merge sbox with mix on the encryption side
[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;
9110a0e3 45 int parse_filter_info = 0;
dd9b8635 46 int coeff_offset = 0;
9110a0e3
AJ
47 int vrt_shift = 0;
48 int sub_version;
5ce117c3
AJ
49 int rows, cols;
50 int res = 1;
dd9b8635 51 int separated_coeff = buf[0] & 1;
5ce117c3
AJ
52
53 s->frames[VP56_FRAME_CURRENT].key_frame = !(buf[0] & 0x80);
54 vp56_init_dequant(s, (buf[0] >> 1) & 0x3F);
55
56 if (s->frames[VP56_FRAME_CURRENT].key_frame) {
9110a0e3
AJ
57 sub_version = buf[1] >> 3;
58 if (sub_version > 8)
59 return 0;
dd9b8635 60 s->filter_header = buf[1] & 0x06;
5ce117c3
AJ
61 if (buf[1] & 1) {
62 av_log(s->avctx, AV_LOG_ERROR, "interlacing not supported\n");
63 return 0;
64 }
dd9b8635
AJ
65 if (separated_coeff || !s->filter_header) {
66 coeff_offset = BE_16(buf+2) - 2;
67 buf += 2;
68 buf_size -= 2;
69 }
5ce117c3
AJ
70
71 rows = buf[2]; /* number of stored macroblock rows */
72 cols = buf[3]; /* number of stored macroblock cols */
73 /* buf[4] is number of displayed macroblock rows */
74 /* buf[5] is number of displayed macroblock cols */
75
76 if (16*cols != s->avctx->coded_width ||
77 16*rows != s->avctx->coded_height) {
78 avcodec_set_dimensions(s->avctx, 16*cols, 16*rows);
9e2424ce
AJ
79 if (s->avctx->extradata_size == 1) {
80 s->avctx->width -= s->avctx->extradata[0] >> 4;
81 s->avctx->height -= s->avctx->extradata[0] & 0x0F;
82 }
5ce117c3
AJ
83 res = 2;
84 }
85
86 vp56_init_range_decoder(c, buf+6, buf_size-6);
87 vp56_rac_gets(c, 2);
88
dd9b8635 89 parse_filter_info = s->filter_header;
9110a0e3
AJ
90 if (sub_version < 8)
91 vrt_shift = 5;
92 s->sub_version = sub_version;
5ce117c3 93 } else {
9110a0e3
AJ
94 if (!s->sub_version)
95 return 0;
96
dd9b8635
AJ
97 if (separated_coeff || !s->filter_header) {
98 coeff_offset = BE_16(buf+1) - 2;
99 buf += 2;
100 buf_size -= 2;
101 }
5ce117c3
AJ
102 vp56_init_range_decoder(c, buf+1, buf_size-1);
103
104 *golden_frame = vp56_rac_get(c);
dd9b8635 105 if (s->filter_header) {
5ce117c3
AJ
106 s->deblock_filtering = vp56_rac_get(c);
107 if (s->deblock_filtering)
108 vp56_rac_get(c);
9110a0e3
AJ
109 if (s->sub_version > 7)
110 parse_filter_info = vp56_rac_get(c);
dd9b8635 111 }
5ce117c3
AJ
112 }
113
114 if (parse_filter_info) {
115 if (vp56_rac_get(c)) {
116 s->filter_mode = 2;
9110a0e3 117 s->sample_variance_threshold = vp56_rac_gets(c, 5) << vrt_shift;
5ce117c3
AJ
118 s->max_vector_length = 2 << vp56_rac_gets(c, 3);
119 } else if (vp56_rac_get(c)) {
120 s->filter_mode = 1;
121 } else {
122 s->filter_mode = 0;
123 }
9110a0e3
AJ
124 if (s->sub_version > 7)
125 s->filter_selection = vp56_rac_gets(c, 4);
126 else
127 s->filter_selection = 16;
5ce117c3
AJ
128 }
129
130 vp56_rac_get(c);
dd9b8635
AJ
131
132 if (coeff_offset) {
133 vp56_init_range_decoder(&s->cc, buf+coeff_offset,
134 buf_size-coeff_offset);
135 s->ccp = &s->cc;
136 } else {
137 s->ccp = &s->c;
138 }
139
5ce117c3
AJ
140 return res;
141}
142
143static void vp6_coeff_order_table_init(vp56_context_t *s)
144{
145 int i, pos, idx = 1;
146
147 s->coeff_index_to_pos[0] = 0;
148 for (i=0; i<16; i++)
149 for (pos=1; pos<64; pos++)
150 if (s->coeff_reorder[pos] == i)
151 s->coeff_index_to_pos[idx++] = pos;
152}
153
154static void vp6_default_models_init(vp56_context_t *s)
155{
156 s->vector_model_dct[0] = 0xA2;
157 s->vector_model_dct[1] = 0xA4;
158 s->vector_model_sig[0] = 0x80;
159 s->vector_model_sig[1] = 0x80;
160
161 memcpy(s->mb_types_stats, vp56_def_mb_types_stats, sizeof(s->mb_types_stats));
162 memcpy(s->vector_model_fdv, vp6_def_fdv_vector_model, sizeof(s->vector_model_fdv));
163 memcpy(s->vector_model_pdv, vp6_def_pdv_vector_model, sizeof(s->vector_model_pdv));
164 memcpy(s->coeff_model_runv, vp6_def_runv_coeff_model, sizeof(s->coeff_model_runv));
165 memcpy(s->coeff_reorder, vp6_def_coeff_reorder, sizeof(s->coeff_reorder));
166
167 vp6_coeff_order_table_init(s);
168}
169
170static void vp6_parse_vector_models(vp56_context_t *s)
171{
172 vp56_range_coder_t *c = &s->c;
173 int comp, node;
174
175 for (comp=0; comp<2; comp++) {
176 if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][0]))
177 s->vector_model_dct[comp] = vp56_rac_gets_nn(c, 7);
178 if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][1]))
179 s->vector_model_sig[comp] = vp56_rac_gets_nn(c, 7);
180 }
181
182 for (comp=0; comp<2; comp++)
183 for (node=0; node<7; node++)
184 if (vp56_rac_get_prob(c, vp6_pdv_pct[comp][node]))
185 s->vector_model_pdv[comp][node] = vp56_rac_gets_nn(c, 7);
186
187 for (comp=0; comp<2; comp++)
188 for (node=0; node<8; node++)
189 if (vp56_rac_get_prob(c, vp6_fdv_pct[comp][node]))
190 s->vector_model_fdv[comp][node] = vp56_rac_gets_nn(c, 7);
191}
192
193static void vp6_parse_coeff_models(vp56_context_t *s)
194{
195 vp56_range_coder_t *c = &s->c;
196 int def_prob[11];
197 int node, cg, ctx, pos;
198 int ct; /* code type */
199 int pt; /* plane type (0 for Y, 1 for U or V) */
200
201 memset(def_prob, 0x80, sizeof(def_prob));
202
203 for (pt=0; pt<2; pt++)
204 for (node=0; node<11; node++)
205 if (vp56_rac_get_prob(c, vp6_dccv_pct[pt][node])) {
206 def_prob[node] = vp56_rac_gets_nn(c, 7);
207 s->coeff_model_dccv[pt][node] = def_prob[node];
208 } else if (s->frames[VP56_FRAME_CURRENT].key_frame) {
209 s->coeff_model_dccv[pt][node] = def_prob[node];
210 }
211
212 if (vp56_rac_get(c)) {
213 for (pos=1; pos<64; pos++)
214 if (vp56_rac_get_prob(c, vp6_coeff_reorder_pct[pos]))
215 s->coeff_reorder[pos] = vp56_rac_gets(c, 4);
216 vp6_coeff_order_table_init(s);
217 }
218
219 for (cg=0; cg<2; cg++)
220 for (node=0; node<14; node++)
221 if (vp56_rac_get_prob(c, vp6_runv_pct[cg][node]))
222 s->coeff_model_runv[cg][node] = vp56_rac_gets_nn(c, 7);
223
224 for (ct=0; ct<3; ct++)
225 for (pt=0; pt<2; pt++)
226 for (cg=0; cg<6; cg++)
227 for (node=0; node<11; node++)
228 if (vp56_rac_get_prob(c, vp6_ract_pct[ct][pt][cg][node])) {
229 def_prob[node] = vp56_rac_gets_nn(c, 7);
230 s->coeff_model_ract[pt][ct][cg][node] = def_prob[node];
231 } else if (s->frames[VP56_FRAME_CURRENT].key_frame) {
232 s->coeff_model_ract[pt][ct][cg][node] = def_prob[node];
233 }
234
235 /* coeff_model_dcct is a linear combination of coeff_model_dccv */
236 for (pt=0; pt<2; pt++)
237 for (ctx=0; ctx<3; ctx++)
238 for (node=0; node<5; node++)
239 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);
240}
241
d120e402 242static void vp6_parse_vector_adjustment(vp56_context_t *s, vp56_mv_t *vect)
5ce117c3
AJ
243{
244 vp56_range_coder_t *c = &s->c;
245 int comp;
246
d120e402 247 *vect = (vp56_mv_t) {0,0};
5ce117c3 248 if (s->vector_candidate_pos < 2)
d120e402 249 *vect = s->vector_candidate[0];
5ce117c3
AJ
250
251 for (comp=0; comp<2; comp++) {
252 int i, delta = 0;
253
254 if (vp56_rac_get_prob(c, s->vector_model_dct[comp])) {
255 static const uint8_t prob_order[] = {0, 1, 2, 7, 6, 5, 4};
256 for (i=0; i<sizeof(prob_order); i++) {
257 int j = prob_order[i];
258 delta |= vp56_rac_get_prob(c, s->vector_model_fdv[comp][j])<<j;
259 }
260 if (delta & 0xF0)
261 delta |= vp56_rac_get_prob(c, s->vector_model_fdv[comp][3])<<3;
262 else
263 delta |= 8;
264 } else {
265 delta = vp56_rac_get_tree(c, vp56_pva_tree,
266 s->vector_model_pdv[comp]);
267 }
268
269 if (delta && vp56_rac_get_prob(c, s->vector_model_sig[comp]))
270 delta = -delta;
271
272 if (!comp)
d120e402 273 vect->x += delta;
5ce117c3 274 else
d120e402 275 vect->y += delta;
5ce117c3
AJ
276 }
277}
278
279static void vp6_parse_coeff(vp56_context_t *s)
280{
dd9b8635 281 vp56_range_coder_t *c = s->ccp;
5ce117c3
AJ
282 uint8_t *permute = s->scantable.permutated;
283 uint8_t *model, *model2, *model3;
284 int coeff, sign, coeff_idx;
285 int b, i, cg, idx, ctx;
286 int pt = 0; /* plane type (0 for Y, 1 for U or V) */
287
288 for (b=0; b<6; b++) {
289 int ct = 1; /* code type */
290 int run = 1;
291
292 if (b > 3) pt = 1;
293
294 ctx = s->left_block[vp56_b6to4[b]].not_null_dc
295 + s->above_blocks[s->above_block_idx[b]].not_null_dc;
296 model = s->coeff_model_dccv[pt];
297 model2 = s->coeff_model_dcct[pt][ctx];
298
299 for (coeff_idx=0; coeff_idx<64; ) {
300 if ((coeff_idx>1 && ct==0) || vp56_rac_get_prob(c, model2[0])) {
301 /* parse a coeff */
302 if (coeff_idx == 0) {
303 s->left_block[vp56_b6to4[b]].not_null_dc = 1;
304 s->above_blocks[s->above_block_idx[b]].not_null_dc = 1;
305 }
306
307 if (vp56_rac_get_prob(c, model2[2])) {
308 if (vp56_rac_get_prob(c, model2[3])) {
309 idx = vp56_rac_get_tree(c, vp56_pc_tree, model);
310 coeff = vp56_coeff_bias[idx];
311 for (i=vp56_coeff_bit_length[idx]; i>=0; i--)
312 coeff += vp56_rac_get_prob(c, vp56_coeff_parse_table[idx][i]) << i;
313 } else {
314 if (vp56_rac_get_prob(c, model2[4]))
315 coeff = 3 + vp56_rac_get_prob(c, model[5]);
316 else
317 coeff = 2;
318 }
319 ct = 2;
320 } else {
321 ct = 1;
322 coeff = 1;
323 }
324 sign = vp56_rac_get(c);
325 coeff = (coeff ^ -sign) + sign;
326 if (coeff_idx)
327 coeff *= s->dequant_ac;
328 idx = s->coeff_index_to_pos[coeff_idx];
329 s->block_coeff[b][permute[idx]] = coeff;
330 run = 1;
331 } else {
332 /* parse a run */
333 ct = 0;
334 if (coeff_idx == 0) {
335 s->left_block[vp56_b6to4[b]].not_null_dc = 0;
336 s->above_blocks[s->above_block_idx[b]].not_null_dc = 0;
337 } else {
338 if (!vp56_rac_get_prob(c, model2[1]))
339 break;
340
341 model3 = s->coeff_model_runv[coeff_idx >= 6];
342 run = vp56_rac_get_tree(c, vp6_pcr_tree, model3);
343 if (!run)
344 for (run=9, i=0; i<6; i++)
345 run += vp56_rac_get_prob(c, model3[i+8]) << i;
346 }
347 }
348
349 cg = vp6_coeff_groups[coeff_idx+=run];
350 model = model2 = s->coeff_model_ract[pt][ct][cg];
351 }
352 }
353}
354
355static int vp6_adjust(int v, int t)
356{
357 int V = v, s = v >> 31;
358 V ^= s;
359 V -= s;
360 if (V-t-1 >= (unsigned)(t-1))
361 return v;
362 V = 2*t - V;
363 V += s;
364 V ^= s;
365 return V;
366}
367
368static int vp6_block_variance(uint8_t *src, int stride)
369{
370 int sum = 0, square_sum = 0;
371 int y, x;
372
373 for (y=0; y<8; y+=2) {
374 for (x=0; x<8; x+=2) {
375 sum += src[x];
376 square_sum += src[x]*src[x];
377 }
378 src += 2*stride;
379 }
c29ff23c 380 return (16*square_sum - sum*sum) >> 8;
5ce117c3
AJ
381}
382
383static void vp6_filter_hv2(vp56_context_t *s, uint8_t *dst, uint8_t *src,
384 int stride, int delta, int16_t weight)
385{
386 s->dsp.put_pixels_tab[1][0](dst, src, stride, 8);
387 s->dsp.biweight_h264_pixels_tab[3](dst, src+delta, stride, 2,
388 8-weight, weight, 0);
389}
390
391static void vp6_filter_hv4(uint8_t *dst, uint8_t *src, int stride,
392 int delta, const int16_t *weights)
393{
394 int x, y;
395
396 for (y=0; y<8; y++) {
397 for (x=0; x<8; x++) {
398 dst[x] = clip_uint8(( src[x-delta ] * weights[0]
399 + src[x ] * weights[1]
400 + src[x+delta ] * weights[2]
401 + src[x+2*delta] * weights[3] + 64) >> 7);
402 }
403 src += stride;
404 dst += stride;
405 }
406}
407
408static void vp6_filter_diag2(vp56_context_t *s, uint8_t *dst, uint8_t *src,
409 int stride, int h_weight, int v_weight)
410{
411 uint8_t *tmp = s->edge_emu_buffer+16;
412 int x, xmax;
413
414 s->dsp.put_pixels_tab[1][0](tmp, src, stride, 8);
415 s->dsp.biweight_h264_pixels_tab[3](tmp, src+1, stride, 2,
416 8-h_weight, h_weight, 0);
417 /* we need a 8x9 block to do vertical filter, so compute one more line */
418 for (x=8*stride, xmax=x+8; x<xmax; x++)
419 tmp[x] = (src[x]*(8-h_weight) + src[x+1]*h_weight + 4) >> 3;
420
421 s->dsp.put_pixels_tab[1][0](dst, tmp, stride, 8);
422 s->dsp.biweight_h264_pixels_tab[3](dst, tmp+stride, stride, 2,
423 8-v_weight, v_weight, 0);
424}
425
426static void vp6_filter_diag4(uint8_t *dst, uint8_t *src, int stride,
427 const int16_t *h_weights,const int16_t *v_weights)
428{
429 int x, y;
430 int tmp[8*11];
431 int *t = tmp;
432
433 src -= stride;
434
435 for (y=0; y<11; y++) {
436 for (x=0; x<8; x++) {
437 t[x] = clip_uint8(( src[x-1] * h_weights[0]
438 + src[x ] * h_weights[1]
439 + src[x+1] * h_weights[2]
440 + src[x+2] * h_weights[3] + 64) >> 7);
441 }
442 src += stride;
443 t += 8;
444 }
445
446 t = tmp + 8;
447 for (y=0; y<8; y++) {
448 for (x=0; x<8; x++) {
449 dst[x] = clip_uint8(( t[x-8 ] * v_weights[0]
450 + t[x ] * v_weights[1]
451 + t[x+8 ] * v_weights[2]
452 + t[x+16] * v_weights[3] + 64) >> 7);
453 }
454 dst += stride;
455 t += 8;
456 }
457}
458
459static void vp6_filter(vp56_context_t *s, uint8_t *dst, uint8_t *src,
460 int offset1, int offset2, int stride,
461 vp56_mv_t mv, int mask, int select, int luma)
462{
463 int filter4 = 0;
464 int x8 = mv.x & mask;
465 int y8 = mv.y & mask;
466
467 if (luma) {
468 x8 *= 2;
469 y8 *= 2;
470 filter4 = s->filter_mode;
471 if (filter4 == 2) {
472 if (s->max_vector_length &&
c26abfa5
DB
473 (FFABS(mv.x) > s->max_vector_length ||
474 FFABS(mv.y) > s->max_vector_length)) {
5ce117c3 475 filter4 = 0;
9110a0e3
AJ
476 } else if (s->sample_variance_threshold
477 && (vp6_block_variance(src+offset1, stride)
5ce117c3
AJ
478 < s->sample_variance_threshold)) {
479 filter4 = 0;
480 }
481 }
482 }
483
484 if ((y8 && (offset2-offset1)*s->flip<0) || (!y8 && offset1 > offset2)) {
485 offset1 = offset2;
486 }
487
488 if (filter4) {
489 if (!y8) { /* left or right combine */
490 vp6_filter_hv4(dst, src+offset1, stride, 1,
491 vp6_block_copy_filter[select][x8]);
492 } else if (!x8) { /* above or below combine */
493 vp6_filter_hv4(dst, src+offset1, stride, stride,
494 vp6_block_copy_filter[select][y8]);
495 } else if ((mv.x^mv.y) >> 31) { /* lower-left or upper-right combine */
496 vp6_filter_diag4(dst, src+offset1-1, stride,
497 vp6_block_copy_filter[select][x8],
498 vp6_block_copy_filter[select][y8]);
499 } else { /* lower-right or upper-left combine */
500 vp6_filter_diag4(dst, src+offset1, stride,
501 vp6_block_copy_filter[select][x8],
502 vp6_block_copy_filter[select][y8]);
503 }
504 } else {
505 if (!y8) { /* left or right combine */
506 vp6_filter_hv2(s, dst, src+offset1, stride, 1, x8);
507 } else if (!x8) { /* above or below combine */
508 vp6_filter_hv2(s, dst, src+offset1, stride, stride, y8);
509 } else if ((mv.x^mv.y) >> 31) { /* lower-left or upper-right combine */
510 vp6_filter_diag2(s, dst, src+offset1-1, stride, x8, y8);
511 } else { /* lower-right or upper-left combine */
512 vp6_filter_diag2(s, dst, src+offset1, stride, x8, y8);
513 }
514 }
515}
516
517static int vp6_decode_init(AVCodecContext *avctx)
518{
519 vp56_context_t *s = avctx->priv_data;
520
521 vp56_init(s, avctx, avctx->codec->id == CODEC_ID_VP6);
522 s->vp56_coord_div = vp6_coord_div;
523 s->parse_vector_adjustment = vp6_parse_vector_adjustment;
524 s->adjust = vp6_adjust;
525 s->filter = vp6_filter;
526 s->parse_coeff = vp6_parse_coeff;
527 s->default_models_init = vp6_default_models_init;
528 s->parse_vector_models = vp6_parse_vector_models;
529 s->parse_coeff_models = vp6_parse_coeff_models;
530 s->parse_header = vp6_parse_header;
531
532 return 0;
533}
534
535AVCodec vp6_decoder = {
536 "vp6",
537 CODEC_TYPE_VIDEO,
538 CODEC_ID_VP6,
539 sizeof(vp56_context_t),
540 vp6_decode_init,
541 NULL,
542 vp56_free,
543 vp56_decode_frame,
544};
545
546/* flash version, not flipped upside-down */
547AVCodec vp6f_decoder = {
548 "vp6f",
549 CODEC_TYPE_VIDEO,
550 CODEC_ID_VP6F,
551 sizeof(vp56_context_t),
552 vp6_decode_init,
553 NULL,
554 vp56_free,
555 vp56_decode_frame,
556};