10l
[libav.git] / libavcodec / vp6.c
CommitLineData
5ce117c3
AJ
1/**
2 * @file vp6.c
3 * VP6 compatible video decoder
4 *
5 * Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org>
6 *
b78e7197
DB
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
5ce117c3
AJ
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
b78e7197 14 * FFmpeg is distributed in the hope that it will be useful,
5ce117c3
AJ
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
b78e7197 20 * License along with FFmpeg; if not, write to the Free Software
5ce117c3 21 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
246f86a4
AJ
22 *
23 *
24 * The VP6F decoder accept an optional 1 byte extradata. It is composed of:
25 * - upper 4bits: difference between encoded width and visible width
26 * - lower 4bits: difference between encoded height and visible height
5ce117c3
AJ
27 */
28
29#include <stdlib.h>
5ce117c3
AJ
30
31#include "avcodec.h"
32#include "dsputil.h"
33#include "bitstream.h"
34#include "mpegvideo.h"
35
36#include "vp56.h"
37#include "vp56data.h"
38#include "vp6data.h"
39
40
41static int vp6_parse_header(vp56_context_t *s, uint8_t *buf, int buf_size,
42 int *golden_frame)
43{
44 vp56_range_coder_t *c = &s->c;
45 int parse_filter_info;
46 int rows, cols;
47 int res = 1;
48
49 if (buf[0] & 1)
50 return 0;
51
52 s->frames[VP56_FRAME_CURRENT].key_frame = !(buf[0] & 0x80);
53 vp56_init_dequant(s, (buf[0] >> 1) & 0x3F);
54
55 if (s->frames[VP56_FRAME_CURRENT].key_frame) {
56 if ((buf[1] & 0xFE) != 0x46) /* would be 0x36 for VP61 */
57 return 0;
58 if (buf[1] & 1) {
59 av_log(s->avctx, AV_LOG_ERROR, "interlacing not supported\n");
60 return 0;
61 }
62
63 rows = buf[2]; /* number of stored macroblock rows */
64 cols = buf[3]; /* number of stored macroblock cols */
65 /* buf[4] is number of displayed macroblock rows */
66 /* buf[5] is number of displayed macroblock cols */
67
68 if (16*cols != s->avctx->coded_width ||
69 16*rows != s->avctx->coded_height) {
70 avcodec_set_dimensions(s->avctx, 16*cols, 16*rows);
9e2424ce
AJ
71 if (s->avctx->extradata_size == 1) {
72 s->avctx->width -= s->avctx->extradata[0] >> 4;
73 s->avctx->height -= s->avctx->extradata[0] & 0x0F;
74 }
5ce117c3
AJ
75 res = 2;
76 }
77
78 vp56_init_range_decoder(c, buf+6, buf_size-6);
79 vp56_rac_gets(c, 2);
80
81 parse_filter_info = 1;
82 } else {
83 vp56_init_range_decoder(c, buf+1, buf_size-1);
84
85 *golden_frame = vp56_rac_get(c);
86 s->deblock_filtering = vp56_rac_get(c);
87 if (s->deblock_filtering)
88 vp56_rac_get(c);
89 parse_filter_info = vp56_rac_get(c);
90 }
91
92 if (parse_filter_info) {
93 if (vp56_rac_get(c)) {
94 s->filter_mode = 2;
95 s->sample_variance_threshold = vp56_rac_gets(c, 5);
96 s->max_vector_length = 2 << vp56_rac_gets(c, 3);
97 } else if (vp56_rac_get(c)) {
98 s->filter_mode = 1;
99 } else {
100 s->filter_mode = 0;
101 }
102 s->filter_selection = vp56_rac_gets(c, 4);
103 }
104
105 vp56_rac_get(c);
106 return res;
107}
108
109static void vp6_coeff_order_table_init(vp56_context_t *s)
110{
111 int i, pos, idx = 1;
112
113 s->coeff_index_to_pos[0] = 0;
114 for (i=0; i<16; i++)
115 for (pos=1; pos<64; pos++)
116 if (s->coeff_reorder[pos] == i)
117 s->coeff_index_to_pos[idx++] = pos;
118}
119
120static void vp6_default_models_init(vp56_context_t *s)
121{
122 s->vector_model_dct[0] = 0xA2;
123 s->vector_model_dct[1] = 0xA4;
124 s->vector_model_sig[0] = 0x80;
125 s->vector_model_sig[1] = 0x80;
126
127 memcpy(s->mb_types_stats, vp56_def_mb_types_stats, sizeof(s->mb_types_stats));
128 memcpy(s->vector_model_fdv, vp6_def_fdv_vector_model, sizeof(s->vector_model_fdv));
129 memcpy(s->vector_model_pdv, vp6_def_pdv_vector_model, sizeof(s->vector_model_pdv));
130 memcpy(s->coeff_model_runv, vp6_def_runv_coeff_model, sizeof(s->coeff_model_runv));
131 memcpy(s->coeff_reorder, vp6_def_coeff_reorder, sizeof(s->coeff_reorder));
132
133 vp6_coeff_order_table_init(s);
134}
135
136static void vp6_parse_vector_models(vp56_context_t *s)
137{
138 vp56_range_coder_t *c = &s->c;
139 int comp, node;
140
141 for (comp=0; comp<2; comp++) {
142 if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][0]))
143 s->vector_model_dct[comp] = vp56_rac_gets_nn(c, 7);
144 if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][1]))
145 s->vector_model_sig[comp] = vp56_rac_gets_nn(c, 7);
146 }
147
148 for (comp=0; comp<2; comp++)
149 for (node=0; node<7; node++)
150 if (vp56_rac_get_prob(c, vp6_pdv_pct[comp][node]))
151 s->vector_model_pdv[comp][node] = vp56_rac_gets_nn(c, 7);
152
153 for (comp=0; comp<2; comp++)
154 for (node=0; node<8; node++)
155 if (vp56_rac_get_prob(c, vp6_fdv_pct[comp][node]))
156 s->vector_model_fdv[comp][node] = vp56_rac_gets_nn(c, 7);
157}
158
159static void vp6_parse_coeff_models(vp56_context_t *s)
160{
161 vp56_range_coder_t *c = &s->c;
162 int def_prob[11];
163 int node, cg, ctx, pos;
164 int ct; /* code type */
165 int pt; /* plane type (0 for Y, 1 for U or V) */
166
167 memset(def_prob, 0x80, sizeof(def_prob));
168
169 for (pt=0; pt<2; pt++)
170 for (node=0; node<11; node++)
171 if (vp56_rac_get_prob(c, vp6_dccv_pct[pt][node])) {
172 def_prob[node] = vp56_rac_gets_nn(c, 7);
173 s->coeff_model_dccv[pt][node] = def_prob[node];
174 } else if (s->frames[VP56_FRAME_CURRENT].key_frame) {
175 s->coeff_model_dccv[pt][node] = def_prob[node];
176 }
177
178 if (vp56_rac_get(c)) {
179 for (pos=1; pos<64; pos++)
180 if (vp56_rac_get_prob(c, vp6_coeff_reorder_pct[pos]))
181 s->coeff_reorder[pos] = vp56_rac_gets(c, 4);
182 vp6_coeff_order_table_init(s);
183 }
184
185 for (cg=0; cg<2; cg++)
186 for (node=0; node<14; node++)
187 if (vp56_rac_get_prob(c, vp6_runv_pct[cg][node]))
188 s->coeff_model_runv[cg][node] = vp56_rac_gets_nn(c, 7);
189
190 for (ct=0; ct<3; ct++)
191 for (pt=0; pt<2; pt++)
192 for (cg=0; cg<6; cg++)
193 for (node=0; node<11; node++)
194 if (vp56_rac_get_prob(c, vp6_ract_pct[ct][pt][cg][node])) {
195 def_prob[node] = vp56_rac_gets_nn(c, 7);
196 s->coeff_model_ract[pt][ct][cg][node] = def_prob[node];
197 } else if (s->frames[VP56_FRAME_CURRENT].key_frame) {
198 s->coeff_model_ract[pt][ct][cg][node] = def_prob[node];
199 }
200
201 /* coeff_model_dcct is a linear combination of coeff_model_dccv */
202 for (pt=0; pt<2; pt++)
203 for (ctx=0; ctx<3; ctx++)
204 for (node=0; node<5; node++)
205 s->coeff_model_dcct[pt][ctx][node] = clip(((s->coeff_model_dccv[pt][node] * vp6_dccv_lc[ctx][node][0] + 128) >> 8) + vp6_dccv_lc[ctx][node][1], 1, 255);
206}
207
d120e402 208static void vp6_parse_vector_adjustment(vp56_context_t *s, vp56_mv_t *vect)
5ce117c3
AJ
209{
210 vp56_range_coder_t *c = &s->c;
211 int comp;
212
d120e402 213 *vect = (vp56_mv_t) {0,0};
5ce117c3 214 if (s->vector_candidate_pos < 2)
d120e402 215 *vect = s->vector_candidate[0];
5ce117c3
AJ
216
217 for (comp=0; comp<2; comp++) {
218 int i, delta = 0;
219
220 if (vp56_rac_get_prob(c, s->vector_model_dct[comp])) {
221 static const uint8_t prob_order[] = {0, 1, 2, 7, 6, 5, 4};
222 for (i=0; i<sizeof(prob_order); i++) {
223 int j = prob_order[i];
224 delta |= vp56_rac_get_prob(c, s->vector_model_fdv[comp][j])<<j;
225 }
226 if (delta & 0xF0)
227 delta |= vp56_rac_get_prob(c, s->vector_model_fdv[comp][3])<<3;
228 else
229 delta |= 8;
230 } else {
231 delta = vp56_rac_get_tree(c, vp56_pva_tree,
232 s->vector_model_pdv[comp]);
233 }
234
235 if (delta && vp56_rac_get_prob(c, s->vector_model_sig[comp]))
236 delta = -delta;
237
238 if (!comp)
d120e402 239 vect->x += delta;
5ce117c3 240 else
d120e402 241 vect->y += delta;
5ce117c3
AJ
242 }
243}
244
245static void vp6_parse_coeff(vp56_context_t *s)
246{
247 vp56_range_coder_t *c = &s->c;
248 uint8_t *permute = s->scantable.permutated;
249 uint8_t *model, *model2, *model3;
250 int coeff, sign, coeff_idx;
251 int b, i, cg, idx, ctx;
252 int pt = 0; /* plane type (0 for Y, 1 for U or V) */
253
254 for (b=0; b<6; b++) {
255 int ct = 1; /* code type */
256 int run = 1;
257
258 if (b > 3) pt = 1;
259
260 ctx = s->left_block[vp56_b6to4[b]].not_null_dc
261 + s->above_blocks[s->above_block_idx[b]].not_null_dc;
262 model = s->coeff_model_dccv[pt];
263 model2 = s->coeff_model_dcct[pt][ctx];
264
265 for (coeff_idx=0; coeff_idx<64; ) {
266 if ((coeff_idx>1 && ct==0) || vp56_rac_get_prob(c, model2[0])) {
267 /* parse a coeff */
268 if (coeff_idx == 0) {
269 s->left_block[vp56_b6to4[b]].not_null_dc = 1;
270 s->above_blocks[s->above_block_idx[b]].not_null_dc = 1;
271 }
272
273 if (vp56_rac_get_prob(c, model2[2])) {
274 if (vp56_rac_get_prob(c, model2[3])) {
275 idx = vp56_rac_get_tree(c, vp56_pc_tree, model);
276 coeff = vp56_coeff_bias[idx];
277 for (i=vp56_coeff_bit_length[idx]; i>=0; i--)
278 coeff += vp56_rac_get_prob(c, vp56_coeff_parse_table[idx][i]) << i;
279 } else {
280 if (vp56_rac_get_prob(c, model2[4]))
281 coeff = 3 + vp56_rac_get_prob(c, model[5]);
282 else
283 coeff = 2;
284 }
285 ct = 2;
286 } else {
287 ct = 1;
288 coeff = 1;
289 }
290 sign = vp56_rac_get(c);
291 coeff = (coeff ^ -sign) + sign;
292 if (coeff_idx)
293 coeff *= s->dequant_ac;
294 idx = s->coeff_index_to_pos[coeff_idx];
295 s->block_coeff[b][permute[idx]] = coeff;
296 run = 1;
297 } else {
298 /* parse a run */
299 ct = 0;
300 if (coeff_idx == 0) {
301 s->left_block[vp56_b6to4[b]].not_null_dc = 0;
302 s->above_blocks[s->above_block_idx[b]].not_null_dc = 0;
303 } else {
304 if (!vp56_rac_get_prob(c, model2[1]))
305 break;
306
307 model3 = s->coeff_model_runv[coeff_idx >= 6];
308 run = vp56_rac_get_tree(c, vp6_pcr_tree, model3);
309 if (!run)
310 for (run=9, i=0; i<6; i++)
311 run += vp56_rac_get_prob(c, model3[i+8]) << i;
312 }
313 }
314
315 cg = vp6_coeff_groups[coeff_idx+=run];
316 model = model2 = s->coeff_model_ract[pt][ct][cg];
317 }
318 }
319}
320
321static int vp6_adjust(int v, int t)
322{
323 int V = v, s = v >> 31;
324 V ^= s;
325 V -= s;
326 if (V-t-1 >= (unsigned)(t-1))
327 return v;
328 V = 2*t - V;
329 V += s;
330 V ^= s;
331 return V;
332}
333
334static int vp6_block_variance(uint8_t *src, int stride)
335{
336 int sum = 0, square_sum = 0;
337 int y, x;
338
339 for (y=0; y<8; y+=2) {
340 for (x=0; x<8; x+=2) {
341 sum += src[x];
342 square_sum += src[x]*src[x];
343 }
344 src += 2*stride;
345 }
c29ff23c 346 return (16*square_sum - sum*sum) >> 8;
5ce117c3
AJ
347}
348
349static void vp6_filter_hv2(vp56_context_t *s, uint8_t *dst, uint8_t *src,
350 int stride, int delta, int16_t weight)
351{
352 s->dsp.put_pixels_tab[1][0](dst, src, stride, 8);
353 s->dsp.biweight_h264_pixels_tab[3](dst, src+delta, stride, 2,
354 8-weight, weight, 0);
355}
356
357static void vp6_filter_hv4(uint8_t *dst, uint8_t *src, int stride,
358 int delta, const int16_t *weights)
359{
360 int x, y;
361
362 for (y=0; y<8; y++) {
363 for (x=0; x<8; x++) {
364 dst[x] = clip_uint8(( src[x-delta ] * weights[0]
365 + src[x ] * weights[1]
366 + src[x+delta ] * weights[2]
367 + src[x+2*delta] * weights[3] + 64) >> 7);
368 }
369 src += stride;
370 dst += stride;
371 }
372}
373
374static void vp6_filter_diag2(vp56_context_t *s, uint8_t *dst, uint8_t *src,
375 int stride, int h_weight, int v_weight)
376{
377 uint8_t *tmp = s->edge_emu_buffer+16;
378 int x, xmax;
379
380 s->dsp.put_pixels_tab[1][0](tmp, src, stride, 8);
381 s->dsp.biweight_h264_pixels_tab[3](tmp, src+1, stride, 2,
382 8-h_weight, h_weight, 0);
383 /* we need a 8x9 block to do vertical filter, so compute one more line */
384 for (x=8*stride, xmax=x+8; x<xmax; x++)
385 tmp[x] = (src[x]*(8-h_weight) + src[x+1]*h_weight + 4) >> 3;
386
387 s->dsp.put_pixels_tab[1][0](dst, tmp, stride, 8);
388 s->dsp.biweight_h264_pixels_tab[3](dst, tmp+stride, stride, 2,
389 8-v_weight, v_weight, 0);
390}
391
392static void vp6_filter_diag4(uint8_t *dst, uint8_t *src, int stride,
393 const int16_t *h_weights,const int16_t *v_weights)
394{
395 int x, y;
396 int tmp[8*11];
397 int *t = tmp;
398
399 src -= stride;
400
401 for (y=0; y<11; y++) {
402 for (x=0; x<8; x++) {
403 t[x] = clip_uint8(( src[x-1] * h_weights[0]
404 + src[x ] * h_weights[1]
405 + src[x+1] * h_weights[2]
406 + src[x+2] * h_weights[3] + 64) >> 7);
407 }
408 src += stride;
409 t += 8;
410 }
411
412 t = tmp + 8;
413 for (y=0; y<8; y++) {
414 for (x=0; x<8; x++) {
415 dst[x] = clip_uint8(( t[x-8 ] * v_weights[0]
416 + t[x ] * v_weights[1]
417 + t[x+8 ] * v_weights[2]
418 + t[x+16] * v_weights[3] + 64) >> 7);
419 }
420 dst += stride;
421 t += 8;
422 }
423}
424
425static void vp6_filter(vp56_context_t *s, uint8_t *dst, uint8_t *src,
426 int offset1, int offset2, int stride,
427 vp56_mv_t mv, int mask, int select, int luma)
428{
429 int filter4 = 0;
430 int x8 = mv.x & mask;
431 int y8 = mv.y & mask;
432
433 if (luma) {
434 x8 *= 2;
435 y8 *= 2;
436 filter4 = s->filter_mode;
437 if (filter4 == 2) {
438 if (s->max_vector_length &&
c26abfa5
DB
439 (FFABS(mv.x) > s->max_vector_length ||
440 FFABS(mv.y) > s->max_vector_length)) {
5ce117c3
AJ
441 filter4 = 0;
442 } else if (!s->sample_variance_threshold
443 || (vp6_block_variance(src+offset1, stride)
444 < s->sample_variance_threshold)) {
445 filter4 = 0;
446 }
447 }
448 }
449
450 if ((y8 && (offset2-offset1)*s->flip<0) || (!y8 && offset1 > offset2)) {
451 offset1 = offset2;
452 }
453
454 if (filter4) {
455 if (!y8) { /* left or right combine */
456 vp6_filter_hv4(dst, src+offset1, stride, 1,
457 vp6_block_copy_filter[select][x8]);
458 } else if (!x8) { /* above or below combine */
459 vp6_filter_hv4(dst, src+offset1, stride, stride,
460 vp6_block_copy_filter[select][y8]);
461 } else if ((mv.x^mv.y) >> 31) { /* lower-left or upper-right combine */
462 vp6_filter_diag4(dst, src+offset1-1, stride,
463 vp6_block_copy_filter[select][x8],
464 vp6_block_copy_filter[select][y8]);
465 } else { /* lower-right or upper-left combine */
466 vp6_filter_diag4(dst, src+offset1, stride,
467 vp6_block_copy_filter[select][x8],
468 vp6_block_copy_filter[select][y8]);
469 }
470 } else {
471 if (!y8) { /* left or right combine */
472 vp6_filter_hv2(s, dst, src+offset1, stride, 1, x8);
473 } else if (!x8) { /* above or below combine */
474 vp6_filter_hv2(s, dst, src+offset1, stride, stride, y8);
475 } else if ((mv.x^mv.y) >> 31) { /* lower-left or upper-right combine */
476 vp6_filter_diag2(s, dst, src+offset1-1, stride, x8, y8);
477 } else { /* lower-right or upper-left combine */
478 vp6_filter_diag2(s, dst, src+offset1, stride, x8, y8);
479 }
480 }
481}
482
483static int vp6_decode_init(AVCodecContext *avctx)
484{
485 vp56_context_t *s = avctx->priv_data;
486
487 vp56_init(s, avctx, avctx->codec->id == CODEC_ID_VP6);
488 s->vp56_coord_div = vp6_coord_div;
489 s->parse_vector_adjustment = vp6_parse_vector_adjustment;
490 s->adjust = vp6_adjust;
491 s->filter = vp6_filter;
492 s->parse_coeff = vp6_parse_coeff;
493 s->default_models_init = vp6_default_models_init;
494 s->parse_vector_models = vp6_parse_vector_models;
495 s->parse_coeff_models = vp6_parse_coeff_models;
496 s->parse_header = vp6_parse_header;
497
498 return 0;
499}
500
501AVCodec vp6_decoder = {
502 "vp6",
503 CODEC_TYPE_VIDEO,
504 CODEC_ID_VP6,
505 sizeof(vp56_context_t),
506 vp6_decode_init,
507 NULL,
508 vp56_free,
509 vp56_decode_frame,
510};
511
512/* flash version, not flipped upside-down */
513AVCodec vp6f_decoder = {
514 "vp6f",
515 CODEC_TYPE_VIDEO,
516 CODEC_ID_VP6F,
517 sizeof(vp56_context_t),
518 vp6_decode_init,
519 NULL,
520 vp56_free,
521 vp56_decode_frame,
522};