment_start[2],
57783884 1012 s->fragment_width[1], s->fragment_height[1]);
138fe832
MM
1013 }
1014
f4433de9 1015 /* fetch the AC table indexes */
d86053a4
MM
1016 ac_y_table = get_bits(gb, 4);
1017 ac_c_table = get_bits(gb, 4);
1018
9d8bb031 1019 /* build tables of AC VLC tables */
d86053a4 1020 for (i = 1; i <= 5; i++) {
9d8bb031
MM
1021 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1022 c_tables[i] = &s->ac_vlc_1[ac_c_table];
d86053a4 1023 }
d86053a4 1024 for (i = 6; i <= 14; i++) {
9d8bb031
MM
1025 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1026 c_tables[i] = &s->ac_vlc_2[ac_c_table];
d86053a4 1027 }
d86053a4 1028 for (i = 15; i <= 27; i++) {
9d8bb031
MM
1029 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1030 c_tables[i] = &s->ac_vlc_3[ac_c_table];
d86053a4 1031 }
d86053a4 1032 for (i = 28; i <= 63; i++) {
9d8bb031
MM
1033 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1034 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1035 }
1036
1037 /* decode all AC coefficents */
1038 for (i = 1; i <= 63; i++) {
9d8bb031 1039 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
c72625f2 1040 0, residual_eob_run);
d86053a4 1041
9d8bb031 1042 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
c72625f2
DC
1043 1, residual_eob_run);
1044 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1045 2, residual_eob_run);
d86053a4 1046 }
892fc83e
MM
1047
1048 return 0;
d86053a4
MM
1049}
1050
1051/*
1052 * This function reverses the DC prediction for each coded fragment in
115329f1 1053 * the frame. Much of this function is adapted directly from the original
d86053a4
MM
1054 * VP3 source code.
1055 */
1056#define COMPATIBLE_FRAME(x) \
1057 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
c72625f2 1058#define DC_COEFF(u) s->all_fragments[u].dc
d86053a4
MM
1059
1060static void reverse_dc_prediction(Vp3DecodeContext *s,
1061 int first_fragment,
1062 int fragment_width,
115329f1 1063 int fragment_height)
d86053a4
MM
1064{
1065
1066#define PUL 8
1067#define PU 4
1068#define PUR 2
1069#define PL 1
1070
1071 int x, y;
1072 int i = first_fragment;
1073
59ef342b 1074 int predicted_dc;
d86053a4 1075
d86053a4
MM
1076 /* DC values for the left, up-left, up, and up-right fragments */
1077 int vl, vul, vu, vur;
1078
f4433de9 1079 /* indexes for the left, up-left, up, and up-right fragments */
d86053a4
MM
1080 int l, ul, u, ur;
1081
115329f1 1082 /*
d86053a4
MM
1083 * The 6 fields mean:
1084 * 0: up-left multiplier
1085 * 1: up multiplier
1086 * 2: up-right multiplier
1087 * 3: left multiplier
d86053a4 1088 */
bb991087 1089 static const int predictor_transform[16][4] = {
006ff1ca
MN
1090 { 0, 0, 0, 0},
1091 { 0, 0, 0,128}, // PL
1092 { 0, 0,128, 0}, // PUR
1093 { 0, 0, 53, 75}, // PUR|PL
1094 { 0,128, 0, 0}, // PU
1095 { 0, 64, 0, 64}, // PU|PL
1096 { 0,128, 0, 0}, // PU|PUR
1097 { 0, 0, 53, 75}, // PU|PUR|PL
1098 {128, 0, 0, 0}, // PUL
1099 { 0, 0, 0,128}, // PUL|PL
1100 { 64, 0, 64, 0}, // PUL|PUR
1101 { 0, 0, 53, 75}, // PUL|PUR|PL
1102 { 0,128, 0, 0}, // PUL|PU
1103 {-104,116, 0,116}, // PUL|PU|PL
1104 { 24, 80, 24, 0}, // PUL|PU|PUR
1105 {-104,116, 0,116} // PUL|PU|PUR|PL
d86053a4
MM
1106 };
1107
1108 /* This table shows which types of blocks can use other blocks for
1109 * prediction. For example, INTRA is the only mode in this table to
1110 * have a frame number of 0. That means INTRA blocks can only predict
115329f1 1111 * from other INTRA blocks. There are 2 golden frame coding types;
d86053a4
MM
1112 * blocks encoding in these modes can only predict from other blocks
1113 * that were encoded with these 1 of these 2 modes. */
50ba3fd7 1114 static const unsigned char compatible_frame[9] = {
d86053a4
MM
1115 1, /* MODE_INTER_NO_MV */
1116 0, /* MODE_INTRA */
1117 1, /* MODE_INTER_PLUS_MV */
1118 1, /* MODE_INTER_LAST_MV */
1119 1, /* MODE_INTER_PRIOR_MV */
1120 2, /* MODE_USING_GOLDEN */
1121 2, /* MODE_GOLDEN_MV */
50ba3fd7
JGG
1122 1, /* MODE_INTER_FOUR_MV */
1123 3 /* MODE_COPY */
d86053a4
MM
1124 };
1125 int current_frame_type;
1126
1127 /* there is a last DC predictor for each of the 3 frame types */
1128 short last_dc[3];
1129
1130 int transform = 0;
1131
d86053a4
MM
1132 vul = vu = vur = vl = 0;
1133 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1134
1135 /* for each fragment row... */
1136 for (y = 0; y < fragment_height; y++) {
1137
1138 /* for each fragment in a row... */
1139 for (x = 0; x < fragment_width; x++, i++) {
1140
1141 /* reverse prediction if this block was coded */
1142 if (s->all_fragments[i].coding_method != MODE_COPY) {
1143
115329f1 1144 current_frame_type =
d86053a4 1145 compatible_frame[s->all_fragments[i].coding_method];
d86053a4 1146
f72f8a77
MN
1147 transform= 0;
1148 if(x){
1149 l= i-1;
7beddb12 1150 vl = DC_COEFF(l);
50ba3fd7 1151 if(COMPATIBLE_FRAME(l))
006ff1ca 1152 transform |= PL;
f72f8a77
MN
1153 }
1154 if(y){
1155 u= i-fragment_width;
7beddb12 1156 vu = DC_COEFF(u);
50ba3fd7 1157 if(COMPATIBLE_FRAME(u))
006ff1ca 1158 transform |= PU;
f72f8a77
MN
1159 if(x){
1160 ul= i-fragment_width-1;
1161 vul = DC_COEFF(ul);
50ba3fd7 1162 if(COMPATIBLE_FRAME(ul))
006ff1ca 1163 transform |= PUL;
f72f8a77
MN
1164 }
1165 if(x + 1 < fragment_width){
1166 ur= i-fragment_width+1;
1167 vur = DC_COEFF(ur);
50ba3fd7 1168 if(COMPATIBLE_FRAME(ur))
006ff1ca 1169 transform |= PUR;
f72f8a77 1170 }
d86053a4
MM
1171 }
1172
d86053a4
MM
1173 if (transform == 0) {
1174
1175 /* if there were no fragments to predict from, use last
1176 * DC saved */
7beddb12 1177 predicted_dc = last_dc[current_frame_type];
d86053a4
MM
1178 } else {
1179
1180 /* apply the appropriate predictor transform */
1181 predicted_dc =
1182 (predictor_transform[transform][0] * vul) +
1183 (predictor_transform[transform][1] * vu) +
1184 (predictor_transform[transform][2] * vur) +
1185 (predictor_transform[transform][3] * vl);
1186
684d9e36 1187 predicted_dc /= 128;
d86053a4
MM
1188
1189 /* check for outranging on the [ul u l] and
1190 * [ul u ur l] predictors */
c11cb375 1191 if ((transform == 15) || (transform == 13)) {
c26abfa5 1192 if (FFABS(predicted_dc - vu) > 128)
d86053a4 1193 predicted_dc = vu;
c26abfa5 1194 else if (FFABS(predicted_dc - vl) > 128)
d86053a4 1195 predicted_dc = vl;
c26abfa5 1196 else if (FFABS(predicted_dc - vul) > 128)
d86053a4
MM
1197 predicted_dc = vul;
1198 }
d86053a4
MM
1199 }
1200
7beddb12 1201 /* at long last, apply the predictor */
c72625f2 1202 DC_COEFF(i) += predicted_dc;
d86053a4 1203 /* save the DC */
7beddb12 1204 last_dc[current_frame_type] = DC_COEFF(i);
d86053a4
MM
1205 }
1206 }
1207 }
1208}
1209
256c0662 1210static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
fe313556 1211{
fe313556
DC
1212 int x, y;
1213 int *bounding_values= s->bounding_values_array+127;
1214
57783884
DC
1215 int width = s->fragment_width[!!plane];
1216 int height = s->fragment_height[!!plane];
621f9a40
DC
1217 int fragment = s->fragment_start [plane] + ystart * width;
1218 int stride = s->current_frame.linesize[plane];
1219 uint8_t *plane_data = s->current_frame.data [plane];
1220 if (!s->flipped_image) stride = -stride;
735acf56 1221 plane_data += s->data_offset[plane] + 8*ystart*stride;
621f9a40
DC
1222
1223 for (y = ystart; y < yend; y++) {
1224
1225 for (x = 0; x < width; x++) {
1226 /* This code basically just deblocks on the edges of coded blocks.
1227 * However, it has to be much more complicated because of the
1228 * braindamaged deblock ordering used in VP3/Theora. Order matters
1229 * because some pixels get filtered twice. */
1230 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1231 {
1232 /* do not perform left edge filter for left columns frags */
1233 if (x > 0) {
1234 s->dsp.vp3_h_loop_filter(
735acf56 1235 plane_data + 8*x,
621f9a40
DC
1236 stride, bounding_values);
1237 }
fe313556 1238
621f9a40
DC
1239 /* do not perform top edge filter for top row fragments */
1240 if (y > 0) {
1241 s->dsp.vp3_v_loop_filter(
735acf56 1242 plane_data + 8*x,
621f9a40
DC
1243 stride, bounding_values);
1244 }
fe313556 1245
621f9a40
DC
1246 /* do not perform right edge filter for right column
1247 * fragments or if right fragment neighbor is also coded
1248 * in this frame (it will be filtered in next iteration) */
1249 if ((x < width - 1) &&
1250 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1251 s->dsp.vp3_h_loop_filter(
735acf56 1252 plane_data + 8*x + 8,
621f9a40 1253 stride, bounding_values);
fe313556
DC
1254 }
1255
621f9a40
DC
1256 /* do not perform bottom edge filter for bottom row
1257 * fragments or if bottom fragment neighbor is also coded
1258 * in this frame (it will be filtered in the next row) */
1259 if ((y < height - 1) &&
1260 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1261 s->dsp.vp3_v_loop_filter(
735acf56 1262 plane_data + 8*x + 8*stride,
621f9a40
DC
1263 stride, bounding_values);
1264 }
fe313556 1265 }
621f9a40
DC
1266
1267 fragment++;
fe313556 1268 }
735acf56 1269 plane_data += 8*stride;
621f9a40 1270 }
fe313556
DC
1271}
1272
a8de3901 1273/**
49bd8e4b 1274 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
c72625f2
DC
1275 * for the next block in coding order
1276 */
1277static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1278 int plane, int inter, DCTELEM block[64])
1279{
1280 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1281 uint8_t *perm = s->scantable.permutated;
1282 int i = 0;
1283
1284 do {
1285 int token = *s->dct_tokens[plane][i];
1286 switch (token & 3) {
1287 case 0: // EOB
1288 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1289 s->dct_tokens[plane][i]++;
1290 else
1291 *s->dct_tokens[plane][i] = token & ~3;
1292 goto end;
1293 case 1: // zero run
1294 s->dct_tokens[plane][i]++;
1295 i += (token >> 2) & 0x7f;
1296 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1297 i++;
1298 break;
1299 case 2: // coeff
1300 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1301 s->dct_tokens[plane][i++]++;
1302 break;
d7097c2d 1303 default: // shouldn't happen
c72625f2
DC
1304 return i;
1305 }
1306 } while (i < 64);
1307end:
1308 // the actual DC+prediction is in the fragment structure
1309 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1310 return i;
1311}
1312
1313/**
a8de3901
DC
1314 * called when all pixels up to row y are complete
1315 */
1316static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1317{
1318 int h, cy;
1319 int offset[4];
1320
1321 if(s->avctx->draw_horiz_band==NULL)
1322 return;
1323
1324 h= y - s->last_slice_end;
3b9ee20f 1325 s->last_slice_end= y;
a8de3901
DC
1326 y -= h;
1327
1328 if (!s->flipped_image) {
83f72f13 1329 y = s->avctx->height - y - h;
a8de3901
DC
1330 }
1331
8764389d 1332 cy = y >> s->chroma_y_shift;
a8de3901
DC
1333 offset[0] = s->current_frame.linesize[0]*y;
1334 offset[1] = s->current_frame.linesize[1]*cy;
1335 offset[2] = s->current_frame.linesize[2]*cy;
1336 offset[3] = 0;
1337
1338 emms_c();
1339 s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
a8de3901
DC
1340}
1341
d86053a4 1342/*
dc4b78d9 1343 * Perform the final rendering for a particular slice of data.
7a095ea6 1344 * The slice number ranges from 0..(c_superblock_height - 1).
dc4b78d9
MM
1345 */
1346static void render_slice(Vp3DecodeContext *s, int slice)
1347{
e8dcd730 1348 int x, y, i, j, fragment;
40d11227 1349 LOCAL_ALIGNED_16(DCTELEM, block, [64]);
dc4b78d9 1350 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
dc4b78d9
MM
1351 int motion_halfpel_index;
1352 uint8_t *motion_source;
7a095ea6 1353 int plane, first_pixel;
dc4b78d9 1354
7a095ea6 1355 if (slice >= s->c_superblock_height)
dc4b78d9
MM
1356 return;
1357
1358 for (plane = 0; plane < 3; plane++) {
735acf56
DC
1359 uint8_t *output_plane = s->current_frame.data [plane] + s->data_offset[plane];
1360 uint8_t * last_plane = s-> last_frame.data [plane] + s->data_offset[plane];
1361 uint8_t *golden_plane = s-> golden_frame.data [plane] + s->data_offset[plane];
1abbf64e 1362 int stride = s->current_frame.linesize[plane];
1e76a1da
DC
1363 int plane_width = s->width >> (plane && s->chroma_x_shift);
1364 int plane_height = s->height >> (plane && s->chroma_y_shift);
14268254 1365 int8_t (*motion_val)[2] = s->motion_val[!!plane];
7a095ea6 1366
1e76a1da
DC
1367 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1368 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
7a095ea6
DC
1369 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1370
57783884
DC
1371 int fragment_width = s->fragment_width[!!plane];
1372 int fragment_height = s->fragment_height[!!plane];
7a095ea6 1373 int fragment_start = s->fragment_start[plane];
1abbf64e
MN
1374
1375 if (!s->flipped_image) stride = -stride;
161e8cf4
DC
1376 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1377 continue;
dc4b78d9 1378
115329f1 1379
c26abfa5 1380 if(FFABS(stride) > 2048)
dc4b78d9
MM
1381 return; //various tables are fixed size
1382
7a095ea6
DC
1383 /* for each superblock row in the slice (both of them)... */
1384 for (; sb_y < slice_height; sb_y++) {
dc4b78d9 1385
7a095ea6
DC
1386 /* for each superblock in a row... */
1387 for (sb_x = 0; sb_x < slice_width; sb_x++) {
dc4b78d9 1388
7a095ea6
DC
1389 /* for each block in a superblock... */
1390 for (j = 0; j < 16; j++) {
1391 x = 4*sb_x + hilbert_offset[j][0];
1392 y = 4*sb_y + hilbert_offset[j][1];
e8dcd730 1393 fragment = y*fragment_width + x;
7a095ea6 1394
e8dcd730 1395 i = fragment_start + fragment;
7a095ea6
DC
1396
1397 // bounds check
1398 if (x >= fragment_width || y >= fragment_height)
1399 continue;
1400
1401 first_pixel = 8*y*stride + 8*x;
dc4b78d9
MM
1402
1403 /* transform if this block was coded */
161e8cf4 1404 if (s->all_fragments[i].coding_method != MODE_COPY) {
dc4b78d9
MM
1405 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1406 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1407 motion_source= golden_plane;
115329f1 1408 else
dc4b78d9
MM
1409 motion_source= last_plane;
1410
735acf56 1411 motion_source += first_pixel;
dc4b78d9
MM
1412 motion_halfpel_index = 0;
1413
1414 /* sort out the motion vector if this fragment is coded
1415 * using a motion vector method */
1416 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1417 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1418 int src_x, src_y;
e8dcd730
AS
1419 motion_x = motion_val[fragment][0];
1420 motion_y = motion_val[fragment][1];
dc4b78d9 1421
7a095ea6
DC
1422 src_x= (motion_x>>1) + 8*x;
1423 src_y= (motion_y>>1) + 8*y;
dc4b78d9
MM
1424
1425 motion_halfpel_index = motion_x & 0x01;
1426 motion_source += (motion_x >> 1);
1427
1428 motion_halfpel_index |= (motion_y & 0x01) << 1;
1429 motion_source += ((motion_y >> 1) * stride);
1430
1431 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1432 uint8_t *temp= s->edge_emu_buffer;
1433 if(stride<0) temp -= 9*stride;
1434 else temp += 9*stride;
1435
2e279598 1436 s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
dc4b78d9
MM
1437 motion_source= temp;
1438 }
1439 }
115329f1 1440
dc4b78d9
MM
1441
1442 /* first, take care of copying a block from either the
1443 * previous or the golden frame */
1444 if (s->all_fragments[i].coding_method != MODE_INTRA) {
115329f1
DB
1445 /* Note, it is possible to implement all MC cases with
1446 put_no_rnd_pixels_l2 which would look more like the
1447 VP3 source but this would be slower as
dc4b78d9
MM
1448 put_no_rnd_pixels_tab is better optimzed */
1449 if(motion_halfpel_index != 3){
1450 s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
735acf56 1451 output_plane + first_pixel,
dc4b78d9
MM
1452 motion_source, stride, 8);
1453 }else{
1454 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1455 s->dsp.put_no_rnd_pixels_l2[1](
735acf56 1456 output_plane + first_pixel,
115329f1
DB
1457 motion_source - d,
1458 motion_source + stride + 1 + d,
dc4b78d9
MM
1459 stride, 8);
1460 }
dc4b78d9
MM
1461 }
1462
5fecfb7d 1463 s->dsp.clear_block(block);
dc4b78d9
MM
1464
1465 /* invert DCT and place (or add) in final output */
115329f1 1466
dc4b78d9 1467 if (s->all_fragments[i].coding_method == MODE_INTRA) {
eb6a6cd7 1468 vp3_dequant(s, s->all_fragments + i, plane, 0, block);
dc4b78d9
MM
1469 if(s->avctx->idct_algo!=FF_IDCT_VP3)
1470 block[0] += 128<<3;
1471 s->dsp.idct_put(
735acf56 1472 output_plane + first_pixel,
dc4b78d9
MM
1473 stride,
1474 block);
1475 } else {
eb6a6cd7 1476 if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
dc4b78d9 1477 s->dsp.idct_add(
735acf56 1478 output_plane + first_pixel,
dc4b78d9
MM
1479 stride,
1480 block);
eb6a6cd7
DC
1481 } else {
1482 s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1483 }
dc4b78d9 1484 }
dc4b78d9
MM
1485 } else {
1486
1487 /* copy directly from the previous frame */
1488 s->dsp.put_pixels_tab[1][0](
735acf56
DC
1489 output_plane + first_pixel,
1490 last_plane + first_pixel,
dc4b78d9
MM
1491 stride, 8);
1492
1493 }
7a095ea6 1494 }
dc4b78d9 1495 }
7a095ea6
DC
1496
1497 // Filter up to the last row in the superblock row
a4501a45
DC
1498 if (!s->skip_loop_filter)
1499 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
dc4b78d9
MM
1500 }
1501 }
1502
dc4b78d9
MM
1503 /* this looks like a good place for slice dispatch... */
1504 /* algorithm:
dc4b78d9 1505 * if (slice == s->macroblock_height - 1)
f44b08a5
MM
1506 * dispatch (both last slice & 2nd-to-last slice);
1507 * else if (slice > 0)
1508 * dispatch (slice - 1);
dc4b78d9
MM
1509 */
1510
3d487db1 1511 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
dc4b78d9
MM
1512}
1513
edbb0c07
AS
1514/// Allocate tables for per-frame data in Vp3DecodeContext
1515static av_cold int allocate_tables(AVCodecContext *avctx)
1516{
1517 Vp3DecodeContext *s = avctx->priv_data;
1518 int y_fragment_count, c_fragment_count;
1519
1520 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1521 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1522
1523 s->superblock_coding = av_malloc(s->superblock_count);
1524 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1525 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1526 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1527 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1528 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1529
1530 /* work out the block mapping tables */
1531 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1532 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1533
1534 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1535 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1536 !s->motion_val[0] || !s->motion_val[1]) {
1537 vp3_decode_end(avctx);
1538 return -1;
1539 }
1540
1541 init_block_mapping(s);
1542
1543 return 0;
1544}
1545
115329f1 1546/*
d86053a4
MM
1547 * This is the ffmpeg/libavcodec API init function.
1548 */
98a6fff9 1549static av_cold int vp3_decode_init(AVCodecContext *avctx)
d86053a4
MM
1550{
1551 Vp3DecodeContext *s = avctx->priv_data;
ae1dd8e1 1552 int i, inter, plane;
892fc83e
MM
1553 int c_width;
1554 int c_height;
1e76a1da 1555 int y_fragment_count, c_fragment_count;
d86053a4 1556
3c3f113e 1557 if (avctx->codec_tag == MKTAG('V','P','3','0'))
bb270c08 1558 s->version = 0;
3c3f113e 1559 else
bb270c08 1560 s->version = 1;
3c3f113e 1561
d86053a4 1562 s->avctx = avctx;
ef516f73
DC
1563 s->width = FFALIGN(avctx->width, 16);
1564 s->height = FFALIGN(avctx->height, 16);
1e76a1da
DC
1565 if (avctx->pix_fmt == PIX_FMT_NONE)
1566 avctx->pix_fmt = PIX_FMT_YUV420P;
580a7465 1567 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
8b6103da
MN
1568 if(avctx->idct_algo==FF_IDCT_AUTO)
1569 avctx->idct_algo=FF_IDCT_VP3;
d86053a4 1570 dsputil_init(&s->dsp, avctx);
115329f1 1571
36af0c95 1572 ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
d86053a4
MM
1573
1574 /* initialize to an impossible value which will force a recalculation
1575 * in the first frame decode */
f2264fa5
DC
1576 for (i = 0; i < 3; i++)
1577 s->qps[i] = -1;
d86053a4 1578
1e76a1da
DC
1579 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1580
892fc83e
MM
1581 s->y_superblock_width = (s->width + 31) / 32;
1582 s->y_superblock_height = (s->height + 31) / 32;
35c28d23 1583 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
892fc83e
MM
1584
1585 /* work out the dimensions for the C planes */
1e76a1da
DC
1586 c_width = s->width >> s->chroma_x_shift;
1587 c_height = s->height >> s->chroma_y_shift;
892fc83e
MM
1588 s->c_superblock_width = (c_width + 31) / 32;
1589 s->c_superblock_height = (c_height + 31) / 32;
35c28d23 1590 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
892fc83e 1591
35c28d23
DC
1592 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1593 s->u_superblock_start = s->y_superblock_count;
1594 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
d86053a4
MM
1595
1596 s->macroblock_width = (s->width + 15) / 16;
1597 s->macroblock_height = (s->height + 15) / 16;
1598 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1599
57783884
DC
1600 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1601 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1e76a1da
DC
1602 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1603 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
d86053a4
MM
1604
1605 /* fragment count covers all 8x8 blocks for all 3 planes */
1e76a1da
DC
1606 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1607 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1608 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1609 s->fragment_start[1] = y_fragment_count;
1610 s->fragment_start[2] = y_fragment_count + c_fragment_count;
d86053a4 1611
f44ee2c3
AB
1612 if (!s->theora_tables)
1613 {
2287c100 1614 for (i = 0; i < 64; i++) {
bb270c08 1615 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
bb270c08 1616 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
ae1dd8e1 1617 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
ae1dd8e1 1618 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
ae1dd8e1 1619 s->base_matrix[2][i] = vp31_inter_dequant[i];
bb270c08 1620 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2287c100 1621 }
f44ee2c3 1622
ae1dd8e1
MN
1623 for(inter=0; inter<2; inter++){
1624 for(plane=0; plane<3; plane++){
1625 s->qr_count[inter][plane]= 1;
1626 s->qr_size [inter][plane][0]= 63;
1627 s->qr_base [inter][plane][0]=
1628 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1629 }
1630 }
1631
39922395
MM
1632 /* init VLC tables */
1633 for (i = 0; i < 16; i++) {
1634
1635 /* DC histograms */
8e6daa4a 1636 init_vlc(&s->dc_vlc[i], 11, 32,
39922395
MM
1637 &dc_bias[i][0][1], 4, 2,
1638 &dc_bias[i][0][0], 4, 2, 0);
1639
1640 /* group 1 AC histograms */
8e6daa4a 1641 init_vlc(&s->ac_vlc_1[i], 11, 32,
39922395
MM
1642 &ac_bias_0[i][0][1], 4, 2,
1643 &ac_bias_0[i][0][0], 4, 2, 0);
1644
1645 /* group 2 AC histograms */
8e6daa4a 1646 init_vlc(&s->ac_vlc_2[i], 11, 32,
39922395
MM
1647 &ac_bias_1[i][0][1], 4, 2,
1648 &ac_bias_1[i][0][0], 4, 2, 0);
1649
1650 /* group 3 AC histograms */
8e6daa4a 1651 init_vlc(&s->ac_vlc_3[i], 11, 32,
39922395
MM
1652 &ac_bias_2[i][0][1], 4, 2,
1653 &ac_bias_2[i][0][0], 4, 2, 0);
1654
1655 /* group 4 AC histograms */
8e6daa4a 1656 init_vlc(&s->ac_vlc_4[i], 11, 32,
39922395
MM
1657 &ac_bias_3[i][0][1], 4, 2,
1658 &ac_bias_3[i][0][0], 4, 2, 0);
1659 }
1660 } else {
39922395 1661
8e6daa4a 1662 for (i = 0; i < 16; i++) {
39922395 1663 /* DC histograms */
8e6daa4a
DC
1664 if (init_vlc(&s->dc_vlc[i], 11, 32,
1665 &s->huffman_table[i][0][1], 8, 4,
1666 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
c4b7b8bf 1667 goto vlc_fail;
39922395
MM
1668
1669 /* group 1 AC histograms */
8e6daa4a
DC
1670 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1671 &s->huffman_table[i+16][0][1], 8, 4,
1672 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
c4b7b8bf 1673 goto vlc_fail;
39922395
MM
1674
1675 /* group 2 AC histograms */
8e6daa4a
DC
1676 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1677 &s->huffman_table[i+16*2][0][1], 8, 4,
1678 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
c4b7b8bf 1679 goto vlc_fail;
39922395
MM
1680
1681 /* group 3 AC histograms */
8e6daa4a
DC
1682 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1683 &s->huffman_table[i+16*3][0][1], 8, 4,
1684 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
c4b7b8bf 1685 goto vlc_fail;
39922395
MM
1686
1687 /* group 4 AC histograms */
8e6daa4a
DC
1688 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1689 &s->huffman_table[i+16*4][0][1], 8, 4,
1690 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
c4b7b8bf 1691 goto vlc_fail;
39922395 1692 }
d86053a4
MM
1693 }
1694
d8278bab
MM
1695 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1696 &superblock_run_length_vlc_table[0][1], 4, 2,
1697 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1698
dd36b667 1699 init_vlc(&s->fragment_run_length_vlc, 5, 30,
0ad72bdd
MM
1700 &fragment_run_length_vlc_table[0][1], 4, 2,
1701 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1702
1703 init_vlc(&s->mode_code_vlc, 3, 8,
1704 &mode_code_vlc_table[0][1], 2, 1,
1705 &mode_code_vlc_table[0][0], 2, 1, 0);
1706
1707 init_vlc(&s->motion_vector_vlc, 6, 63,
1708 &motion_vector_vlc_table[0][1], 2, 1,
1709 &motion_vector_vlc_table[0][0], 2, 1, 0);
1710
44ae98dd
MM
1711 for (i = 0; i < 3; i++) {
1712 s->current_frame.data[i] = NULL;
1713 s->last_frame.data[i] = NULL;
1714 s->golden_frame.data[i] = NULL;
61873c4a
MM
1715 }
1716
edbb0c07 1717 return allocate_tables(avctx);
c4b7b8bf
RD
1718
1719vlc_fail:
1720 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1721 return -1;
d86053a4
MM
1722}
1723
1724/*
1725 * This is the ffmpeg/libavcodec API frame decode function.
1726 */
115329f1 1727static int vp3_decode_frame(AVCodecContext *avctx,
d86053a4 1728 void *data, int *data_size,
7a00bbad 1729 AVPacket *avpkt)
d86053a4 1730{
7a00bbad
TB
1731 const uint8_t *buf = avpkt->data;
1732 int buf_size = avpkt->size;
d86053a4
MM
1733 Vp3DecodeContext *s = avctx->priv_data;
1734 GetBitContext gb;
dc4b78d9 1735 int i;
d86053a4 1736
d86053a4 1737 init_get_bits(&gb, buf, buf_size * 8);
115329f1 1738
f44ee2c3
AB
1739 if (s->theora && get_bits1(&gb))
1740 {
bb270c08
DB
1741 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1742 return -1;
f44ee2c3 1743 }
3c3f113e
AB
1744
1745 s->keyframe = !get_bits1(&gb);
1746 if (!s->theora)
bb270c08 1747 skip_bits(&gb, 1);
f2264fa5
DC
1748 for (i = 0; i < 3; i++)
1749 s->last_qps[i] = s->qps[i];
efea8528 1750
f2264fa5 1751 s->nqps=0;
efea8528 1752 do{
f2264fa5
DC
1753 s->qps[s->nqps++]= get_bits(&gb, 6);
1754 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1755 for (i = s->nqps; i < 3; i++)
1756 s->qps[i] = -1;
d86053a4 1757
f8830383 1758 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
bb270c08 1759 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
6ee99a7e 1760 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
d86053a4 1761
a4501a45
DC
1762 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1763 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1764
f2264fa5 1765 if (s->qps[0] != s->last_qps[0])
f44b08a5 1766 init_loop_filter(s);
f2264fa5
DC
1767
1768 for (i = 0; i < s->nqps; i++)
1769 // reinit all dequantizers if the first one changed, because
1770 // the DC of the first quantizer must be used for all matrices
1771 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1772 init_dequantizer(s, i);
642d7e84 1773
068e82ba
DC
1774 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1775 return buf_size;
1776
739b5090 1777 s->current_frame.reference = 3;
ff65969f 1778 s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
739b5090
DC
1779 if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1780 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
e72d2d12 1781 goto error;
739b5090
DC
1782 }
1783
d86053a4 1784 if (s->keyframe) {
bb270c08
DB
1785 if (!s->theora)
1786 {
1787 skip_bits(&gb, 4); /* width code */
1788 skip_bits(&gb, 4); /* height code */
1789 if (s->version)
1790 {
1791 s->version = get_bits(&gb, 5);
6ee99a7e 1792 if (avctx->frame_number == 0)
bb270c08
DB
1793 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1794 }
1795 }
1796 if (s->version || s->theora)
1797 {
1798 if (get_bits1(&gb))
1799 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1800 skip_bits(&gb, 2); /* reserved? */
1801 }
d86053a4 1802 } else {
735acf56 1803 if (!s->golden_frame.data[0]) {
7a4e8b59 1804 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
ff65969f 1805
7a4e8b59 1806 s->golden_frame.reference = 3;
ff65969f 1807 s->golden_frame.pict_type = FF_I_TYPE;
7a4e8b59
DC
1808 if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1809 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1810 goto error;
1811 }
1812 s->last_frame = s->golden_frame;
1813 s->last_frame.type = FF_BUFFER_TYPE_COPY;
d86053a4 1814 }
d86053a4
MM
1815 }
1816
b928ec64
MN
1817 s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1818 s->current_frame.qstride= 0;
1819
703acd54 1820 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
d86053a4 1821
220a6f40
MN
1822 if (unpack_superblocks(s, &gb)){
1823 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
e72d2d12 1824 goto error;
220a6f40 1825 }
220a6f40
MN
1826 if (unpack_modes(s, &gb)){
1827 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
e72d2d12 1828 goto error;
220a6f40 1829 }
220a6f40
MN
1830 if (unpack_vectors(s, &gb)){
1831 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
e72d2d12 1832 goto error;
220a6f40 1833 }
f2264fa5
DC
1834 if (unpack_block_qpis(s, &gb)){
1835 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
e72d2d12 1836 goto error;
f2264fa5 1837 }
220a6f40
MN
1838 if (unpack_dct_coeffs(s, &gb)){
1839 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
e72d2d12 1840 goto error;
892fc83e 1841 }
735acf56
DC
1842
1843 for (i = 0; i < 3; i++) {
1e76a1da 1844 int height = s->height >> (i && s->chroma_y_shift);
735acf56
DC
1845 if (s->flipped_image)
1846 s->data_offset[i] = 0;
1847 else
1e76a1da 1848 s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
735acf56 1849 }
d86053a4 1850
a8de3901 1851 s->last_slice_end = 0;
7a095ea6 1852 for (i = 0; i < s->c_superblock_height; i++)
dc4b78d9 1853 render_slice(s, i);
d86053a4 1854
256c0662
DC
1855 // filter the last row
1856 for (i = 0; i < 3; i++) {
1e76a1da 1857 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
256c0662
DC
1858 apply_loop_filter(s, i, row, row+1);
1859 }
83f72f13 1860 vp3_draw_horiz_band(s, s->avctx->height);
892fc83e 1861
d86053a4
MM
1862 *data_size=sizeof(AVFrame);
1863 *(AVFrame*)data= s->current_frame;
1864
44ae98dd
MM
1865 /* release the last frame, if it is allocated and if it is not the
1866 * golden frame */
0c9afacb 1867 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
44ae98dd 1868 avctx->release_buffer(avctx, &s->last_frame);
d86053a4 1869
61873c4a 1870 /* shuffle frames (last = current) */
4501e678 1871 s->last_frame= s->current_frame;
739b5090
DC
1872
1873 if (s->keyframe) {
1874 if (s->golden_frame.data[0])
1875 avctx->release_buffer(avctx, &s->golden_frame);
1876 s->golden_frame = s->current_frame;
0c9afacb 1877 s->last_frame.type = FF_BUFFER_TYPE_COPY;
739b5090
DC
1878 }
1879
8e39d4a7 1880 s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
d86053a4
MM
1881
1882 return buf_size;
e72d2d12
DC
1883
1884error:
1885 if (s->current_frame.data[0])
1886 avctx->release_buffer(avctx, &s->current_frame);
1887 return -1;
d86053a4
MM
1888}
1889
1890/*
1891 * This is the ffmpeg/libavcodec API module cleanup function.
1892 */
98a6fff9 1893static av_cold int vp3_decode_end(AVCodecContext *avctx)
d86053a4
MM
1894{
1895 Vp3DecodeContext *s = avctx->priv_data;
6f4e2b5a 1896 int i;
d86053a4 1897
6f4e2b5a 1898 av_free(s->superblock_coding);
d86053a4 1899 av_free(s->all_fragments);
c72625f2
DC
1900 av_free(s->coded_fragment_list[0]);
1901 av_free(s->dct_tokens_base);
d86053a4 1902 av_free(s->superblock_fragments);
96a7e73b 1903 av_free(s->macroblock_coding);
14268254
DC
1904 av_free(s->motion_val[0]);
1905 av_free(s->motion_val[1]);
115329f1 1906
6f4e2b5a
MR
1907 for (i = 0; i < 16; i++) {
1908 free_vlc(&s->dc_vlc[i]);
1909 free_vlc(&s->ac_vlc_1[i]);
1910 free_vlc(&s->ac_vlc_2[i]);
1911 free_vlc(&s->ac_vlc_3[i]);
1912 free_vlc(&s->ac_vlc_4[i]);
1913 }
1914
1915 free_vlc(&s->superblock_run_length_vlc);
1916 free_vlc(&s->fragment_run_length_vlc);
1917 free_vlc(&s->mode_code_vlc);
1918 free_vlc(&s->motion_vector_vlc);
1919
d86053a4 1920 /* release all frames */
0c9afacb 1921 if (s->golden_frame.data[0])
892fc83e 1922 avctx->release_buffer(avctx, &s->golden_frame);
0c9afacb 1923 if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
892fc83e
MM
1924 avctx->release_buffer(avctx, &s->last_frame);
1925 /* no need to release the current_frame since it will always be pointing
1926 * to the same frame as either the golden or last frame */
d86053a4
MM
1927
1928 return 0;
1929}
1930
39922395
MM
1931static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1932{
1933 Vp3DecodeContext *s = avctx->priv_data;
1934
5fc32c27 1935 if (get_bits1(gb)) {
39922395
MM
1936 int token;
1937 if (s->entries >= 32) { /* overflow */
1938 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1939 return -1;
1940 }
1941 token = get_bits(gb, 5);
1942 //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size);
1943 s->huffman_table[s->hti][token][0] = s->hbits;
1944 s->huffman_table[s->hti][token][1] = s->huff_code_size;
1945 s->entries++;
1946 }
1947 else {
1948 if (s->huff_code_size >= 32) {/* overflow */
1949 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1950 return -1;
1951 }
1952 s->huff_code_size++;
1953 s->hbits <<= 1;
00bbe276
AC
1954 if (read_huffman_tree(avctx, gb))
1955 return -1;
39922395 1956 s->hbits |= 1;
00bbe276
AC
1957 if (read_huffman_tree(avctx, gb))
1958 return -1;
39922395
MM
1959 s->hbits >>= 1;
1960 s->huff_code_size--;
1961 }
1962 return 0;
1963}
1964
b250f9c6 1965#if CONFIG_THEORA_DECODER
1e76a1da
DC
1966static const enum PixelFormat theora_pix_fmts[4] = {
1967 PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
1968};
1969
e278056f 1970static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
1971{
1972 Vp3DecodeContext *s = avctx->priv_data;
ea3c2d53 1973 int visible_width, visible_height, colorspace;
ddc7e438 1974 int offset_x = 0, offset_y = 0;
6974952d 1975 AVRational fps, aspect;
9a7ad925 1976
e278056f 1977 s->theora = get_bits_long(gb, 24);
356306ac 1978 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
105c3d25 1979
ba7ee4a4 1980 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
9a7ad925 1981 /* but previous versions have the image flipped relative to vp3 */
ba7ee4a4 1982 if (s->theora < 0x030200)
9a7ad925 1983 {
bb270c08 1984 s->flipped_image = 1;
9a7ad925
AB
1985 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
1986 }
f44ee2c3 1987
277e3e53
DC
1988 visible_width = s->width = get_bits(gb, 16) << 4;
1989 visible_height = s->height = get_bits(gb, 16) << 4;
115329f1 1990
e16f217c 1991 if(av_image_check_size(s->width, s->height, 0, avctx)){
7146d2c2 1992 av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
0ecca7a4
MN
1993 s->width= s->height= 0;
1994 return -1;
1995 }
7146d2c2 1996
277e3e53 1997 if (s->theora >= 0x030200) {
a0ce2d1b
DC
1998 visible_width = get_bits_long(gb, 24);
1999 visible_height = get_bits_long(gb, 24);
c0f716b8 2000
ddc7e438
DC
2001 offset_x = get_bits(gb, 8); /* offset x */
2002 offset_y = get_bits(gb, 8); /* offset y, from bottom */
ba4816a0 2003 }
f44ee2c3 2004
8099d6c9
DC
2005 fps.num = get_bits_long(gb, 32);
2006 fps.den = get_bits_long(gb, 32);
2007 if (fps.num && fps.den) {
33e62383
DC
2008 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2009 fps.den, fps.num, 1<<30);
8099d6c9
DC
2010 }
2011
6974952d
RD
2012 aspect.num = get_bits_long(gb, 24);
2013 aspect.den = get_bits_long(gb, 24);
2014 if (aspect.num && aspect.den) {
2015 av_reduce(&avctx->sample_aspect_ratio.num,
2016 &avctx->sample_aspect_ratio.den,
2017 aspect.num, aspect.den, 1<<30);
2018 }
115329f1 2019
ba7ee4a4 2020 if (s->theora < 0x030200)
e278056f 2021 skip_bits(gb, 5); /* keyframe frequency force */
ea3c2d53 2022 colorspace = get_bits(gb, 8);
e278056f 2023 skip_bits(gb, 24); /* bitrate */
f44ee2c3 2024
e278056f 2025 skip_bits(gb, 6); /* quality hint */
115329f1 2026
ba7ee4a4 2027 if (s->theora >= 0x030200)
105c3d25 2028 {
e278056f 2029 skip_bits(gb, 5); /* keyframe frequency force */
1e76a1da 2030 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
337f5c6e 2031 skip_bits(gb, 3); /* reserved */
105c3d25 2032 }
115329f1 2033
e278056f 2034// align_get_bits(gb);
115329f1 2035
c0f716b8 2036 if ( visible_width <= s->width && visible_width > s->width-16
ddc7e438
DC
2037 && visible_height <= s->height && visible_height > s->height-16
2038 && !offset_x && (offset_y == s->height - visible_height))
c0f716b8
AJ
2039 avcodec_set_dimensions(avctx, visible_width, visible_height);
2040 else
2041 avcodec_set_dimensions(avctx, s->width, s->height);
f44ee2c3 2042
ea3c2d53
DC
2043 if (colorspace == 1) {
2044 avctx->color_primaries = AVCOL_PRI_BT470M;
2045 } else if (colorspace == 2) {
2046 avctx->color_primaries = AVCOL_PRI_BT470BG;
2047 }
2048 if (colorspace == 1 || colorspace == 2) {
2049 avctx->colorspace = AVCOL_SPC_BT470BG;
2050 avctx->color_trc = AVCOL_TRC_BT709;
2051 }
2052
f44ee2c3
AB
2053 return 0;
2054}
2055
e278056f 2056static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
2057{
2058 Vp3DecodeContext *s = avctx->priv_data;
ae1dd8e1 2059 int i, n, matrices, inter, plane;
ba7ee4a4
MC
2060
2061 if (s->theora >= 0x030200) {
e278056f 2062 n = get_bits(gb, 3);
9c7154c7 2063 /* loop filter limit values table */
9d777508
JGG
2064 if (n)
2065 for (i = 0; i < 64; i++)
10f69158 2066 s->filter_limit_values[i] = get_bits(gb, n);
ba7ee4a4 2067 }
115329f1 2068
ba7ee4a4 2069 if (s->theora >= 0x030200)
e278056f 2070 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2071 else
2072 n = 16;
f44ee2c3
AB
2073 /* quality threshold table */
2074 for (i = 0; i < 64; i++)
e278056f 2075 s->coded_ac_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2076
ba7ee4a4 2077 if (s->theora >= 0x030200)
e278056f 2078 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2079 else
2080 n = 16;
f44ee2c3
AB
2081 /* dc scale factor table */
2082 for (i = 0; i < 64; i++)
e278056f 2083 s->coded_dc_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2084
ba7ee4a4 2085 if (s->theora >= 0x030200)
e278056f 2086 matrices = get_bits(gb, 9) + 1;
ba7ee4a4 2087 else
2da2ba03 2088 matrices = 3;
f44ee2c3 2089
ae1dd8e1
MN
2090 if(matrices > 384){
2091 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2092 return -1;
2093 }
3c3f113e 2094
ae1dd8e1 2095 for(n=0; n<matrices; n++){
bb270c08 2096 for (i = 0; i < 64; i++)
ae1dd8e1
MN
2097 s->base_matrix[n][i]= get_bits(gb, 8);
2098 }
2da2ba03 2099
ae1dd8e1
MN
2100 for (inter = 0; inter <= 1; inter++) {
2101 for (plane = 0; plane <= 2; plane++) {
2102 int newqr= 1;
2103 if (inter || plane > 0)
5fc32c27 2104 newqr = get_bits1(gb);
39922395 2105 if (!newqr) {
ae1dd8e1 2106 int qtj, plj;
5fc32c27 2107 if(inter && get_bits1(gb)){
ae1dd8e1
MN
2108 qtj = 0;
2109 plj = plane;
2110 }else{
2111 qtj= (3*inter + plane - 1) / 3;
2112 plj= (plane + 2) % 3;
2113 }
2114 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2115 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2116 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2117 } else {
2118 int qri= 0;
39922395 2119 int qi = 0;
ae1dd8e1
MN
2120
2121 for(;;){
2122 i= get_bits(gb, av_log2(matrices-1)+1);
2123 if(i>= matrices){
2124 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2125 return -1;
2126 }
2127 s->qr_base[inter][plane][qri]= i;
2128 if(qi >= 63)
2129 break;
2130 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2131 s->qr_size[inter][plane][qri++]= i;
2132 qi += i;
39922395 2133 }
ae1dd8e1 2134
2da2ba03 2135 if (qi > 63) {
7146d2c2 2136 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
bb270c08
DB
2137 return -1;
2138 }
ae1dd8e1 2139 s->qr_count[inter][plane]= qri;
39922395
MM
2140 }
2141 }
2142 }
2143
2da2ba03 2144 /* Huffman tables */
39922395
MM
2145 for (s->hti = 0; s->hti < 80; s->hti++) {
2146 s->entries = 0;
2147 s->huff_code_size = 1;
5fc32c27 2148 if (!get_bits1(gb)) {
39922395 2149 s->hbits = 0;
00bbe276
AC
2150 if(read_huffman_tree(avctx, gb))
2151 return -1;
39922395 2152 s->hbits = 1;
00bbe276
AC
2153 if(read_huffman_tree(avctx, gb))
2154 return -1;
39922395
MM
2155 }
2156 }
115329f1 2157
f44ee2c3 2158 s->theora_tables = 1;
115329f1 2159
f44ee2c3
AB
2160 return 0;
2161}
2162
5ef251e5 2163static av_cold int theora_decode_init(AVCodecContext *avctx)
f44ee2c3
AB
2164{
2165 Vp3DecodeContext *s = avctx->priv_data;
2166 GetBitContext gb;
2167 int ptype;
da91ed59
AJ
2168 uint8_t *header_start[3];
2169 int header_len[3];
2170 int i;
115329f1 2171
f44ee2c3
AB
2172 s->theora = 1;
2173
2174 if (!avctx->extradata_size)
7146d2c2
AB
2175 {
2176 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
bb270c08 2177 return -1;
7146d2c2 2178 }
f44ee2c3 2179
da91ed59
AJ
2180 if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2181 42, header_start, header_len) < 0) {
2182 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2183 return -1;
2184 }
ee89b2b9 2185
da91ed59 2186 for(i=0;i<3;i++) {
fa6f2751 2187 init_get_bits(&gb, header_start[i], header_len[i] * 8);
f44ee2c3
AB
2188
2189 ptype = get_bits(&gb, 8);
115329f1 2190
7146d2c2
AB
2191 if (!(ptype & 0x80))
2192 {
2193 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
e278056f 2194// return -1;
115329f1 2195 }
7146d2c2 2196
3700dab4 2197 // FIXME: Check for this as well.
0a8dedc9 2198 skip_bits_long(&gb, 6*8); /* "theora" */
115329f1 2199
f44ee2c3
AB
2200 switch(ptype)
2201 {
2202 case 0x80:
e278056f 2203 theora_decode_header(avctx, &gb);
bb270c08
DB
2204 break;
2205 case 0x81:
2da2ba03 2206// FIXME: is this needed? it breaks sometimes
bb270c08
DB
2207// theora_decode_comments(avctx, gb);
2208 break;
2209 case 0x82:
00bbe276
AC
2210 if (theora_decode_tables(avctx, &gb))
2211 return -1;
bb270c08
DB
2212 break;
2213 default:
2214 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2215 break;
f44ee2c3 2216 }
12ce1f3f
DC
2217 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2218 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
116d866c
MC
2219 if (s->theora < 0x030200)
2220 break;
ee89b2b9 2221 }
f44ee2c3 2222
c79c960a 2223 return vp3_decode_init(avctx);
f44ee2c3
AB
2224}
2225
d36beb3f 2226AVCodec ff_theora_decoder = {
6f6a3e2a 2227 "theora",
72415b2a 2228 AVMEDIA_TYPE_VIDEO,
6f6a3e2a 2229 CODEC_ID_THEORA,
d86053a4 2230 sizeof(Vp3DecodeContext),
6f6a3e2a 2231 theora_decode_init,
d86053a4
MM
2232 NULL,
2233 vp3_decode_end,
2234 vp3_decode_frame,
a8de3901 2235 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
d5202e4f 2236 NULL,
fe4bf374 2237 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
d86053a4 2238};
6f6a3e2a 2239#endif
f44ee2c3 2240
d36beb3f 2241AVCodec ff_vp3_decoder = {
6f6a3e2a 2242 "vp3",
72415b2a 2243 AVMEDIA_TYPE_VIDEO,
6f6a3e2a 2244 CODEC_ID_VP3,
f44ee2c3 2245 sizeof(Vp3DecodeContext),
6f6a3e2a 2246 vp3_decode_init,
f44ee2c3
AB
2247 NULL,
2248 vp3_decode_end,
2249 vp3_decode_frame,
a8de3901 2250 CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
d5202e4f 2251 NULL,
fe4bf374 2252 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
f44ee2c3 2253};