Give IDCT matrix transpose macro a more descriptive name
[libav.git] / libavcodec / vp3.c
CommitLineData
d86053a4 1/*
67335dbc 2 * Copyright (C) 2003-2004 the ffmpeg project
d86053a4 3 *
2912e87a 4 * This file is part of Libav.
b78e7197 5 *
2912e87a 6 * Libav 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 *
2912e87a 11 * Libav 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
2912e87a 17 * License along with Libav; 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
737eb597 36#include "libavutil/imgutils.h"
d86053a4 37#include "avcodec.h"
f3a29b75 38#include "internal.h"
d86053a4 39#include "dsputil.h"
9106a698 40#include "get_bits.h"
3bd062bf 41#include "hpeldsp.h"
8c53d39e 42#include "videodsp.h"
d86053a4 43#include "vp3data.h"
28f9ab70 44#include "vp3dsp.h"
da91ed59 45#include "xiph.h"
d23845f3 46#include "thread.h"
d86053a4
MM
47
48#define FRAGMENT_PIXELS 8
49
7beddb12 50//FIXME split things out into their own arrays
d86053a4 51typedef struct Vp3Fragment {
c72625f2 52 int16_t dc;
288774bb 53 uint8_t coding_method;
f2264fa5 54 uint8_t qpi;
d86053a4
MM
55} Vp3Fragment;
56
57#define SB_NOT_CODED 0
58#define SB_PARTIALLY_CODED 1
59#define SB_FULLY_CODED 2
60
ecb51b25
DC
61// This is the maximum length of a single long bit run that can be encoded
62// for superblock coding or block qps. Theora special-cases this to read a
63// bit instead of flipping the current bit to allow for runs longer than 4129.
64#define MAXIMUM_LONG_BIT_RUN 4129
65
d86053a4
MM
66#define MODE_INTER_NO_MV 0
67#define MODE_INTRA 1
68#define MODE_INTER_PLUS_MV 2
69#define MODE_INTER_LAST_MV 3
70#define MODE_INTER_PRIOR_LAST 4
71#define MODE_USING_GOLDEN 5
72#define MODE_GOLDEN_MV 6
73#define MODE_INTER_FOURMV 7
74#define CODING_MODE_COUNT 8
75
76/* special internal mode */
77#define MODE_COPY 8
78
79/* There are 6 preset schemes, plus a free-form scheme */
e8e47435 80static const int ModeAlphabet[6][CODING_MODE_COUNT] =
d86053a4 81{
d86053a4 82 /* scheme 1: Last motion vector dominates */
115329f1 83 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
d86053a4 84 MODE_INTER_PLUS_MV, MODE_INTER_NO_MV,
115329f1 85 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
86 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
87
88 /* scheme 2 */
115329f1 89 { MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
d86053a4 90 MODE_INTER_NO_MV, MODE_INTER_PLUS_MV,
115329f1 91 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
92 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
93
94 /* scheme 3 */
115329f1 95 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
d86053a4 96 MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
115329f1 97 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
98 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
99
100 /* scheme 4 */
115329f1 101 { MODE_INTER_LAST_MV, MODE_INTER_PLUS_MV,
d86053a4 102 MODE_INTER_NO_MV, MODE_INTER_PRIOR_LAST,
115329f1 103 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
104 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
105
106 /* scheme 5: No motion vector dominates */
115329f1 107 { MODE_INTER_NO_MV, MODE_INTER_LAST_MV,
d86053a4 108 MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
115329f1 109 MODE_INTRA, MODE_USING_GOLDEN,
d86053a4
MM
110 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
111
112 /* scheme 6 */
115329f1 113 { MODE_INTER_NO_MV, MODE_USING_GOLDEN,
d86053a4 114 MODE_INTER_LAST_MV, MODE_INTER_PRIOR_LAST,
115329f1 115 MODE_INTER_PLUS_MV, MODE_INTRA,
d86053a4
MM
116 MODE_GOLDEN_MV, MODE_INTER_FOURMV },
117
118};
119
7a095ea6
DC
120static const uint8_t hilbert_offset[16][2] = {
121 {0,0}, {1,0}, {1,1}, {0,1},
122 {0,2}, {0,3}, {1,3}, {1,2},
123 {2,2}, {2,3}, {3,3}, {3,2},
124 {3,1}, {2,1}, {2,0}, {3,0}
125};
126
d86053a4
MM
127#define MIN_DEQUANT_VAL 2
128
129typedef struct Vp3DecodeContext {
130 AVCodecContext *avctx;
f44ee2c3 131 int theora, theora_tables;
3c3f113e 132 int version;
d86053a4 133 int width, height;
1e76a1da 134 int chroma_x_shift, chroma_y_shift;
759001c5
AK
135 ThreadFrame golden_frame;
136 ThreadFrame last_frame;
137 ThreadFrame current_frame;
d86053a4 138 int keyframe;
18df366a 139 uint8_t idct_permutation[64];
01582122 140 uint8_t idct_scantable[64];
3bd062bf 141 HpelDSPContext hdsp;
8c53d39e 142 VideoDSPContext vdsp;
28f9ab70 143 VP3DSPContext vp3dsp;
88bd7fdc 144 DECLARE_ALIGNED(16, int16_t, block)[64];
9a7ad925 145 int flipped_image;
a8de3901 146 int last_slice_end;
a4501a45 147 int skip_loop_filter;
d86053a4 148
f2264fa5
DC
149 int qps[3];
150 int nqps;
151 int last_qps[3];
d86053a4
MM
152
153 int superblock_count;
892fc83e
MM
154 int y_superblock_width;
155 int y_superblock_height;
35c28d23 156 int y_superblock_count;
892fc83e
MM
157 int c_superblock_width;
158 int c_superblock_height;
35c28d23 159 int c_superblock_count;
d86053a4
MM
160 int u_superblock_start;
161 int v_superblock_start;
162 unsigned char *superblock_coding;
163
164 int macroblock_count;
165 int macroblock_width;
166 int macroblock_height;
167
168 int fragment_count;
57783884
DC
169 int fragment_width[2];
170 int fragment_height[2];
d86053a4
MM
171
172 Vp3Fragment *all_fragments;
1abbf64e 173 int fragment_start[3];
735acf56 174 int data_offset[3];
115329f1 175
14268254
DC
176 int8_t (*motion_val[2])[2];
177
f44ee2c3
AB
178 /* tables */
179 uint16_t coded_dc_scale_factor[64];
67335dbc 180 uint32_t coded_ac_scale_factor[64];
ae1dd8e1
MN
181 uint8_t base_matrix[384][64];
182 uint8_t qr_count[2][3];
183 uint8_t qr_size [2][3][64];
184 uint16_t qr_base[2][3][64];
d86053a4 185
c72625f2
DC
186 /**
187 * This is a list of all tokens in bitstream order. Reordering takes place
188 * by pulling from each level during IDCT. As a consequence, IDCT must be
189 * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
190 * otherwise. The 32 different tokens with up to 12 bits of extradata are
191 * collapsed into 3 types, packed as follows:
192 * (from the low to high bits)
193 *
194 * 2 bits: type (0,1,2)
195 * 0: EOB run, 14 bits for run length (12 needed)
196 * 1: zero run, 7 bits for run length
197 * 7 bits for the next coefficient (3 needed)
198 * 2: coefficient, 14 bits (11 needed)
199 *
200 * Coefficients are signed, so are packed in the highest bits for automatic
201 * sign extension.
202 */
203 int16_t *dct_tokens[3][64];
204 int16_t *dct_tokens_base;
205#define TOKEN_EOB(eob_run) ((eob_run) << 2)
206#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
207#define TOKEN_COEFF(coeff) (((coeff) << 2) + 2)
208
209 /**
210 * number of blocks that contain DCT coefficients at the given level or higher
211 */
212 int num_coded_frags[3][64];
213 int total_num_coded_frags;
214
f4433de9 215 /* this is a list of indexes into the all_fragments array indicating
d86053a4 216 * which of the fragments are coded */
c72625f2 217 int *coded_fragment_list[3];
098523eb 218
d86053a4
MM
219 VLC dc_vlc[16];
220 VLC ac_vlc_1[16];
221 VLC ac_vlc_2[16];
222 VLC ac_vlc_3[16];
223 VLC ac_vlc_4[16];
224
0ad72bdd
MM
225 VLC superblock_run_length_vlc;
226 VLC fragment_run_length_vlc;
227 VLC mode_code_vlc;
228 VLC motion_vector_vlc;
229
38acbc3c
MM
230 /* these arrays need to be on 16-byte boundaries since SSE2 operations
231 * index into them */
02494787 232 DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64]; ///< qmat[qpi][is_inter][plane]
d86053a4
MM
233
234 /* This table contains superblock_count * 16 entries. Each set of 16
f4433de9 235 * numbers corresponds to the fragment indexes 0..15 of the superblock.
d86053a4
MM
236 * An entry will be -1 to indicate that no entry corresponds to that
237 * index. */
238 int *superblock_fragments;
239
115329f1 240 /* This is an array that indicates how a particular macroblock
74c0ac12 241 * is coded. */
96a7e73b 242 unsigned char *macroblock_coding;
d86053a4 243
902685b8 244 uint8_t *edge_emu_buffer;
39922395 245
f44b08a5
MM
246 /* Huffman decode */
247 int hti;
248 unsigned int hbits;
249 int entries;
250 int huff_code_size;
8e6daa4a 251 uint32_t huffman_table[80][32][2];
f44b08a5 252
51ace577 253 uint8_t filter_limit_values[64];
84dc2d8a 254 DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
d86053a4
MM
255} Vp3DecodeContext;
256
257/************************************************************************
258 * VP3 specific functions
259 ************************************************************************/
260
bc4d1561
DB
261static void vp3_decode_flush(AVCodecContext *avctx)
262{
263 Vp3DecodeContext *s = avctx->priv_data;
264
759001c5 265 if (s->golden_frame.f)
bc4d1561 266 ff_thread_release_buffer(avctx, &s->golden_frame);
759001c5 267 if (s->last_frame.f)
bc4d1561 268 ff_thread_release_buffer(avctx, &s->last_frame);
759001c5 269 if (s->current_frame.f)
bc4d1561
DB
270 ff_thread_release_buffer(avctx, &s->current_frame);
271}
272
273static av_cold int vp3_decode_end(AVCodecContext *avctx)
274{
275 Vp3DecodeContext *s = avctx->priv_data;
276 int i;
277
ec86ba57
RB
278 av_freep(&s->superblock_coding);
279 av_freep(&s->all_fragments);
280 av_freep(&s->coded_fragment_list[0]);
281 av_freep(&s->dct_tokens_base);
282 av_freep(&s->superblock_fragments);
283 av_freep(&s->macroblock_coding);
284 av_freep(&s->motion_val[0]);
285 av_freep(&s->motion_val[1]);
286 av_freep(&s->edge_emu_buffer);
bc4d1561 287
759001c5
AK
288 /* release all frames */
289 vp3_decode_flush(avctx);
290 av_frame_free(&s->current_frame.f);
291 av_frame_free(&s->last_frame.f);
292 av_frame_free(&s->golden_frame.f);
293
bc4d1561
DB
294 if (avctx->internal->is_copy)
295 return 0;
296
297 for (i = 0; i < 16; i++) {
e96b4a53
MS
298 ff_free_vlc(&s->dc_vlc[i]);
299 ff_free_vlc(&s->ac_vlc_1[i]);
300 ff_free_vlc(&s->ac_vlc_2[i]);
301 ff_free_vlc(&s->ac_vlc_3[i]);
302 ff_free_vlc(&s->ac_vlc_4[i]);
bc4d1561
DB
303 }
304
e96b4a53
MS
305 ff_free_vlc(&s->superblock_run_length_vlc);
306 ff_free_vlc(&s->fragment_run_length_vlc);
307 ff_free_vlc(&s->mode_code_vlc);
308 ff_free_vlc(&s->motion_vector_vlc);
bc4d1561 309
bc4d1561
DB
310
311 return 0;
312}
313
d86053a4
MM
314/*
315 * This function sets up all of the various blocks mappings:
316 * superblocks <-> fragments, macroblocks <-> fragments,
317 * superblocks <-> macroblocks
892fc83e 318 *
49bd8e4b 319 * @return 0 is successful; returns 1 if *anything* went wrong.
d86053a4 320 */
115329f1 321static int init_block_mapping(Vp3DecodeContext *s)
d86053a4 322{
a16389c1
DC
323 int sb_x, sb_y, plane;
324 int x, y, i, j = 0;
325
326 for (plane = 0; plane < 3; plane++) {
327 int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
328 int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
57783884
DC
329 int frag_width = s->fragment_width[!!plane];
330 int frag_height = s->fragment_height[!!plane];
a16389c1
DC
331
332 for (sb_y = 0; sb_y < sb_height; sb_y++)
333 for (sb_x = 0; sb_x < sb_width; sb_x++)
334 for (i = 0; i < 16; i++) {
335 x = 4*sb_x + hilbert_offset[i][0];
336 y = 4*sb_y + hilbert_offset[i][1];
337
338 if (x < frag_width && y < frag_height)
339 s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
340 else
341 s->superblock_fragments[j++] = -1;
342 }
d86053a4
MM
343 }
344
892fc83e 345 return 0; /* successful path out */
d86053a4
MM
346}
347
348/*
f44b08a5 349 * This function sets up the dequantization tables used for a particular
d86053a4
MM
350 * frame.
351 */
f2264fa5 352static void init_dequantizer(Vp3DecodeContext *s, int qpi)
d86053a4 353{
f2264fa5
DC
354 int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
355 int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
36c32bdd 356 int i, plane, inter, qri, bmi, bmj, qistart;
d86053a4 357
ae1dd8e1
MN
358 for(inter=0; inter<2; inter++){
359 for(plane=0; plane<3; plane++){
360 int sum=0;
361 for(qri=0; qri<s->qr_count[inter][plane]; qri++){
362 sum+= s->qr_size[inter][plane][qri];
f2264fa5 363 if(s->qps[qpi] <= sum)
ae1dd8e1
MN
364 break;
365 }
366 qistart= sum - s->qr_size[inter][plane][qri];
367 bmi= s->qr_base[inter][plane][qri ];
368 bmj= s->qr_base[inter][plane][qri+1];
369 for(i=0; i<64; i++){
f2264fa5
DC
370 int coeff= ( 2*(sum -s->qps[qpi])*s->base_matrix[bmi][i]
371 - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
ae1dd8e1
MN
372 + s->qr_size[inter][plane][qri])
373 / (2*s->qr_size[inter][plane][qri]);
374
a14ab4e4 375 int qmin= 8<<(inter + !i);
ae1dd8e1
MN
376 int qscale= i ? ac_scale_factor : dc_scale_factor;
377
18df366a
RB
378 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
379 av_clip((qscale * coeff) / 100 * 4, qmin, 4096);
ae1dd8e1 380 }
f2264fa5
DC
381 // all DC coefficients use the same quant so as not to interfere with DC prediction
382 s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
ae1dd8e1 383 }
d86053a4 384 }
d86053a4
MM
385}
386
387/*
f44b08a5
MM
388 * This function initializes the loop filter boundary limits if the frame's
389 * quality index is different from the previous frame's.
7fa5f999
RD
390 *
391 * The filter_limit_values may not be larger than 127.
f44b08a5
MM
392 */
393static void init_loop_filter(Vp3DecodeContext *s)
394{
395 int *bounding_values= s->bounding_values_array+127;
396 int filter_limit;
397 int x;
7fa5f999 398 int value;
f44b08a5 399
f2264fa5 400 filter_limit = s->filter_limit_values[s->qps[0]];
27c2fb0f 401 assert(filter_limit < 128);
f44b08a5
MM
402
403 /* set up the bounding values */
404 memset(s->bounding_values_array, 0, 256 * sizeof(int));
405 for (x = 0; x < filter_limit; x++) {
f44b08a5
MM
406 bounding_values[-x] = -x;
407 bounding_values[x] = x;
f44b08a5 408 }
7fa5f999
RD
409 for (x = value = filter_limit; x < 128 && value; x++, value--) {
410 bounding_values[ x] = value;
411 bounding_values[-x] = -value;
412 }
413 if (value)
414 bounding_values[128] = value;
357f45d9 415 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
f44b08a5
MM
416}
417
418/*
115329f1 419 * This function unpacks all of the superblock/macroblock/fragment coding
d86053a4
MM
420 * information from the bitstream.
421 */
892fc83e 422static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 423{
e2720b6b 424 int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
d86053a4
MM
425 int bit = 0;
426 int current_superblock = 0;
427 int current_run = 0;
855c720c 428 int num_partial_superblocks = 0;
d86053a4
MM
429
430 int i, j;
431 int current_fragment;
35c28d23 432 int plane;
d86053a4 433
d86053a4 434 if (s->keyframe) {
d86053a4
MM
435 memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
436
437 } else {
438
439 /* unpack the list of partially-coded superblocks */
cc46005f
DC
440 bit = get_bits1(gb) ^ 1;
441 current_run = 0;
442
60867312 443 while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
cc46005f
DC
444 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
445 bit = get_bits1(gb);
446 else
447 bit ^= 1;
448
115329f1 449 current_run = get_vlc2(gb,
33dbc1b7
DC
450 s->superblock_run_length_vlc.table, 6, 2) + 1;
451 if (current_run == 34)
d8278bab 452 current_run += get_bits(gb, 12);
d86053a4 453
33dbc1b7
DC
454 if (current_superblock + current_run > s->superblock_count) {
455 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
456 return -1;
457 }
458
459 memset(s->superblock_coding + current_superblock, bit, current_run);
460
461 current_superblock += current_run;
855c720c
DC
462 if (bit)
463 num_partial_superblocks += current_run;
d86053a4
MM
464 }
465
466 /* unpack the list of fully coded superblocks if any of the blocks were
467 * not marked as partially coded in the previous step */
855c720c
DC
468 if (num_partial_superblocks < s->superblock_count) {
469 int superblocks_decoded = 0;
d86053a4
MM
470
471 current_superblock = 0;
cc46005f
DC
472 bit = get_bits1(gb) ^ 1;
473 current_run = 0;
474
60867312
DC
475 while (superblocks_decoded < s->superblock_count - num_partial_superblocks
476 && get_bits_left(gb) > 0) {
cc46005f
DC
477
478 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
479 bit = get_bits1(gb);
480 else
481 bit ^= 1;
482
115329f1 483 current_run = get_vlc2(gb,
855c720c
DC
484 s->superblock_run_length_vlc.table, 6, 2) + 1;
485 if (current_run == 34)
d8278bab 486 current_run += get_bits(gb, 12);
855c720c
DC
487
488 for (j = 0; j < current_run; current_superblock++) {
489 if (current_superblock >= s->superblock_count) {
490 av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
491 return -1;
d86053a4 492 }
855c720c
DC
493
494 /* skip any superblocks already marked as partially coded */
495 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
b5da3635 496 s->superblock_coding[current_superblock] = 2*bit;
855c720c
DC
497 j++;
498 }
d86053a4 499 }
855c720c 500 superblocks_decoded += current_run;
d86053a4
MM
501 }
502 }
503
504 /* if there were partial blocks, initialize bitstream for
505 * unpacking fragment codings */
855c720c 506 if (num_partial_superblocks) {
d86053a4
MM
507
508 current_run = 0;
5fc32c27 509 bit = get_bits1(gb);
115329f1 510 /* toggle the bit because as soon as the first run length is
d86053a4
MM
511 * fetched the bit will be toggled again */
512 bit ^= 1;
513 }
514 }
515
516 /* figure out which fragments are coded; iterate through each
517 * superblock (all planes) */
c72625f2 518 s->total_num_coded_frags = 0;
96a7e73b 519 memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
35c28d23
DC
520
521 for (plane = 0; plane < 3; plane++) {
e2720b6b 522 int sb_start = superblock_starts[plane];
35c28d23 523 int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
c72625f2 524 int num_coded_frags = 0;
35c28d23 525
60867312 526 for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
d86053a4
MM
527
528 /* iterate through all 16 fragments in a superblock */
529 for (j = 0; j < 16; j++) {
530
531 /* if the fragment is in bounds, check its coding status */
532 current_fragment = s->superblock_fragments[i * 16 + j];
533 if (current_fragment != -1) {
6cb35b45 534 int coded = s->superblock_coding[i];
d86053a4 535
6cb35b45 536 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
d86053a4
MM
537
538 /* fragment may or may not be coded; this is the case
539 * that cares about the fragment coding runs */
b5da3635 540 if (current_run-- == 0) {
d86053a4 541 bit ^= 1;
115329f1 542 current_run = get_vlc2(gb,
b5da3635 543 s->fragment_run_length_vlc.table, 5, 2);
d86053a4 544 }
6cb35b45
DC
545 coded = bit;
546 }
d86053a4 547
6cb35b45 548 if (coded) {
115329f1 549 /* default mode; actual mode will be decoded in
22493ab9 550 * the next phase */
115329f1 551 s->all_fragments[current_fragment].coding_method =
d86053a4 552 MODE_INTER_NO_MV;
c72625f2 553 s->coded_fragment_list[plane][num_coded_frags++] =
d86053a4 554 current_fragment;
d86053a4
MM
555 } else {
556 /* not coded; copy this fragment from the prior frame */
557 s->all_fragments[current_fragment].coding_method =
558 MODE_COPY;
d86053a4 559 }
d86053a4
MM
560 }
561 }
562 }
c72625f2
DC
563 s->total_num_coded_frags += num_coded_frags;
564 for (i = 0; i < 64; i++)
565 s->num_coded_frags[plane][i] = num_coded_frags;
566 if (plane < 2)
567 s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
35c28d23 568 }
892fc83e 569 return 0;
d86053a4
MM
570}
571
572/*
573 * This function unpacks all the coding mode data for individual macroblocks
574 * from the bitstream.
575 */
892fc83e 576static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 577{
19cd517d 578 int i, j, k, sb_x, sb_y;
d86053a4
MM
579 int scheme;
580 int current_macroblock;
581 int current_fragment;
582 int coding_mode;
e8e47435 583 int custom_mode_alphabet[CODING_MODE_COUNT];
7c2e31d1 584 const int *alphabet;
1e76a1da 585 Vp3Fragment *frag;
d86053a4 586
d86053a4 587 if (s->keyframe) {
d86053a4
MM
588 for (i = 0; i < s->fragment_count; i++)
589 s->all_fragments[i].coding_method = MODE_INTRA;
590
591 } else {
592
593 /* fetch the mode coding scheme for this frame */
594 scheme = get_bits(gb, 3);
d86053a4
MM
595
596 /* is it a custom coding scheme? */
597 if (scheme == 0) {
d86053a4 598 for (i = 0; i < 8; i++)
2c823b3c
AC
599 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
600 for (i = 0; i < 8; i++)
e8e47435 601 custom_mode_alphabet[get_bits(gb, 3)] = i;
7c2e31d1
DC
602 alphabet = custom_mode_alphabet;
603 } else
604 alphabet = ModeAlphabet[scheme-1];
d86053a4 605
d86053a4
MM
606 /* iterate through all of the macroblocks that contain 1 or more
607 * coded fragments */
19cd517d
DC
608 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
609 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
60867312
DC
610 if (get_bits_left(gb) <= 0)
611 return -1;
d86053a4
MM
612
613 for (j = 0; j < 4; j++) {
19cd517d
DC
614 int mb_x = 2*sb_x + (j>>1);
615 int mb_y = 2*sb_y + (((j>>1)+j)&1);
616 current_macroblock = mb_y * s->macroblock_width + mb_x;
617
15675ce6 618 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
d86053a4
MM
619 continue;
620
ea676144
DC
621#define BLOCK_X (2*mb_x + (k&1))
622#define BLOCK_Y (2*mb_y + (k>>1))
15675ce6
DC
623 /* coding modes are only stored if the macroblock has at least one
624 * luma block coded, otherwise it must be INTER_NO_MV */
625 for (k = 0; k < 4; k++) {
57783884 626 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
15675ce6
DC
627 if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
628 break;
629 }
630 if (k == 4) {
631 s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
632 continue;
633 }
ea676144 634
d86053a4
MM
635 /* mode 7 means get 3 bits for each coding mode */
636 if (scheme == 7)
637 coding_mode = get_bits(gb, 3);
638 else
7c2e31d1 639 coding_mode = alphabet
0ad72bdd 640 [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
d86053a4 641
96a7e73b 642 s->macroblock_coding[current_macroblock] = coding_mode;
ea676144 643 for (k = 0; k < 4; k++) {
1e76a1da
DC
644 frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
645 if (frag->coding_method != MODE_COPY)
646 frag->coding_method = coding_mode;
ea676144 647 }
1e76a1da
DC
648
649#define SET_CHROMA_MODES \
650 if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
651 frag[s->fragment_start[1]].coding_method = coding_mode;\
652 if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
653 frag[s->fragment_start[2]].coding_method = coding_mode;
654
655 if (s->chroma_y_shift) {
656 frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
657 SET_CHROMA_MODES
658 } else if (s->chroma_x_shift) {
659 frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
660 for (k = 0; k < 2; k++) {
661 SET_CHROMA_MODES
662 frag += s->fragment_width[1];
663 }
664 } else {
665 for (k = 0; k < 4; k++) {
666 frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
667 SET_CHROMA_MODES
668 }
d86053a4 669 }
d86053a4 670 }
19cd517d 671 }
d86053a4
MM
672 }
673 }
892fc83e
MM
674
675 return 0;
44ae98dd
MM
676}
677
678/*
d86053a4
MM
679 * This function unpacks all the motion vectors for the individual
680 * macroblocks from the bitstream.
681 */
892fc83e 682static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4 683{
eb691ef2 684 int j, k, sb_x, sb_y;
d86053a4 685 int coding_mode;
1c183aa4
DC
686 int motion_x[4];
687 int motion_y[4];
d86053a4
MM
688 int last_motion_x = 0;
689 int last_motion_y = 0;
690 int prior_last_motion_x = 0;
691 int prior_last_motion_y = 0;
692 int current_macroblock;
693 int current_fragment;
14268254 694 int frag;
d86053a4 695
6599e2a7 696 if (s->keyframe)
6298f49f 697 return 0;
10f38380 698
1ae4518d
DC
699 /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
700 coding_mode = get_bits1(gb);
d86053a4 701
1ae4518d
DC
702 /* iterate through all of the macroblocks that contain 1 or more
703 * coded fragments */
19cd517d
DC
704 for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
705 for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
60867312
DC
706 if (get_bits_left(gb) <= 0)
707 return -1;
d86053a4 708
1ae4518d 709 for (j = 0; j < 4; j++) {
19cd517d
DC
710 int mb_x = 2*sb_x + (j>>1);
711 int mb_y = 2*sb_y + (((j>>1)+j)&1);
712 current_macroblock = mb_y * s->macroblock_width + mb_x;
713
714 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
1ae4518d
DC
715 (s->macroblock_coding[current_macroblock] == MODE_COPY))
716 continue;
d86053a4 717
1ae4518d
DC
718 switch (s->macroblock_coding[current_macroblock]) {
719
720 case MODE_INTER_PLUS_MV:
721 case MODE_GOLDEN_MV:
722 /* all 6 fragments use the same motion vector */
723 if (coding_mode == 0) {
724 motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
725 motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
726 } else {
727 motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
728 motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
892fc83e 729 }
7f9926a4 730
1ae4518d
DC
731 /* vector maintenance, only on MODE_INTER_PLUS_MV */
732 if (s->macroblock_coding[current_macroblock] ==
733 MODE_INTER_PLUS_MV) {
e32e2d56
AJ
734 prior_last_motion_x = last_motion_x;
735 prior_last_motion_y = last_motion_y;
1ae4518d
DC
736 last_motion_x = motion_x[0];
737 last_motion_y = motion_y[0];
738 }
739 break;
740
741 case MODE_INTER_FOURMV:
742 /* vector maintenance */
743 prior_last_motion_x = last_motion_x;
744 prior_last_motion_y = last_motion_y;
745
746 /* fetch 4 vectors from the bitstream, one for each
747 * Y fragment, then average for the C fragment vectors */
1ae4518d 748 for (k = 0; k < 4; k++) {
57783884 749 current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
eb691ef2 750 if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
1ae4518d
DC
751 if (coding_mode == 0) {
752 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
753 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
feaf1a73 754 } else {
1ae4518d
DC
755 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
756 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
feaf1a73 757 }
1ae4518d
DC
758 last_motion_x = motion_x[k];
759 last_motion_y = motion_y[k];
760 } else {
761 motion_x[k] = 0;
762 motion_y[k] = 0;
d86053a4 763 }
1ae4518d 764 }
1ae4518d
DC
765 break;
766
767 case MODE_INTER_LAST_MV:
768 /* all 6 fragments use the last motion vector */
769 motion_x[0] = last_motion_x;
770 motion_y[0] = last_motion_y;
d86053a4 771
1ae4518d
DC
772 /* no vector maintenance (last vector remains the
773 * last vector) */
774 break;
775
776 case MODE_INTER_PRIOR_LAST:
777 /* all 6 fragments use the motion vector prior to the
778 * last motion vector */
779 motion_x[0] = prior_last_motion_x;
780 motion_y[0] = prior_last_motion_y;
d86053a4 781
1ae4518d
DC
782 /* vector maintenance */
783 prior_last_motion_x = last_motion_x;
784 prior_last_motion_y = last_motion_y;
785 last_motion_x = motion_x[0];
786 last_motion_y = motion_y[0];
787 break;
44ae98dd 788
1ae4518d
DC
789 default:
790 /* covers intra, inter without MV, golden without MV */
e6e32bdc
MM
791 motion_x[0] = 0;
792 motion_y[0] = 0;
44ae98dd 793
1ae4518d
DC
794 /* no vector maintenance */
795 break;
796 }
d86053a4 797
1ae4518d 798 /* assign the motion vectors to the correct fragments */
ea676144 799 for (k = 0; k < 4; k++) {
1ae4518d 800 current_fragment =
57783884 801 BLOCK_Y*s->fragment_width[0] + BLOCK_X;
e6e32bdc 802 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
14268254
DC
803 s->motion_val[0][current_fragment][0] = motion_x[k];
804 s->motion_val[0][current_fragment][1] = motion_y[k];
e6e32bdc 805 } else {
14268254
DC
806 s->motion_val[0][current_fragment][0] = motion_x[0];
807 s->motion_val[0][current_fragment][1] = motion_y[0];
e6e32bdc 808 }
d86053a4 809 }
1e76a1da 810
1e76a1da 811 if (s->chroma_y_shift) {
1c183aa4
DC
812 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
813 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
814 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
815 }
1e76a1da
DC
816 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
817 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
14268254
DC
818 frag = mb_y*s->fragment_width[1] + mb_x;
819 s->motion_val[1][frag][0] = motion_x[0];
820 s->motion_val[1][frag][1] = motion_y[0];
1e76a1da
DC
821 } else if (s->chroma_x_shift) {
822 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
823 motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
824 motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
825 motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
826 motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
827 } else {
828 motion_x[1] = motion_x[0];
829 motion_y[1] = motion_y[0];
830 }
831 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
832 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
833
14268254 834 frag = 2*mb_y*s->fragment_width[1] + mb_x;
1e76a1da 835 for (k = 0; k < 2; k++) {
14268254
DC
836 s->motion_val[1][frag][0] = motion_x[k];
837 s->motion_val[1][frag][1] = motion_y[k];
1e76a1da
DC
838 frag += s->fragment_width[1];
839 }
840 } else {
841 for (k = 0; k < 4; k++) {
14268254 842 frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
1e76a1da 843 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
14268254
DC
844 s->motion_val[1][frag][0] = motion_x[k];
845 s->motion_val[1][frag][1] = motion_y[k];
1e76a1da 846 } else {
14268254
DC
847 s->motion_val[1][frag][0] = motion_x[0];
848 s->motion_val[1][frag][1] = motion_y[0];
1e76a1da
DC
849 }
850 }
ea676144 851 }
d86053a4 852 }
19cd517d 853 }
1ae4518d 854 }
892fc83e
MM
855
856 return 0;
d86053a4
MM
857}
858
f2264fa5
DC
859static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
860{
861 int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
c72625f2 862 int num_blocks = s->total_num_coded_frags;
f2264fa5
DC
863
864 for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
865 i = blocks_decoded = num_blocks_at_qpi = 0;
866
cc46005f
DC
867 bit = get_bits1(gb) ^ 1;
868 run_length = 0;
f2264fa5
DC
869
870 do {
cc46005f
DC
871 if (run_length == MAXIMUM_LONG_BIT_RUN)
872 bit = get_bits1(gb);
873 else
874 bit ^= 1;
875
f2264fa5
DC
876 run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
877 if (run_length == 34)
878 run_length += get_bits(gb, 12);
879 blocks_decoded += run_length;
880
881 if (!bit)
882 num_blocks_at_qpi += run_length;
883
884 for (j = 0; j < run_length; i++) {
c72625f2 885 if (i >= s->total_num_coded_frags)
f2264fa5
DC
886 return -1;
887
c72625f2
DC
888 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
889 s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
f2264fa5
DC
890 j++;
891 }
892 }
60867312 893 } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
f2264fa5
DC
894
895 num_blocks -= num_blocks_at_qpi;
896 }
897
898 return 0;
899}
900
115329f1 901/*
d86053a4
MM
902 * This function is called by unpack_dct_coeffs() to extract the VLCs from
903 * the bitstream. The VLCs encode tokens which are used to unpack DCT
904 * data. This function unpacks all the VLCs for either the Y plane or both
905 * C planes, and is called for DC coefficients or different AC coefficient
906 * levels (since different coefficient types require different VLC tables.
907 *
908 * This function returns a residual eob run. E.g, if a particular token gave
909 * instructions to EOB the next 5 fragments and there were only 2 fragments
910 * left in the current fragment range, 3 would be returned so that it could
911 * be passed into the next call to this same function.
912 */
913static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
914 VLC *table, int coeff_index,
c72625f2 915 int plane,
d86053a4
MM
916 int eob_run)
917{
c72625f2 918 int i, j = 0;
d86053a4 919 int token;
d3076955 920 int zero_run = 0;
88bd7fdc 921 int16_t coeff = 0;
d3076955 922 int bits_to_get;
c72625f2
DC
923 int blocks_ended;
924 int coeff_i = 0;
925 int num_coeffs = s->num_coded_frags[plane][coeff_index];
926 int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
d86053a4 927
ee3d7f58 928 /* local references to structure members to avoid repeated deferences */
c72625f2 929 int *coded_fragment_list = s->coded_fragment_list[plane];
ee3d7f58 930 Vp3Fragment *all_fragments = s->all_fragments;
ee3d7f58
MM
931 VLC_TYPE (*vlc_table)[2] = table->table;
932
c72625f2
DC
933 if (num_coeffs < 0)
934 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
935
936 if (eob_run > num_coeffs) {
937 coeff_i = blocks_ended = num_coeffs;
938 eob_run -= num_coeffs;
098523eb 939 } else {
c72625f2
DC
940 coeff_i = blocks_ended = eob_run;
941 eob_run = 0;
74c0ac12
MM
942 }
943
c72625f2
DC
944 // insert fake EOB token to cover the split between planes or zzi
945 if (blocks_ended)
946 dct_tokens[j++] = blocks_ended << 2;
d86053a4 947
f50dafa8 948 while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
d86053a4 949 /* decode a VLC into a token */
8e6daa4a 950 token = get_vlc2(gb, vlc_table, 11, 3);
d86053a4 951 /* use the token to get a zero run, a coefficient, and an eob run */
8370e426 952 if ((unsigned) token <= 6U) {
d3076955
MM
953 eob_run = eob_run_base[token];
954 if (eob_run_get_bits[token])
955 eob_run += get_bits(gb, eob_run_get_bits[token]);
c72625f2
DC
956
957 // record only the number of blocks ended in this plane,
958 // any spill will be recorded in the next plane.
959 if (eob_run > num_coeffs - coeff_i) {
960 dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
961 blocks_ended += num_coeffs - coeff_i;
962 eob_run -= num_coeffs - coeff_i;
963 coeff_i = num_coeffs;
964 } else {
965 dct_tokens[j++] = TOKEN_EOB(eob_run);
966 blocks_ended += eob_run;
967 coeff_i += eob_run;
968 eob_run = 0;
969 }
8370e426 970 } else if (token >= 0) {
d3076955 971 bits_to_get = coeff_get_bits[token];
428984b0
MM
972 if (bits_to_get)
973 bits_to_get = get_bits(gb, bits_to_get);
974 coeff = coeff_tables[token][bits_to_get];
d3076955
MM
975
976 zero_run = zero_run_base[token];
977 if (zero_run_get_bits[token])
978 zero_run += get_bits(gb, zero_run_get_bits[token]);
d86053a4 979
c72625f2
DC
980 if (zero_run) {
981 dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
982 } else {
983 // Save DC into the fragment structure. DC prediction is
984 // done in raster order, so the actual DC can't be in with
985 // other tokens. We still need the token in dct_tokens[]
986 // however, or else the structure collapses on itself.
987 if (!coeff_index)
988 all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
989
990 dct_tokens[j++] = TOKEN_COEFF(coeff);
991 }
992
993 if (coeff_index + zero_run > 64) {
01f9640b 994 av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
c72625f2
DC
995 " %d coeffs left\n", zero_run, 64-coeff_index);
996 zero_run = 64 - coeff_index;
997 }
098523eb 998
c72625f2
DC
999 // zero runs code multiple coefficients,
1000 // so don't try to decode coeffs for those higher levels
1001 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
1002 s->num_coded_frags[plane][i]--;
1003 coeff_i++;
8370e426
RB
1004 } else {
1005 av_log(s->avctx, AV_LOG_ERROR,
1006 "Invalid token %d\n", token);
1007 return -1;
c72625f2 1008 }
d86053a4
MM
1009 }
1010
c72625f2
DC
1011 if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1012 av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1013
1014 // decrement the number of blocks that have higher coeffecients for each
1015 // EOB run at this level
1016 if (blocks_ended)
1017 for (i = coeff_index+1; i < 64; i++)
1018 s->num_coded_frags[plane][i] -= blocks_ended;
1019
1020 // setup the next buffer
1021 if (plane < 2)
1022 s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
1023 else if (coeff_index < 63)
1024 s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
1025
d86053a4
MM
1026 return eob_run;
1027}
1028
138fe832
MM
1029static void reverse_dc_prediction(Vp3DecodeContext *s,
1030 int first_fragment,
1031 int fragment_width,
1032 int fragment_height);
d86053a4
MM
1033/*
1034 * This function unpacks all of the DCT coefficient data from the
1035 * bitstream.
1036 */
892fc83e 1037static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
d86053a4
MM
1038{
1039 int i;
1040 int dc_y_table;
1041 int dc_c_table;
1042 int ac_y_table;
1043 int ac_c_table;
1044 int residual_eob_run = 0;
9d8bb031
MM
1045 VLC *y_tables[64];
1046 VLC *c_tables[64];
d86053a4 1047
c72625f2
DC
1048 s->dct_tokens[0][0] = s->dct_tokens_base;
1049
f4433de9 1050 /* fetch the DC table indexes */
d86053a4
MM
1051 dc_y_table = get_bits(gb, 4);
1052 dc_c_table = get_bits(gb, 4);
1053
1054 /* unpack the Y plane DC coefficients */
115329f1 1055 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
c72625f2 1056 0, residual_eob_run);
8370e426
RB
1057 if (residual_eob_run < 0)
1058 return residual_eob_run;
d86053a4 1059
138fe832 1060 /* reverse prediction of the Y-plane DC coefficients */
57783884 1061 reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
138fe832 1062
d86053a4 1063 /* unpack the C plane DC coefficients */
d86053a4 1064 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
c72625f2 1065 1, residual_eob_run);
8370e426
RB
1066 if (residual_eob_run < 0)
1067 return residual_eob_run;
c72625f2
DC
1068 residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1069 2, residual_eob_run);
8370e426
RB
1070 if (residual_eob_run < 0)
1071 return residual_eob_run;
d86053a4 1072
138fe832
MM
1073 /* reverse prediction of the C-plane DC coefficients */
1074 if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1075 {
1076 reverse_dc_prediction(s, s->fragment_start[1],
57783884 1077 s->fragment_width[1], s->fragment_height[1]);
138fe832 1078 reverse_dc_prediction(s, s->fragment_start[2],
57783884 1079 s->fragment_width[1], s->fragment_height[1]);
138fe832
MM
1080 }
1081
f4433de9 1082 /* fetch the AC table indexes */
d86053a4
MM
1083 ac_y_table = get_bits(gb, 4);
1084 ac_c_table = get_bits(gb, 4);
1085
9d8bb031 1086 /* build tables of AC VLC tables */
d86053a4 1087 for (i = 1; i <= 5; i++) {
9d8bb031
MM
1088 y_tables[i] = &s->ac_vlc_1[ac_y_table];
1089 c_tables[i] = &s->ac_vlc_1[ac_c_table];
d86053a4 1090 }
d86053a4 1091 for (i = 6; i <= 14; i++) {
9d8bb031
MM
1092 y_tables[i] = &s->ac_vlc_2[ac_y_table];
1093 c_tables[i] = &s->ac_vlc_2[ac_c_table];
d86053a4 1094 }
d86053a4 1095 for (i = 15; i <= 27; i++) {
9d8bb031
MM
1096 y_tables[i] = &s->ac_vlc_3[ac_y_table];
1097 c_tables[i] = &s->ac_vlc_3[ac_c_table];
d86053a4 1098 }
d86053a4 1099 for (i = 28; i <= 63; i++) {
9d8bb031
MM
1100 y_tables[i] = &s->ac_vlc_4[ac_y_table];
1101 c_tables[i] = &s->ac_vlc_4[ac_c_table];
1102 }
1103
1104 /* decode all AC coefficents */
1105 for (i = 1; i <= 63; i++) {
9d8bb031 1106 residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
c72625f2 1107 0, residual_eob_run);
8370e426
RB
1108 if (residual_eob_run < 0)
1109 return residual_eob_run;
d86053a4 1110
9d8bb031 1111 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
c72625f2 1112 1, residual_eob_run);
8370e426
RB
1113 if (residual_eob_run < 0)
1114 return residual_eob_run;
c72625f2
DC
1115 residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1116 2, residual_eob_run);
8370e426
RB
1117 if (residual_eob_run < 0)
1118 return residual_eob_run;
d86053a4 1119 }
892fc83e
MM
1120
1121 return 0;
d86053a4
MM
1122}
1123
1124/*
1125 * This function reverses the DC prediction for each coded fragment in
115329f1 1126 * the frame. Much of this function is adapted directly from the original
d86053a4
MM
1127 * VP3 source code.
1128 */
1129#define COMPATIBLE_FRAME(x) \
1130 (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
c72625f2 1131#define DC_COEFF(u) s->all_fragments[u].dc
d86053a4
MM
1132
1133static void reverse_dc_prediction(Vp3DecodeContext *s,
1134 int first_fragment,
1135 int fragment_width,
115329f1 1136 int fragment_height)
d86053a4
MM
1137{
1138
1139#define PUL 8
1140#define PU 4
1141#define PUR 2
1142#define PL 1
1143
1144 int x, y;
1145 int i = first_fragment;
1146
59ef342b 1147 int predicted_dc;
d86053a4 1148
d86053a4
MM
1149 /* DC values for the left, up-left, up, and up-right fragments */
1150 int vl, vul, vu, vur;
1151
f4433de9 1152 /* indexes for the left, up-left, up, and up-right fragments */
d86053a4
MM
1153 int l, ul, u, ur;
1154
115329f1 1155 /*
d86053a4
MM
1156 * The 6 fields mean:
1157 * 0: up-left multiplier
1158 * 1: up multiplier
1159 * 2: up-right multiplier
1160 * 3: left multiplier
d86053a4 1161 */
bb991087 1162 static const int predictor_transform[16][4] = {
006ff1ca
MN
1163 { 0, 0, 0, 0},
1164 { 0, 0, 0,128}, // PL
1165 { 0, 0,128, 0}, // PUR
1166 { 0, 0, 53, 75}, // PUR|PL
1167 { 0,128, 0, 0}, // PU
1168 { 0, 64, 0, 64}, // PU|PL
1169 { 0,128, 0, 0}, // PU|PUR
1170 { 0, 0, 53, 75}, // PU|PUR|PL
1171 {128, 0, 0, 0}, // PUL
1172 { 0, 0, 0,128}, // PUL|PL
1173 { 64, 0, 64, 0}, // PUL|PUR
1174 { 0, 0, 53, 75}, // PUL|PUR|PL
1175 { 0,128, 0, 0}, // PUL|PU
1176 {-104,116, 0,116}, // PUL|PU|PL
1177 { 24, 80, 24, 0}, // PUL|PU|PUR
1178 {-104,116, 0,116} // PUL|PU|PUR|PL
d86053a4
MM
1179 };
1180
1181 /* This table shows which types of blocks can use other blocks for
1182 * prediction. For example, INTRA is the only mode in this table to
1183 * have a frame number of 0. That means INTRA blocks can only predict
115329f1 1184 * from other INTRA blocks. There are 2 golden frame coding types;
d86053a4
MM
1185 * blocks encoding in these modes can only predict from other blocks
1186 * that were encoded with these 1 of these 2 modes. */
50ba3fd7 1187 static const unsigned char compatible_frame[9] = {
d86053a4
MM
1188 1, /* MODE_INTER_NO_MV */
1189 0, /* MODE_INTRA */
1190 1, /* MODE_INTER_PLUS_MV */
1191 1, /* MODE_INTER_LAST_MV */
1192 1, /* MODE_INTER_PRIOR_MV */
1193 2, /* MODE_USING_GOLDEN */
1194 2, /* MODE_GOLDEN_MV */
50ba3fd7
JGG
1195 1, /* MODE_INTER_FOUR_MV */
1196 3 /* MODE_COPY */
d86053a4
MM
1197 };
1198 int current_frame_type;
1199
1200 /* there is a last DC predictor for each of the 3 frame types */
1201 short last_dc[3];
1202
1203 int transform = 0;
1204
d86053a4
MM
1205 vul = vu = vur = vl = 0;
1206 last_dc[0] = last_dc[1] = last_dc[2] = 0;
1207
1208 /* for each fragment row... */
1209 for (y = 0; y < fragment_height; y++) {
1210
1211 /* for each fragment in a row... */
1212 for (x = 0; x < fragment_width; x++, i++) {
1213
1214 /* reverse prediction if this block was coded */
1215 if (s->all_fragments[i].coding_method != MODE_COPY) {
1216
115329f1 1217 current_frame_type =
d86053a4 1218 compatible_frame[s->all_fragments[i].coding_method];
d86053a4 1219
f72f8a77
MN
1220 transform= 0;
1221 if(x){
1222 l= i-1;
7beddb12 1223 vl = DC_COEFF(l);
50ba3fd7 1224 if(COMPATIBLE_FRAME(l))
006ff1ca 1225 transform |= PL;
f72f8a77
MN
1226 }
1227 if(y){
1228 u= i-fragment_width;
7beddb12 1229 vu = DC_COEFF(u);
50ba3fd7 1230 if(COMPATIBLE_FRAME(u))
006ff1ca 1231 transform |= PU;
f72f8a77
MN
1232 if(x){
1233 ul= i-fragment_width-1;
1234 vul = DC_COEFF(ul);
50ba3fd7 1235 if(COMPATIBLE_FRAME(ul))
006ff1ca 1236 transform |= PUL;
f72f8a77
MN
1237 }
1238 if(x + 1 < fragment_width){
1239 ur= i-fragment_width+1;
1240 vur = DC_COEFF(ur);
50ba3fd7 1241 if(COMPATIBLE_FRAME(ur))
006ff1ca 1242 transform |= PUR;
f72f8a77 1243 }
d86053a4
MM
1244 }
1245
d86053a4
MM
1246 if (transform == 0) {
1247
1248 /* if there were no fragments to predict from, use last
1249 * DC saved */
7beddb12 1250 predicted_dc = last_dc[current_frame_type];
d86053a4
MM
1251 } else {
1252
1253 /* apply the appropriate predictor transform */
1254 predicted_dc =
1255 (predictor_transform[transform][0] * vul) +
1256 (predictor_transform[transform][1] * vu) +
1257 (predictor_transform[transform][2] * vur) +
1258 (predictor_transform[transform][3] * vl);
1259
684d9e36 1260 predicted_dc /= 128;
d86053a4
MM
1261
1262 /* check for outranging on the [ul u l] and
1263 * [ul u ur l] predictors */
c11cb375 1264 if ((transform == 15) || (transform == 13)) {
c26abfa5 1265 if (FFABS(predicted_dc - vu) > 128)
d86053a4 1266 predicted_dc = vu;
c26abfa5 1267 else if (FFABS(predicted_dc - vl) > 128)
d86053a4 1268 predicted_dc = vl;
c26abfa5 1269 else if (FFABS(predicted_dc - vul) > 128)
d86053a4
MM
1270 predicted_dc = vul;
1271 }
d86053a4
MM
1272 }
1273
7beddb12 1274 /* at long last, apply the predictor */
c72625f2 1275 DC_COEFF(i) += predicted_dc;
d86053a4 1276 /* save the DC */
7beddb12 1277 last_dc[current_frame_type] = DC_COEFF(i);
d86053a4
MM
1278 }
1279 }
1280 }
1281}
1282
256c0662 1283static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
fe313556 1284{
fe313556
DC
1285 int x, y;
1286 int *bounding_values= s->bounding_values_array+127;
1287
57783884
DC
1288 int width = s->fragment_width[!!plane];
1289 int height = s->fragment_height[!!plane];
621f9a40 1290 int fragment = s->fragment_start [plane] + ystart * width;
93f30547 1291 ptrdiff_t stride = s->current_frame.f->linesize[plane];
759001c5 1292 uint8_t *plane_data = s->current_frame.f->data [plane];
621f9a40 1293 if (!s->flipped_image) stride = -stride;
735acf56 1294 plane_data += s->data_offset[plane] + 8*ystart*stride;
621f9a40
DC
1295
1296 for (y = ystart; y < yend; y++) {
1297
1298 for (x = 0; x < width; x++) {
1299 /* This code basically just deblocks on the edges of coded blocks.
1300 * However, it has to be much more complicated because of the
1301 * braindamaged deblock ordering used in VP3/Theora. Order matters
1302 * because some pixels get filtered twice. */
1303 if( s->all_fragments[fragment].coding_method != MODE_COPY )
1304 {
1305 /* do not perform left edge filter for left columns frags */
1306 if (x > 0) {
28f9ab70 1307 s->vp3dsp.h_loop_filter(
735acf56 1308 plane_data + 8*x,
621f9a40
DC
1309 stride, bounding_values);
1310 }
fe313556 1311
621f9a40
DC
1312 /* do not perform top edge filter for top row fragments */
1313 if (y > 0) {
28f9ab70 1314 s->vp3dsp.v_loop_filter(
735acf56 1315 plane_data + 8*x,
621f9a40
DC
1316 stride, bounding_values);
1317 }
fe313556 1318
621f9a40
DC
1319 /* do not perform right edge filter for right column
1320 * fragments or if right fragment neighbor is also coded
1321 * in this frame (it will be filtered in next iteration) */
1322 if ((x < width - 1) &&
1323 (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
28f9ab70 1324 s->vp3dsp.h_loop_filter(
735acf56 1325 plane_data + 8*x + 8,
621f9a40 1326 stride, bounding_values);
fe313556
DC
1327 }
1328
621f9a40
DC
1329 /* do not perform bottom edge filter for bottom row
1330 * fragments or if bottom fragment neighbor is also coded
1331 * in this frame (it will be filtered in the next row) */
1332 if ((y < height - 1) &&
1333 (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
28f9ab70 1334 s->vp3dsp.v_loop_filter(
735acf56 1335 plane_data + 8*x + 8*stride,
621f9a40
DC
1336 stride, bounding_values);
1337 }
fe313556 1338 }
621f9a40
DC
1339
1340 fragment++;
fe313556 1341 }
735acf56 1342 plane_data += 8*stride;
621f9a40 1343 }
fe313556
DC
1344}
1345
a8de3901 1346/**
49bd8e4b 1347 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
c72625f2
DC
1348 * for the next block in coding order
1349 */
1350static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
88bd7fdc 1351 int plane, int inter, int16_t block[64])
c72625f2
DC
1352{
1353 int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
01582122 1354 uint8_t *perm = s->idct_scantable;
c72625f2
DC
1355 int i = 0;
1356
1357 do {
1358 int token = *s->dct_tokens[plane][i];
1359 switch (token & 3) {
1360 case 0: // EOB
1361 if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1362 s->dct_tokens[plane][i]++;
1363 else
1364 *s->dct_tokens[plane][i] = token & ~3;
1365 goto end;
1366 case 1: // zero run
1367 s->dct_tokens[plane][i]++;
1368 i += (token >> 2) & 0x7f;
8b94df0f
RT
1369 if (i > 63) {
1370 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
1371 return i;
1372 }
c72625f2
DC
1373 block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1374 i++;
1375 break;
1376 case 2: // coeff
1377 block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1378 s->dct_tokens[plane][i++]++;
1379 break;
d7097c2d 1380 default: // shouldn't happen
c72625f2
DC
1381 return i;
1382 }
1383 } while (i < 64);
9b4767e4
JG
1384 // return value is expected to be a valid level
1385 i--;
c72625f2
DC
1386end:
1387 // the actual DC+prediction is in the fragment structure
1388 block[0] = frag->dc * s->qmat[0][inter][plane][0];
1389 return i;
1390}
1391
1392/**
a8de3901
DC
1393 * called when all pixels up to row y are complete
1394 */
1395static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1396{
560f773c
JR
1397 int h, cy, i;
1398 int offset[AV_NUM_DATA_POINTERS];
a8de3901 1399
27237d52 1400 if (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
d23845f3
AS
1401 int y_flipped = s->flipped_image ? s->avctx->height-y : y;
1402
1403 // At the end of the frame, report INT_MAX instead of the height of the frame.
1404 // This makes the other threads' ff_thread_await_progress() calls cheaper, because
1405 // they don't have to clip their values.
1406 ff_thread_report_progress(&s->current_frame, y_flipped==s->avctx->height ? INT_MAX : y_flipped-1, 0);
1407 }
1408
a8de3901
DC
1409 if(s->avctx->draw_horiz_band==NULL)
1410 return;
1411
1412 h= y - s->last_slice_end;
3b9ee20f 1413 s->last_slice_end= y;
a8de3901
DC
1414 y -= h;
1415
1416 if (!s->flipped_image) {
83f72f13 1417 y = s->avctx->height - y - h;
a8de3901
DC
1418 }
1419
8764389d 1420 cy = y >> s->chroma_y_shift;
759001c5
AK
1421 offset[0] = s->current_frame.f->linesize[0]*y;
1422 offset[1] = s->current_frame.f->linesize[1]*cy;
1423 offset[2] = s->current_frame.f->linesize[2]*cy;
560f773c
JR
1424 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
1425 offset[i] = 0;
a8de3901
DC
1426
1427 emms_c();
759001c5 1428 s->avctx->draw_horiz_band(s->avctx, s->current_frame.f, offset, y, 3, h);
a8de3901
DC
1429}
1430
d23845f3
AS
1431/**
1432 * Wait for the reference frame of the current fragment.
1433 * The progress value is in luma pixel rows.
1434 */
1435static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment, int motion_y, int y)
1436{
759001c5 1437 ThreadFrame *ref_frame;
d23845f3
AS
1438 int ref_row;
1439 int border = motion_y&1;
1440
1441 if (fragment->coding_method == MODE_USING_GOLDEN ||
1442 fragment->coding_method == MODE_GOLDEN_MV)
1443 ref_frame = &s->golden_frame;
1444 else
1445 ref_frame = &s->last_frame;
1446
1447 ref_row = y + (motion_y>>1);
1448 ref_row = FFMAX(FFABS(ref_row), ref_row + 8 + border);
1449
1450 ff_thread_await_progress(ref_frame, ref_row, 0);
1451}
1452
d86053a4 1453/*
dc4b78d9 1454 * Perform the final rendering for a particular slice of data.
7a095ea6 1455 * The slice number ranges from 0..(c_superblock_height - 1).
dc4b78d9
MM
1456 */
1457static void render_slice(Vp3DecodeContext *s, int slice)
1458{
e8dcd730 1459 int x, y, i, j, fragment;
88bd7fdc 1460 int16_t *block = s->block;
dc4b78d9 1461 int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
dc4b78d9
MM
1462 int motion_halfpel_index;
1463 uint8_t *motion_source;
7a095ea6 1464 int plane, first_pixel;
dc4b78d9 1465
7a095ea6 1466 if (slice >= s->c_superblock_height)
dc4b78d9
MM
1467 return;
1468
1469 for (plane = 0; plane < 3; plane++) {
759001c5
AK
1470 uint8_t *output_plane = s->current_frame.f->data [plane] + s->data_offset[plane];
1471 uint8_t * last_plane = s-> last_frame.f->data [plane] + s->data_offset[plane];
1472 uint8_t *golden_plane = s-> golden_frame.f->data [plane] + s->data_offset[plane];
93f30547 1473 ptrdiff_t stride = s->current_frame.f->linesize[plane];
1e76a1da
DC
1474 int plane_width = s->width >> (plane && s->chroma_x_shift);
1475 int plane_height = s->height >> (plane && s->chroma_y_shift);
14268254 1476 int8_t (*motion_val)[2] = s->motion_val[!!plane];
7a095ea6 1477
1e76a1da
DC
1478 int sb_x, sb_y = slice << (!plane && s->chroma_y_shift);
1479 int slice_height = sb_y + 1 + (!plane && s->chroma_y_shift);
7a095ea6
DC
1480 int slice_width = plane ? s->c_superblock_width : s->y_superblock_width;
1481
57783884
DC
1482 int fragment_width = s->fragment_width[!!plane];
1483 int fragment_height = s->fragment_height[!!plane];
7a095ea6 1484 int fragment_start = s->fragment_start[plane];
27237d52 1485 int do_await = !plane && HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_FRAME);
1abbf64e
MN
1486
1487 if (!s->flipped_image) stride = -stride;
161e8cf4
DC
1488 if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1489 continue;
dc4b78d9 1490
7a095ea6
DC
1491 /* for each superblock row in the slice (both of them)... */
1492 for (; sb_y < slice_height; sb_y++) {
dc4b78d9 1493
7a095ea6
DC
1494 /* for each superblock in a row... */
1495 for (sb_x = 0; sb_x < slice_width; sb_x++) {
dc4b78d9 1496
7a095ea6
DC
1497 /* for each block in a superblock... */
1498 for (j = 0; j < 16; j++) {
1499 x = 4*sb_x + hilbert_offset[j][0];
1500 y = 4*sb_y + hilbert_offset[j][1];
e8dcd730 1501 fragment = y*fragment_width + x;
7a095ea6 1502
e8dcd730 1503 i = fragment_start + fragment;
7a095ea6
DC
1504
1505 // bounds check
1506 if (x >= fragment_width || y >= fragment_height)
1507 continue;
1508
1509 first_pixel = 8*y*stride + 8*x;
dc4b78d9 1510
d23845f3
AS
1511 if (do_await && s->all_fragments[i].coding_method != MODE_INTRA)
1512 await_reference_row(s, &s->all_fragments[i], motion_val[fragment][1], (16*y) >> s->chroma_y_shift);
1513
dc4b78d9 1514 /* transform if this block was coded */
161e8cf4 1515 if (s->all_fragments[i].coding_method != MODE_COPY) {
dc4b78d9
MM
1516 if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1517 (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1518 motion_source= golden_plane;
115329f1 1519 else
dc4b78d9
MM
1520 motion_source= last_plane;
1521
735acf56 1522 motion_source += first_pixel;
dc4b78d9
MM
1523 motion_halfpel_index = 0;
1524
1525 /* sort out the motion vector if this fragment is coded
1526 * using a motion vector method */
1527 if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1528 (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1529 int src_x, src_y;
e8dcd730
AS
1530 motion_x = motion_val[fragment][0];
1531 motion_y = motion_val[fragment][1];
dc4b78d9 1532
7a095ea6
DC
1533 src_x= (motion_x>>1) + 8*x;
1534 src_y= (motion_y>>1) + 8*y;
dc4b78d9
MM
1535
1536 motion_halfpel_index = motion_x & 0x01;
1537 motion_source += (motion_x >> 1);
1538
1539 motion_halfpel_index |= (motion_y & 0x01) << 1;
1540 motion_source += ((motion_y >> 1) * stride);
1541
1542 if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1543 uint8_t *temp= s->edge_emu_buffer;
a89f4ca0 1544 if(stride<0) temp -= 8*stride;
dc4b78d9 1545
458446ac
RB
1546 s->vdsp.emulated_edge_mc(temp, motion_source,
1547 stride, stride,
1548 9, 9, src_x, src_y,
1549 plane_width,
1550 plane_height);
dc4b78d9
MM
1551 motion_source= temp;
1552 }
1553 }
115329f1 1554
dc4b78d9
MM
1555
1556 /* first, take care of copying a block from either the
1557 * previous or the golden frame */
1558 if (s->all_fragments[i].coding_method != MODE_INTRA) {
115329f1
DB
1559 /* Note, it is possible to implement all MC cases with
1560 put_no_rnd_pixels_l2 which would look more like the
1561 VP3 source but this would be slower as
dc4b78d9
MM
1562 put_no_rnd_pixels_tab is better optimzed */
1563 if(motion_halfpel_index != 3){
3bd062bf 1564 s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
735acf56 1565 output_plane + first_pixel,
dc4b78d9
MM
1566 motion_source, stride, 8);
1567 }else{
1568 int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
4a73fbd9 1569 s->vp3dsp.put_no_rnd_pixels_l2(
735acf56 1570 output_plane + first_pixel,
115329f1
DB
1571 motion_source - d,
1572 motion_source + stride + 1 + d,
dc4b78d9
MM
1573 stride, 8);
1574 }
dc4b78d9
MM
1575 }
1576
dc4b78d9 1577 /* invert DCT and place (or add) in final output */
115329f1 1578
dc4b78d9 1579 if (s->all_fragments[i].coding_method == MODE_INTRA) {
8b94df0f
RT
1580 int index;
1581 index = vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1582 if (index > 63)
1583 continue;
28f9ab70 1584 s->vp3dsp.idct_put(
735acf56 1585 output_plane + first_pixel,
dc4b78d9
MM
1586 stride,
1587 block);
1588 } else {
8b94df0f
RT
1589 int index = vp3_dequant(s, s->all_fragments + i, plane, 1, block);
1590 if (index > 63)
1591 continue;
1592 if (index > 0) {
28f9ab70 1593 s->vp3dsp.idct_add(
735acf56 1594 output_plane + first_pixel,
dc4b78d9
MM
1595 stride,
1596 block);
eb6a6cd7 1597 } else {
28f9ab70 1598 s->vp3dsp.idct_dc_add(output_plane + first_pixel, stride, block);
eb6a6cd7 1599 }
dc4b78d9 1600 }
dc4b78d9
MM
1601 } else {
1602
1603 /* copy directly from the previous frame */
3bd062bf 1604 s->hdsp.put_pixels_tab[1][0](
735acf56
DC
1605 output_plane + first_pixel,
1606 last_plane + first_pixel,
dc4b78d9
MM
1607 stride, 8);
1608
1609 }
7a095ea6 1610 }
dc4b78d9 1611 }
7a095ea6
DC
1612
1613 // Filter up to the last row in the superblock row
a4501a45
DC
1614 if (!s->skip_loop_filter)
1615 apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
dc4b78d9
MM
1616 }
1617 }
1618
dc4b78d9
MM
1619 /* this looks like a good place for slice dispatch... */
1620 /* algorithm:
dc4b78d9 1621 * if (slice == s->macroblock_height - 1)
f44b08a5
MM
1622 * dispatch (both last slice & 2nd-to-last slice);
1623 * else if (slice > 0)
1624 * dispatch (slice - 1);
dc4b78d9
MM
1625 */
1626
3d487db1 1627 vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
dc4b78d9
MM
1628}
1629
edbb0c07
AS
1630/// Allocate tables for per-frame data in Vp3DecodeContext
1631static av_cold int allocate_tables(AVCodecContext *avctx)
1632{
1633 Vp3DecodeContext *s = avctx->priv_data;
1634 int y_fragment_count, c_fragment_count;
1635
1636 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1637 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1638
1639 s->superblock_coding = av_malloc(s->superblock_count);
1640 s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1641 s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1642 s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1643 s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1644 s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1645
1646 /* work out the block mapping tables */
1647 s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1648 s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1649
1650 if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1651 !s->coded_fragment_list[0] || !s->superblock_fragments || !s->macroblock_coding ||
1652 !s->motion_val[0] || !s->motion_val[1]) {
1653 vp3_decode_end(avctx);
1654 return -1;
1655 }
1656
1657 init_block_mapping(s);
1658
1659 return 0;
1660}
1661
759001c5
AK
1662static av_cold int init_frames(Vp3DecodeContext *s)
1663{
1664 s->current_frame.f = av_frame_alloc();
1665 s->last_frame.f = av_frame_alloc();
1666 s->golden_frame.f = av_frame_alloc();
1667
1668 if (!s->current_frame.f || !s->last_frame.f || !s->golden_frame.f) {
1669 av_frame_free(&s->current_frame.f);
1670 av_frame_free(&s->last_frame.f);
1671 av_frame_free(&s->golden_frame.f);
1672 return AVERROR(ENOMEM);
1673 }
1674
1675 return 0;
1676}
1677
98a6fff9 1678static av_cold int vp3_decode_init(AVCodecContext *avctx)
d86053a4
MM
1679{
1680 Vp3DecodeContext *s = avctx->priv_data;
759001c5 1681 int i, inter, plane, ret;
892fc83e
MM
1682 int c_width;
1683 int c_height;
1e76a1da 1684 int y_fragment_count, c_fragment_count;
d86053a4 1685
759001c5
AK
1686 ret = init_frames(s);
1687 if (ret < 0)
1688 return ret;
1689
1690 avctx->internal->allocate_progress = 1;
1691
3c3f113e 1692 if (avctx->codec_tag == MKTAG('V','P','3','0'))
bb270c08 1693 s->version = 0;
3c3f113e 1694 else
bb270c08 1695 s->version = 1;
3c3f113e 1696
d86053a4 1697 s->avctx = avctx;
ef516f73
DC
1698 s->width = FFALIGN(avctx->width, 16);
1699 s->height = FFALIGN(avctx->height, 16);
716d413c
AK
1700 if (avctx->pix_fmt == AV_PIX_FMT_NONE)
1701 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
580a7465 1702 avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
3bd062bf 1703 ff_hpeldsp_init(&s->hdsp, avctx->flags | CODEC_FLAG_BITEXACT);
8c53d39e 1704 ff_videodsp_init(&s->vdsp, 8);
28f9ab70 1705 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
115329f1 1706
01582122 1707 for (i = 0; i < 64; i++) {
f2408ec9
DB
1708#define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
1709 s->idct_permutation[i] = TRANSPOSE(i);
1710 s->idct_scantable[i] = TRANSPOSE(ff_zigzag_direct[i]);
1711#undef TRANSPOSE
01582122 1712 }
d86053a4
MM
1713
1714 /* initialize to an impossible value which will force a recalculation
1715 * in the first frame decode */
f2264fa5
DC
1716 for (i = 0; i < 3; i++)
1717 s->qps[i] = -1;
d86053a4 1718
c1a02e88
LB
1719 av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift,
1720 &s->chroma_y_shift);
1e76a1da 1721
892fc83e
MM
1722 s->y_superblock_width = (s->width + 31) / 32;
1723 s->y_superblock_height = (s->height + 31) / 32;
35c28d23 1724 s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
892fc83e
MM
1725
1726 /* work out the dimensions for the C planes */
1e76a1da
DC
1727 c_width = s->width >> s->chroma_x_shift;
1728 c_height = s->height >> s->chroma_y_shift;
892fc83e
MM
1729 s->c_superblock_width = (c_width + 31) / 32;
1730 s->c_superblock_height = (c_height + 31) / 32;
35c28d23 1731 s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
892fc83e 1732
35c28d23
DC
1733 s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1734 s->u_superblock_start = s->y_superblock_count;
1735 s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
d86053a4
MM
1736
1737 s->macroblock_width = (s->width + 15) / 16;
1738 s->macroblock_height = (s->height + 15) / 16;
1739 s->macroblock_count = s->macroblock_width * s->macroblock_height;
1740
57783884
DC
1741 s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1742 s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1e76a1da
DC
1743 s->fragment_width[1] = s->fragment_width[0] >> s->chroma_x_shift;
1744 s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
d86053a4
MM
1745
1746 /* fragment count covers all 8x8 blocks for all 3 planes */
1e76a1da
DC
1747 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1748 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1749 s->fragment_count = y_fragment_count + 2*c_fragment_count;
1750 s->fragment_start[1] = y_fragment_count;
1751 s->fragment_start[2] = y_fragment_count + c_fragment_count;
d86053a4 1752
f44ee2c3
AB
1753 if (!s->theora_tables)
1754 {
2287c100 1755 for (i = 0; i < 64; i++) {
bb270c08 1756 s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
bb270c08 1757 s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
ae1dd8e1 1758 s->base_matrix[0][i] = vp31_intra_y_dequant[i];
ae1dd8e1 1759 s->base_matrix[1][i] = vp31_intra_c_dequant[i];
ae1dd8e1 1760 s->base_matrix[2][i] = vp31_inter_dequant[i];
bb270c08 1761 s->filter_limit_values[i] = vp31_filter_limit_values[i];
2287c100 1762 }
f44ee2c3 1763
ae1dd8e1
MN
1764 for(inter=0; inter<2; inter++){
1765 for(plane=0; plane<3; plane++){
1766 s->qr_count[inter][plane]= 1;
1767 s->qr_size [inter][plane][0]= 63;
1768 s->qr_base [inter][plane][0]=
1769 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1770 }
1771 }
1772
39922395
MM
1773 /* init VLC tables */
1774 for (i = 0; i < 16; i++) {
1775
1776 /* DC histograms */
8e6daa4a 1777 init_vlc(&s->dc_vlc[i], 11, 32,
39922395
MM
1778 &dc_bias[i][0][1], 4, 2,
1779 &dc_bias[i][0][0], 4, 2, 0);
1780
1781 /* group 1 AC histograms */
8e6daa4a 1782 init_vlc(&s->ac_vlc_1[i], 11, 32,
39922395
MM
1783 &ac_bias_0[i][0][1], 4, 2,
1784 &ac_bias_0[i][0][0], 4, 2, 0);
1785
1786 /* group 2 AC histograms */
8e6daa4a 1787 init_vlc(&s->ac_vlc_2[i], 11, 32,
39922395
MM
1788 &ac_bias_1[i][0][1], 4, 2,
1789 &ac_bias_1[i][0][0], 4, 2, 0);
1790
1791 /* group 3 AC histograms */
8e6daa4a 1792 init_vlc(&s->ac_vlc_3[i], 11, 32,
39922395
MM
1793 &ac_bias_2[i][0][1], 4, 2,
1794 &ac_bias_2[i][0][0], 4, 2, 0);
1795
1796 /* group 4 AC histograms */
8e6daa4a 1797 init_vlc(&s->ac_vlc_4[i], 11, 32,
39922395
MM
1798 &ac_bias_3[i][0][1], 4, 2,
1799 &ac_bias_3[i][0][0], 4, 2, 0);
1800 }
1801 } else {
39922395 1802
8e6daa4a 1803 for (i = 0; i < 16; i++) {
39922395 1804 /* DC histograms */
8e6daa4a
DC
1805 if (init_vlc(&s->dc_vlc[i], 11, 32,
1806 &s->huffman_table[i][0][1], 8, 4,
1807 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
c4b7b8bf 1808 goto vlc_fail;
39922395
MM
1809
1810 /* group 1 AC histograms */
8e6daa4a
DC
1811 if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1812 &s->huffman_table[i+16][0][1], 8, 4,
1813 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
c4b7b8bf 1814 goto vlc_fail;
39922395
MM
1815
1816 /* group 2 AC histograms */
8e6daa4a
DC
1817 if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1818 &s->huffman_table[i+16*2][0][1], 8, 4,
1819 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
c4b7b8bf 1820 goto vlc_fail;
39922395
MM
1821
1822 /* group 3 AC histograms */
8e6daa4a
DC
1823 if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1824 &s->huffman_table[i+16*3][0][1], 8, 4,
1825 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
c4b7b8bf 1826 goto vlc_fail;
39922395
MM
1827
1828 /* group 4 AC histograms */
8e6daa4a
DC
1829 if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1830 &s->huffman_table[i+16*4][0][1], 8, 4,
1831 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
c4b7b8bf 1832 goto vlc_fail;
39922395 1833 }
d86053a4
MM
1834 }
1835
d8278bab
MM
1836 init_vlc(&s->superblock_run_length_vlc, 6, 34,
1837 &superblock_run_length_vlc_table[0][1], 4, 2,
1838 &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1839
dd36b667 1840 init_vlc(&s->fragment_run_length_vlc, 5, 30,
0ad72bdd
MM
1841 &fragment_run_length_vlc_table[0][1], 4, 2,
1842 &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1843
1844 init_vlc(&s->mode_code_vlc, 3, 8,
1845 &mode_code_vlc_table[0][1], 2, 1,
1846 &mode_code_vlc_table[0][0], 2, 1, 0);
1847
1848 init_vlc(&s->motion_vector_vlc, 6, 63,
1849 &motion_vector_vlc_table[0][1], 2, 1,
1850 &motion_vector_vlc_table[0][0], 2, 1, 0);
1851
edbb0c07 1852 return allocate_tables(avctx);
c4b7b8bf
RD
1853
1854vlc_fail:
1855 av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1856 return -1;
d86053a4
MM
1857}
1858
d23845f3 1859/// Release and shuffle frames after decode finishes
759001c5 1860static int update_frames(AVCodecContext *avctx)
d23845f3
AS
1861{
1862 Vp3DecodeContext *s = avctx->priv_data;
759001c5 1863 int ret = 0;
d23845f3 1864
d23845f3
AS
1865
1866 /* shuffle frames (last = current) */
759001c5
AK
1867 ff_thread_release_buffer(avctx, &s->last_frame);
1868 ret = ff_thread_ref_frame(&s->last_frame, &s->current_frame);
1869 if (ret < 0)
1870 goto fail;
d23845f3
AS
1871
1872 if (s->keyframe) {
759001c5
AK
1873 ff_thread_release_buffer(avctx, &s->golden_frame);
1874 ret = ff_thread_ref_frame(&s->golden_frame, &s->current_frame);
d23845f3
AS
1875 }
1876
759001c5
AK
1877fail:
1878 ff_thread_release_buffer(avctx, &s->current_frame);
1879 return ret;
1880}
1881
1882static int ref_frame(Vp3DecodeContext *s, ThreadFrame *dst, ThreadFrame *src)
1883{
1884 ff_thread_release_buffer(s->avctx, dst);
1885 if (src->f->data[0])
1886 return ff_thread_ref_frame(dst, src);
1887 return 0;
1888}
1889
1890static int ref_frames(Vp3DecodeContext *dst, Vp3DecodeContext *src)
1891{
1892 int ret;
1893 if ((ret = ref_frame(dst, &dst->current_frame, &src->current_frame)) < 0 ||
1894 (ret = ref_frame(dst, &dst->golden_frame, &src->golden_frame)) < 0 ||
1895 (ret = ref_frame(dst, &dst->last_frame, &src->last_frame)) < 0)
1896 return ret;
1897 return 0;
d23845f3
AS
1898}
1899
1900static int vp3_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1901{
1902 Vp3DecodeContext *s = dst->priv_data, *s1 = src->priv_data;
1903 int qps_changed = 0, i, err;
1904
8370e426
RB
1905#define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1906
759001c5 1907 if (!s1->current_frame.f->data[0]
d23845f3 1908 ||s->width != s1->width
8370e426
RB
1909 ||s->height!= s1->height) {
1910 if (s != s1)
759001c5 1911 ref_frames(s, s1);
d23845f3 1912 return -1;
8370e426 1913 }
d23845f3
AS
1914
1915 if (s != s1) {
1916 // init tables if the first frame hasn't been decoded
759001c5 1917 if (!s->current_frame.f->data[0]) {
d23845f3
AS
1918 int y_fragment_count, c_fragment_count;
1919 s->avctx = dst;
1920 err = allocate_tables(dst);
1921 if (err)
1922 return err;
1923 y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
1924 c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
1925 memcpy(s->motion_val[0], s1->motion_val[0], y_fragment_count * sizeof(*s->motion_val[0]));
1926 memcpy(s->motion_val[1], s1->motion_val[1], c_fragment_count * sizeof(*s->motion_val[1]));
1927 }
1928
d23845f3 1929 // copy previous frame data
759001c5
AK
1930 if ((err = ref_frames(s, s1)) < 0)
1931 return err;
1932
1933 s->keyframe = s1->keyframe;
d23845f3
AS
1934
1935 // copy qscale data if necessary
1936 for (i = 0; i < 3; i++) {
1937 if (s->qps[i] != s1->qps[1]) {
1938 qps_changed = 1;
1939 memcpy(&s->qmat[i], &s1->qmat[i], sizeof(s->qmat[i]));
1940 }
1941 }
1942
902685b8 1943 if (s->qps[0] != s1->qps[0])
d23845f3 1944 memcpy(&s->bounding_values_array, &s1->bounding_values_array, sizeof(s->bounding_values_array));
d23845f3
AS
1945
1946 if (qps_changed)
1947 copy_fields(s, s1, qps, superblock_count);
1948#undef copy_fields
1949 }
1950
759001c5 1951 return update_frames(dst);
d23845f3
AS
1952}
1953
115329f1 1954static int vp3_decode_frame(AVCodecContext *avctx,
df9b9567 1955 void *data, int *got_frame,
7a00bbad 1956 AVPacket *avpkt)
d86053a4 1957{
7a00bbad
TB
1958 const uint8_t *buf = avpkt->data;
1959 int buf_size = avpkt->size;
d86053a4
MM
1960 Vp3DecodeContext *s = avctx->priv_data;
1961 GetBitContext gb;
759001c5 1962 int i, ret;
d86053a4 1963
d86053a4 1964 init_get_bits(&gb, buf, buf_size * 8);
115329f1 1965
f44ee2c3
AB
1966 if (s->theora && get_bits1(&gb))
1967 {
bb270c08
DB
1968 av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1969 return -1;
f44ee2c3 1970 }
3c3f113e
AB
1971
1972 s->keyframe = !get_bits1(&gb);
1973 if (!s->theora)
bb270c08 1974 skip_bits(&gb, 1);
f2264fa5
DC
1975 for (i = 0; i < 3; i++)
1976 s->last_qps[i] = s->qps[i];
efea8528 1977
f2264fa5 1978 s->nqps=0;
efea8528 1979 do{
f2264fa5
DC
1980 s->qps[s->nqps++]= get_bits(&gb, 6);
1981 } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1982 for (i = s->nqps; i < 3; i++)
1983 s->qps[i] = -1;
d86053a4 1984
f8830383 1985 if (s->avctx->debug & FF_DEBUG_PICT_INFO)
bb270c08 1986 av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
6ee99a7e 1987 s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
d86053a4 1988
a4501a45
DC
1989 s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1990 avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1991
f2264fa5 1992 if (s->qps[0] != s->last_qps[0])
f44b08a5 1993 init_loop_filter(s);
f2264fa5
DC
1994
1995 for (i = 0; i < s->nqps; i++)
1996 // reinit all dequantizers if the first one changed, because
1997 // the DC of the first quantizer must be used for all matrices
1998 if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1999 init_dequantizer(s, i);
642d7e84 2000
068e82ba
DC
2001 if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
2002 return buf_size;
2003
759001c5
AK
2004 s->current_frame.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
2005 if (ff_thread_get_buffer(avctx, &s->current_frame, AV_GET_BUFFER_FLAG_REF) < 0) {
739b5090 2006 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
e72d2d12 2007 goto error;
739b5090
DC
2008 }
2009
902685b8 2010 if (!s->edge_emu_buffer)
759001c5 2011 s->edge_emu_buffer = av_malloc(9*FFABS(s->current_frame.f->linesize[0]));
902685b8 2012
d86053a4 2013 if (s->keyframe) {
bb270c08
DB
2014 if (!s->theora)
2015 {
2016 skip_bits(&gb, 4); /* width code */
2017 skip_bits(&gb, 4); /* height code */
2018 if (s->version)
2019 {
2020 s->version = get_bits(&gb, 5);
6ee99a7e 2021 if (avctx->frame_number == 0)
bb270c08
DB
2022 av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2023 }
2024 }
2025 if (s->version || s->theora)
2026 {
2027 if (get_bits1(&gb))
2028 av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2029 skip_bits(&gb, 2); /* reserved? */
2030 }
d86053a4 2031 } else {
759001c5 2032 if (!s->golden_frame.f->data[0]) {
7a4e8b59 2033 av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
ff65969f 2034
759001c5
AK
2035 s->golden_frame.f->pict_type = AV_PICTURE_TYPE_I;
2036 if (ff_thread_get_buffer(avctx, &s->golden_frame, AV_GET_BUFFER_FLAG_REF) < 0) {
7a4e8b59
DC
2037 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2038 goto error;
2039 }
759001c5
AK
2040 ff_thread_release_buffer(avctx, &s->last_frame);
2041 if ((ret = ff_thread_ref_frame(&s->last_frame, &s->golden_frame)) < 0)
2042 goto error;
8cf9a09d 2043 ff_thread_report_progress(&s->last_frame, INT_MAX, 0);
d86053a4 2044 }
d86053a4
MM
2045 }
2046
703acd54 2047 memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
d23845f3 2048 ff_thread_finish_setup(avctx);
d86053a4 2049
220a6f40
MN
2050 if (unpack_superblocks(s, &gb)){
2051 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
e72d2d12 2052 goto error;
220a6f40 2053 }
220a6f40
MN
2054 if (unpack_modes(s, &gb)){
2055 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
e72d2d12 2056 goto error;
220a6f40 2057 }
220a6f40
MN
2058 if (unpack_vectors(s, &gb)){
2059 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
e72d2d12 2060 goto error;
220a6f40 2061 }
f2264fa5
DC
2062 if (unpack_block_qpis(s, &gb)){
2063 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
e72d2d12 2064 goto error;
f2264fa5 2065 }
220a6f40
MN
2066 if (unpack_dct_coeffs(s, &gb)){
2067 av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
e72d2d12 2068 goto error;
892fc83e 2069 }
735acf56
DC
2070
2071 for (i = 0; i < 3; i++) {
1e76a1da 2072 int height = s->height >> (i && s->chroma_y_shift);
735acf56
DC
2073 if (s->flipped_image)
2074 s->data_offset[i] = 0;
2075 else
759001c5 2076 s->data_offset[i] = (height-1) * s->current_frame.f->linesize[i];
735acf56 2077 }
d86053a4 2078
a8de3901 2079 s->last_slice_end = 0;
7a095ea6 2080 for (i = 0; i < s->c_superblock_height; i++)
dc4b78d9 2081 render_slice(s, i);
d86053a4 2082
256c0662
DC
2083 // filter the last row
2084 for (i = 0; i < 3; i++) {
1e76a1da 2085 int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
256c0662
DC
2086 apply_loop_filter(s, i, row, row+1);
2087 }
83f72f13 2088 vp3_draw_horiz_band(s, s->avctx->height);
892fc83e 2089
759001c5
AK
2090 if ((ret = av_frame_ref(data, s->current_frame.f)) < 0)
2091 return ret;
df9b9567 2092 *got_frame = 1;
d86053a4 2093
759001c5
AK
2094 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME)) {
2095 ret = update_frames(avctx);
2096 if (ret < 0)
2097 return ret;
2098 }
d86053a4
MM
2099
2100 return buf_size;
e72d2d12
DC
2101
2102error:
d23845f3
AS
2103 ff_thread_report_progress(&s->current_frame, INT_MAX, 0);
2104
27237d52 2105 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_FRAME))
759001c5 2106 av_frame_unref(s->current_frame.f);
d23845f3 2107
e72d2d12 2108 return -1;
d86053a4
MM
2109}
2110
39922395
MM
2111static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2112{
2113 Vp3DecodeContext *s = avctx->priv_data;
2114
5fc32c27 2115 if (get_bits1(gb)) {
39922395
MM
2116 int token;
2117 if (s->entries >= 32) { /* overflow */
2118 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2119 return -1;
2120 }
2121 token = get_bits(gb, 5);
1218777f
DB
2122 av_dlog(avctx, "hti %d hbits %x token %d entry : %d size %d\n",
2123 s->hti, s->hbits, token, s->entries, s->huff_code_size);
39922395
MM
2124 s->huffman_table[s->hti][token][0] = s->hbits;
2125 s->huffman_table[s->hti][token][1] = s->huff_code_size;
2126 s->entries++;
2127 }
2128 else {
2129 if (s->huff_code_size >= 32) {/* overflow */
2130 av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2131 return -1;
2132 }
2133 s->huff_code_size++;
2134 s->hbits <<= 1;
00bbe276
AC
2135 if (read_huffman_tree(avctx, gb))
2136 return -1;
39922395 2137 s->hbits |= 1;
00bbe276
AC
2138 if (read_huffman_tree(avctx, gb))
2139 return -1;
39922395
MM
2140 s->hbits >>= 1;
2141 s->huff_code_size--;
2142 }
2143 return 0;
2144}
2145
bc4d1561
DB
2146static int vp3_init_thread_copy(AVCodecContext *avctx)
2147{
2148 Vp3DecodeContext *s = avctx->priv_data;
2149
2150 s->superblock_coding = NULL;
2151 s->all_fragments = NULL;
2152 s->coded_fragment_list[0] = NULL;
2153 s->dct_tokens_base = NULL;
2154 s->superblock_fragments = NULL;
2155 s->macroblock_coding = NULL;
2156 s->motion_val[0] = NULL;
2157 s->motion_val[1] = NULL;
2158 s->edge_emu_buffer = NULL;
2159
759001c5 2160 return init_frames(s);
bc4d1561
DB
2161}
2162
b250f9c6 2163#if CONFIG_THEORA_DECODER
716d413c
AK
2164static const enum AVPixelFormat theora_pix_fmts[4] = {
2165 AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P
1e76a1da
DC
2166};
2167
e278056f 2168static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
2169{
2170 Vp3DecodeContext *s = avctx->priv_data;
ea3c2d53 2171 int visible_width, visible_height, colorspace;
ddc7e438 2172 int offset_x = 0, offset_y = 0;
b53febc1 2173 int ret;
6974952d 2174 AVRational fps, aspect;
9a7ad925 2175
e278056f 2176 s->theora = get_bits_long(gb, 24);
356306ac 2177 av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
105c3d25 2178
ba7ee4a4 2179 /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
9a7ad925 2180 /* but previous versions have the image flipped relative to vp3 */
ba7ee4a4 2181 if (s->theora < 0x030200)
9a7ad925 2182 {
bb270c08 2183 s->flipped_image = 1;
9a7ad925
AB
2184 av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2185 }
f44ee2c3 2186
277e3e53
DC
2187 visible_width = s->width = get_bits(gb, 16) << 4;
2188 visible_height = s->height = get_bits(gb, 16) << 4;
115329f1 2189
277e3e53 2190 if (s->theora >= 0x030200) {
a0ce2d1b
DC
2191 visible_width = get_bits_long(gb, 24);
2192 visible_height = get_bits_long(gb, 24);
c0f716b8 2193
ddc7e438
DC
2194 offset_x = get_bits(gb, 8); /* offset x */
2195 offset_y = get_bits(gb, 8); /* offset y, from bottom */
ba4816a0 2196 }
f44ee2c3 2197
8099d6c9
DC
2198 fps.num = get_bits_long(gb, 32);
2199 fps.den = get_bits_long(gb, 32);
2200 if (fps.num && fps.den) {
6fc8226e
MS
2201 if (fps.num < 0 || fps.den < 0) {
2202 av_log(avctx, AV_LOG_ERROR, "Invalid framerate\n");
2203 return AVERROR_INVALIDDATA;
2204 }
33e62383
DC
2205 av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2206 fps.den, fps.num, 1<<30);
8099d6c9
DC
2207 }
2208
6974952d
RD
2209 aspect.num = get_bits_long(gb, 24);
2210 aspect.den = get_bits_long(gb, 24);
2211 if (aspect.num && aspect.den) {
2212 av_reduce(&avctx->sample_aspect_ratio.num,
2213 &avctx->sample_aspect_ratio.den,
2214 aspect.num, aspect.den, 1<<30);
2215 }
115329f1 2216
ba7ee4a4 2217 if (s->theora < 0x030200)
e278056f 2218 skip_bits(gb, 5); /* keyframe frequency force */
ea3c2d53 2219 colorspace = get_bits(gb, 8);
e278056f 2220 skip_bits(gb, 24); /* bitrate */
f44ee2c3 2221
e278056f 2222 skip_bits(gb, 6); /* quality hint */
115329f1 2223
ba7ee4a4 2224 if (s->theora >= 0x030200)
105c3d25 2225 {
e278056f 2226 skip_bits(gb, 5); /* keyframe frequency force */
1e76a1da 2227 avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
337f5c6e 2228 skip_bits(gb, 3); /* reserved */
105c3d25 2229 }
115329f1 2230
e278056f 2231// align_get_bits(gb);
115329f1 2232
c0f716b8 2233 if ( visible_width <= s->width && visible_width > s->width-16
ddc7e438
DC
2234 && visible_height <= s->height && visible_height > s->height-16
2235 && !offset_x && (offset_y == s->height - visible_height))
b53febc1 2236 ret = ff_set_dimensions(avctx, visible_width, visible_height);
c0f716b8 2237 else
b53febc1
AK
2238 ret = ff_set_dimensions(avctx, s->width, s->height);
2239 if (ret < 0)
2240 return ret;
f44ee2c3 2241
ea3c2d53
DC
2242 if (colorspace == 1) {
2243 avctx->color_primaries = AVCOL_PRI_BT470M;
2244 } else if (colorspace == 2) {
2245 avctx->color_primaries = AVCOL_PRI_BT470BG;
2246 }
2247 if (colorspace == 1 || colorspace == 2) {
2248 avctx->colorspace = AVCOL_SPC_BT470BG;
2249 avctx->color_trc = AVCOL_TRC_BT709;
2250 }
2251
f44ee2c3
AB
2252 return 0;
2253}
2254
e278056f 2255static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
f44ee2c3
AB
2256{
2257 Vp3DecodeContext *s = avctx->priv_data;
ae1dd8e1 2258 int i, n, matrices, inter, plane;
ba7ee4a4
MC
2259
2260 if (s->theora >= 0x030200) {
e278056f 2261 n = get_bits(gb, 3);
9c7154c7 2262 /* loop filter limit values table */
9d777508
JGG
2263 if (n)
2264 for (i = 0; i < 64; i++)
10f69158 2265 s->filter_limit_values[i] = get_bits(gb, n);
ba7ee4a4 2266 }
115329f1 2267
ba7ee4a4 2268 if (s->theora >= 0x030200)
e278056f 2269 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2270 else
2271 n = 16;
f44ee2c3
AB
2272 /* quality threshold table */
2273 for (i = 0; i < 64; i++)
e278056f 2274 s->coded_ac_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2275
ba7ee4a4 2276 if (s->theora >= 0x030200)
e278056f 2277 n = get_bits(gb, 4) + 1;
ba7ee4a4
MC
2278 else
2279 n = 16;
f44ee2c3
AB
2280 /* dc scale factor table */
2281 for (i = 0; i < 64; i++)
e278056f 2282 s->coded_dc_scale_factor[i] = get_bits(gb, n);
f44ee2c3 2283
ba7ee4a4 2284 if (s->theora >= 0x030200)
e278056f 2285 matrices = get_bits(gb, 9) + 1;
ba7ee4a4 2286 else
2da2ba03 2287 matrices = 3;
f44ee2c3 2288
ae1dd8e1
MN
2289 if(matrices > 384){
2290 av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2291 return -1;
2292 }
3c3f113e 2293
ae1dd8e1 2294 for(n=0; n<matrices; n++){
bb270c08 2295 for (i = 0; i < 64; i++)
ae1dd8e1
MN
2296 s->base_matrix[n][i]= get_bits(gb, 8);
2297 }
2da2ba03 2298
ae1dd8e1
MN
2299 for (inter = 0; inter <= 1; inter++) {
2300 for (plane = 0; plane <= 2; plane++) {
2301 int newqr= 1;
2302 if (inter || plane > 0)
5fc32c27 2303 newqr = get_bits1(gb);
39922395 2304 if (!newqr) {
ae1dd8e1 2305 int qtj, plj;
5fc32c27 2306 if(inter && get_bits1(gb)){
ae1dd8e1
MN
2307 qtj = 0;
2308 plj = plane;
2309 }else{
2310 qtj= (3*inter + plane - 1) / 3;
2311 plj= (plane + 2) % 3;
2312 }
2313 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2314 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2315 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2316 } else {
2317 int qri= 0;
39922395 2318 int qi = 0;
ae1dd8e1
MN
2319
2320 for(;;){
2321 i= get_bits(gb, av_log2(matrices-1)+1);
2322 if(i>= matrices){
2323 av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2324 return -1;
2325 }
2326 s->qr_base[inter][plane][qri]= i;
2327 if(qi >= 63)
2328 break;
2329 i = get_bits(gb, av_log2(63-qi)+1) + 1;
2330 s->qr_size[inter][plane][qri++]= i;
2331 qi += i;
39922395 2332 }
ae1dd8e1 2333
2da2ba03 2334 if (qi > 63) {
7146d2c2 2335 av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
bb270c08
DB
2336 return -1;
2337 }
ae1dd8e1 2338 s->qr_count[inter][plane]= qri;
39922395
MM
2339 }
2340 }
2341 }
2342
2da2ba03 2343 /* Huffman tables */
39922395
MM
2344 for (s->hti = 0; s->hti < 80; s->hti++) {
2345 s->entries = 0;
2346 s->huff_code_size = 1;
5fc32c27 2347 if (!get_bits1(gb)) {
39922395 2348 s->hbits = 0;
00bbe276
AC
2349 if(read_huffman_tree(avctx, gb))
2350 return -1;
39922395 2351 s->hbits = 1;
00bbe276
AC
2352 if(read_huffman_tree(avctx, gb))
2353 return -1;
39922395
MM
2354 }
2355 }
115329f1 2356
f44ee2c3 2357 s->theora_tables = 1;
115329f1 2358
f44ee2c3
AB
2359 return 0;
2360}
2361
5ef251e5 2362static av_cold int theora_decode_init(AVCodecContext *avctx)
f44ee2c3
AB
2363{
2364 Vp3DecodeContext *s = avctx->priv_data;
2365 GetBitContext gb;
2366 int ptype;
da91ed59
AJ
2367 uint8_t *header_start[3];
2368 int header_len[3];
2369 int i;
115329f1 2370
f44ee2c3
AB
2371 s->theora = 1;
2372
2373 if (!avctx->extradata_size)
7146d2c2
AB
2374 {
2375 av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
bb270c08 2376 return -1;
7146d2c2 2377 }
f44ee2c3 2378
357db4c2 2379 if (avpriv_split_xiph_headers(avctx->extradata, avctx->extradata_size,
da91ed59
AJ
2380 42, header_start, header_len) < 0) {
2381 av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2382 return -1;
2383 }
ee89b2b9 2384
da91ed59 2385 for(i=0;i<3;i++) {
e33db35b
MS
2386 if (header_len[i] <= 0)
2387 continue;
fa6f2751 2388 init_get_bits(&gb, header_start[i], header_len[i] * 8);
f44ee2c3
AB
2389
2390 ptype = get_bits(&gb, 8);
115329f1 2391
7146d2c2
AB
2392 if (!(ptype & 0x80))
2393 {
2394 av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
e278056f 2395// return -1;
115329f1 2396 }
7146d2c2 2397
3700dab4 2398 // FIXME: Check for this as well.
0a8dedc9 2399 skip_bits_long(&gb, 6*8); /* "theora" */
115329f1 2400
f44ee2c3
AB
2401 switch(ptype)
2402 {
2403 case 0x80:
e278056f 2404 theora_decode_header(avctx, &gb);
bb270c08
DB
2405 break;
2406 case 0x81:
2da2ba03 2407// FIXME: is this needed? it breaks sometimes
bb270c08
DB
2408// theora_decode_comments(avctx, gb);
2409 break;
2410 case 0x82:
00bbe276
AC
2411 if (theora_decode_tables(avctx, &gb))
2412 return -1;
bb270c08
DB
2413 break;
2414 default:
2415 av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2416 break;
f44ee2c3 2417 }
12ce1f3f
DC
2418 if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2419 av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
116d866c
MC
2420 if (s->theora < 0x030200)
2421 break;
ee89b2b9 2422 }
f44ee2c3 2423
c79c960a 2424 return vp3_decode_init(avctx);
f44ee2c3
AB
2425}
2426
d36beb3f 2427AVCodec ff_theora_decoder = {
00c3b67b 2428 .name = "theora",
b2bed932 2429 .long_name = NULL_IF_CONFIG_SMALL("Theora"),
00c3b67b 2430 .type = AVMEDIA_TYPE_VIDEO,
36ef5369 2431 .id = AV_CODEC_ID_THEORA,
00c3b67b
MS
2432 .priv_data_size = sizeof(Vp3DecodeContext),
2433 .init = theora_decode_init,
2434 .close = vp3_decode_end,
2435 .decode = vp3_decode_frame,
2436 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2437 CODEC_CAP_FRAME_THREADS,
2438 .flush = vp3_decode_flush,
8370e426 2439 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
d23845f3 2440 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context)
d86053a4 2441};
6f6a3e2a 2442#endif
f44ee2c3 2443
d36beb3f 2444AVCodec ff_vp3_decoder = {
00c3b67b 2445 .name = "vp3",
b2bed932 2446 .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
00c3b67b 2447 .type = AVMEDIA_TYPE_VIDEO,
36ef5369 2448 .id = AV_CODEC_ID_VP3,
00c3b67b
MS
2449 .priv_data_size = sizeof(Vp3DecodeContext),
2450 .init = vp3_decode_init,
2451 .close = vp3_decode_end,
2452 .decode = vp3_decode_frame,
2453 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND |
2454 CODEC_CAP_FRAME_THREADS,
2455 .flush = vp3_decode_flush,
8370e426 2456 .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp3_init_thread_copy),
00c3b67b 2457 .update_thread_context = ONLY_IF_THREADS_ENABLED(vp3_update_thread_context),
f44ee2c3 2458};