svq3 decoder by anonymous
[libav.git] / libavcodec / svq3.c
CommitLineData
8b82a956
MN
1/*
2 * Copyright (c) 2003 The FFmpeg Project.
3 *
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
8 *
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
13 *
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 *
18 *
19 * How to use this decoder:
20 * SVQ3 data is transported within Apple Quicktime files. Quicktime files
21 * have stsd atoms to describe media trak properties. Sometimes the stsd
22 * atom contains information that the decoder must know in order to function
23 * properly. Such is the case with SVQ3. In order to get the best use out
24 * of this decoder, the calling app must make the video stsd atom available
25 * via the AVCodecContext's extradata[_size] field:
26 *
27 * AVCodecContext.extradata = pointer to stsd, first characters are expected
28 * to be 's', 't', 's', and 'd', NOT the atom length
29 * AVCodecContext.extradata_size = size of stsd atom memory buffer (which
30 * will be the same as the stsd atom size field from the QT file, minus 4
31 * bytes since the length is missing.
32 *
33 */
34
35/**
36 * @file svq3.c
37 * svq3 decoder.
38 */
39
40static const uint8_t svq3_scan[16]={
41 0+0*4, 1+0*4, 2+0*4, 2+1*4,
42 2+2*4, 3+0*4, 3+1*4, 3+2*4,
43 0+1*4, 0+2*4, 1+1*4, 1+2*4,
44 0+3*4, 1+3*4, 2+3*4, 3+3*4,
45};
46
47static const uint8_t svq3_pred_0[25][2] = {
48 { 0, 0 },
49 { 1, 0 }, { 0, 1 },
50 { 0, 2 }, { 1, 1 }, { 2, 0 },
51 { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },
52 { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },
53 { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },
54 { 2, 4 }, { 3, 3 }, { 4, 2 },
55 { 4, 3 }, { 3, 4 },
56 { 4, 4 }
57};
58
59static const int8_t svq3_pred_1[6][6][5] = {
60 { { 2,-1,-1,-1,-1 }, { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 },
61 { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 }, { 1, 2,-1,-1,-1 } },
62 { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 4, 3 }, { 0, 1, 2, 4, 3 },
63 { 0, 2, 1, 4, 3 }, { 2, 0, 1, 3, 4 }, { 0, 4, 2, 1, 3 } },
64 { { 2, 0,-1,-1,-1 }, { 2, 1, 0, 4, 3 }, { 1, 2, 4, 0, 3 },
65 { 2, 1, 0, 4, 3 }, { 2, 1, 4, 3, 0 }, { 1, 2, 4, 0, 3 } },
66 { { 2, 0,-1,-1,-1 }, { 2, 0, 1, 4, 3 }, { 1, 2, 0, 4, 3 },
67 { 2, 1, 0, 4, 3 }, { 2, 1, 3, 4, 0 }, { 2, 4, 1, 0, 3 } },
68 { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 3, 4 }, { 1, 2, 3, 0, 4 },
69 { 2, 0, 1, 3, 4 }, { 2, 1, 3, 0, 4 }, { 2, 0, 4, 3, 1 } },
70 { { 0, 2,-1,-1,-1 }, { 0, 2, 4, 1, 3 }, { 1, 4, 2, 0, 3 },
71 { 4, 2, 0, 1, 3 }, { 2, 0, 1, 4, 3 }, { 4, 2, 1, 0, 3 } },
72};
73
74static const struct { uint8_t run; uint8_t level; } svq3_dct_tables[2][16] = {
75 { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },
76 { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },
77 { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },
78 { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }
79};
80
81static const uint32_t svq3_dequant_coeff[32] = {
82 3881, 4351, 4890, 5481, 6154, 6914, 7761, 8718,
83 9781, 10987, 12339, 13828, 15523, 17435, 19561, 21873,
84 24552, 27656, 30847, 34870, 38807, 43747, 49103, 54683,
85 61694, 68745, 77615, 89113,100253,109366,126635,141533
86};
87
88
89static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
90 const int qmul= svq3_dequant_coeff[qp];
91#define stride 16
92 int i;
93 int temp[16];
94 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
95 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
96
97 for(i=0; i<4; i++){
98 const int offset= y_offset[i];
99 const int z0= 13*(block[offset+stride*0] + block[offset+stride*4]);
100 const int z1= 13*(block[offset+stride*0] - block[offset+stride*4]);
101 const int z2= 7* block[offset+stride*1] - 17*block[offset+stride*5];
102 const int z3= 17* block[offset+stride*1] + 7*block[offset+stride*5];
103
104 temp[4*i+0]= z0+z3;
105 temp[4*i+1]= z1+z2;
106 temp[4*i+2]= z1-z2;
107 temp[4*i+3]= z0-z3;
108 }
109
110 for(i=0; i<4; i++){
111 const int offset= x_offset[i];
112 const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
113 const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
114 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
115 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
116
117 block[stride*0 +offset]= ((z0 + z3)*qmul + 0x80000)>>20;
118 block[stride*2 +offset]= ((z1 + z2)*qmul + 0x80000)>>20;
119 block[stride*8 +offset]= ((z1 - z2)*qmul + 0x80000)>>20;
120 block[stride*10+offset]= ((z0 - z3)*qmul + 0x80000)>>20;
121 }
122}
123#undef stride
124
125static void svq3_add_idct_c (uint8_t *dst, DCTELEM *block, int stride, int qp, int dc){
126 const int qmul= svq3_dequant_coeff[qp];
127 int i;
128 uint8_t *cm = cropTbl + MAX_NEG_CROP;
129
130 if (dc) {
131 dc = 13*13*((dc == 1) ? 1538*block[0] : ((qmul*(block[0] >> 3)) / 2));
132 block[0] = 0;
133 }
134
135 for (i=0; i < 4; i++) {
136 const int z0= 13*(block[0 + 4*i] + block[2 + 4*i]);
137 const int z1= 13*(block[0 + 4*i] - block[2 + 4*i]);
138 const int z2= 7* block[1 + 4*i] - 17*block[3 + 4*i];
139 const int z3= 17* block[1 + 4*i] + 7*block[3 + 4*i];
140
141 block[0 + 4*i]= z0 + z3;
142 block[1 + 4*i]= z1 + z2;
143 block[2 + 4*i]= z1 - z2;
144 block[3 + 4*i]= z0 - z3;
145 }
146
147 for (i=0; i < 4; i++) {
148 const int z0= 13*(block[i + 4*0] + block[i + 4*2]);
149 const int z1= 13*(block[i + 4*0] - block[i + 4*2]);
150 const int z2= 7* block[i + 4*1] - 17*block[i + 4*3];
151 const int z3= 17* block[i + 4*1] + 7*block[i + 4*3];
152 const int rr= (dc + 0x80000);
153
154 dst[i + stride*0]= cm[ dst[i + stride*0] + (((z0 + z3)*qmul + rr) >> 20) ];
155 dst[i + stride*1]= cm[ dst[i + stride*1] + (((z1 + z2)*qmul + rr) >> 20) ];
156 dst[i + stride*2]= cm[ dst[i + stride*2] + (((z1 - z2)*qmul + rr) >> 20) ];
157 dst[i + stride*3]= cm[ dst[i + stride*3] + (((z0 - z3)*qmul + rr) >> 20) ];
158 }
159}
160
161static void pred4x4_down_left_svq3_c(uint8_t *src, uint8_t *topright, int stride){
162 LOAD_TOP_EDGE
163 LOAD_LEFT_EDGE
164 const __attribute__((unused)) int unu0= t0;
165 const __attribute__((unused)) int unu1= l0;
166
167 src[0+0*stride]=(l1 + t1)>>1;
168 src[1+0*stride]=
169 src[0+1*stride]=(l2 + t2)>>1;
170 src[2+0*stride]=
171 src[1+1*stride]=
172 src[0+2*stride]=
173 src[3+0*stride]=
174 src[2+1*stride]=
175 src[1+2*stride]=
176 src[0+3*stride]=
177 src[3+1*stride]=
178 src[2+2*stride]=
179 src[1+3*stride]=
180 src[3+2*stride]=
181 src[2+3*stride]=
182 src[3+3*stride]=(l3 + t3)>>1;
183};
184
185static void pred16x16_plane_svq3_c(uint8_t *src, int stride){
186 pred16x16_plane_compat_c(src, stride, 1);
187}
188
189static inline int svq3_decode_block (GetBitContext *gb, DCTELEM *block,
190 int index, const int type) {
191
192 static const uint8_t *const scan_patterns[4] =
193 { luma_dc_zigzag_scan, zigzag_scan, svq3_scan, chroma_dc_scan };
194
195 int run, level, sign, vlc, limit;
196 const int intra = (3 * type) >> 2;
197 const uint8_t *const scan = scan_patterns[type];
198
199 for (limit=(16 >> intra); index < 16; index=limit, limit+=8) {
200 for (; (vlc = svq3_get_ue_golomb (gb)) != 0; index++) {
201
202 if (vlc == INVALID_VLC)
203 return -1;
204
205 sign = (vlc & 0x1) - 1;
206 vlc = (vlc + 1) >> 1;
207
208 if (type == 3) {
209 if (vlc < 3) {
210 run = 0;
211 level = vlc;
212 } else if (vlc < 4) {
213 run = 1;
214 level = 1;
215 } else {
216 run = (vlc & 0x3);
217 level = ((vlc + 9) >> 2) - run;
218 }
219 } else {
220 if (vlc < 16) {
221 run = svq3_dct_tables[intra][vlc].run;
222 level = svq3_dct_tables[intra][vlc].level;
223 } else if (intra) {
224 run = (vlc & 0x7);
225 level = (vlc >> 3) + ((run == 0) ? 8 : ((run < 2) ? 2 : ((run < 5) ? 0 : -1)));
226 } else {
227 run = (vlc & 0xF);
228 level = (vlc >> 4) + ((run == 0) ? 4 : ((run < 3) ? 2 : ((run < 10) ? 1 : 0)));
229 }
230 }
231
232 if ((index += run) >= limit)
233 return -1;
234
235 block[scan[index]] = (level ^ sign) - sign;
236 }
237
238 if (type != 2) {
239 break;
240 }
241 }
242
243 return 0;
244}
245
246static void sixpel_mc_put (MpegEncContext *s,
247 uint8_t *src, uint8_t *dst, int stride,
248 int dxy, int width, int height) {
249 int i, j;
250
251 switch (dxy) {
252 case 6*0+0:
253 for (i=0; i < height; i++) {
254 memcpy (dst, src, width);
255 src += stride;
256 dst += stride;
257 }
258 break;
259 case 6*0+2:
260 for (i=0; i < height; i++) {
261 for (j=0; j < width; j++) {
262 dst[j] = (683*(2*src[j] + src[j+1] + 1)) >> 11;
263 }
264 src += stride;
265 dst += stride;
266 }
267 break;
268 case 6*0+3:
269 for (i=0; i < height; i++) {
270 for (j=0; j < width; j++) {
271 dst[j] = (src[j] + src[j+1] + 1) >> 1;
272 }
273 src += stride;
274 dst += stride;
275 }
276 break;
277 case 6*0+4:
278 for (i=0; i < height; i++) {
279 for (j=0; j < width; j++) {
280 dst[j] = (683*(src[j] + 2*src[j+1] + 1)) >> 11;
281 }
282 src += stride;
283 dst += stride;
284 }
285 break;
286 case 6*2+0:
287 for (i=0; i < height; i++) {
288 for (j=0; j < width; j++) {
289 dst[j] = (683*(2*src[j] + src[j+stride] + 1)) >> 11;
290 }
291 src += stride;
292 dst += stride;
293 }
294 break;
295 case 6*2+2:
296 for (i=0; i < height; i++) {
297 for (j=0; j < width; j++) {
298 dst[j] = (2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15;
299 }
300 src += stride;
301 dst += stride;
302 }
303 break;
304 case 6*2+4:
305 for (i=0; i < height; i++) {
306 for (j=0; j < width; j++) {
307 dst[j] = (2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
308 }
309 src += stride;
310 dst += stride;
311 }
312 break;
313 case 6*3+0:
314 for (i=0; i < height; i++) {
315 for (j=0; j < width; j++) {
316 dst[j] = (src[j] + src[j+stride]+1) >> 1;
317 }
318 src += stride;
319 dst += stride;
320 }
321 break;
322 case 6*3+3:
323 for (i=0; i < height; i++) {
324 for (j=0; j < width; j++) {
325 dst[j] = (src[j] + src[j+1] + src[j+stride] + src[j+stride+1] + 2) >> 2;
326 }
327 src += stride;
328 dst += stride;
329 }
330 break;
331 case 6*4+0:
332 for (i=0; i < height; i++) {
333 for (j=0; j < width; j++) {
334 dst[j] = (683*(src[j] + 2*src[j+stride] + 1)) >> 11;
335 }
336 src += stride;
337 dst += stride;
338 }
339 break;
340 case 6*4+2:
341 for (i=0; i < height; i++) {
342 for (j=0; j < width; j++) {
343 dst[j] = (2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
344 }
345 src += stride;
346 dst += stride;
347 }
348 break;
349 case 6*4+4:
350 for (i=0; i < height; i++) {
351 for (j=0; j < width; j++) {
352 dst[j] = (2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15;
353 }
354 src += stride;
355 dst += stride;
356 }
357 break;
358 }
359}
360
361static inline void svq3_mc_dir_part (MpegEncContext *s, int x, int y,
362 int width, int height, int mx, int my) {
363 uint8_t *src, *dest;
364 int i, emu = 0;
365 const int sx = ((unsigned) (mx + 0x7FFFFFFE)) % 6;
366 const int sy = ((unsigned) (my + 0x7FFFFFFE)) % 6;
367 const int dxy= 6*sy + sx;
368
369 /* decode and clip motion vector to frame border (+16) */
370 mx = x + (mx - sx) / 6;
371 my = y + (my - sy) / 6;
372
373 if (mx < 0 || mx >= (s->width - width - 1) ||
374 my < 0 || my >= (s->height - height - 1)) {
375
376 if ((s->flags & CODEC_FLAG_EMU_EDGE)) {
377 emu = 1;
378 }
379
380 mx = clip (mx, -16, (s->width - width + 15));
381 my = clip (my, -16, (s->height - height + 15));
382 }
383
384 /* form component predictions */
385 dest = s->current_picture.data[0] + x + y*s->linesize;
386 src = s->last_picture.data[0] + mx + my*s->linesize;
387
388 if (emu) {
389 ff_emulated_edge_mc (s, src, s->linesize, (width + 1), (height + 1),
390 mx, my, s->width, s->height);
391 src = s->edge_emu_buffer;
392 }
393 sixpel_mc_put (s, src, dest, s->linesize, dxy, width, height);
394
395 if (!(s->flags & CODEC_FLAG_GRAY)) {
396 mx = (mx + (mx < (int) x)) >> 1;
397 my = (my + (my < (int) y)) >> 1;
398 width = (width >> 1);
399 height = (height >> 1);
400
401 for (i=1; i < 3; i++) {
402 dest = s->current_picture.data[i] + (x >> 1) + (y >> 1)*s->uvlinesize;
403 src = s->last_picture.data[i] + mx + my*s->uvlinesize;
404
405 if (emu) {
406 ff_emulated_edge_mc (s, src, s->uvlinesize, (width + 1), (height + 1),
407 mx, my, (s->width >> 1), (s->height >> 1));
408 src = s->edge_emu_buffer;
409 }
410 sixpel_mc_put (s, src, dest, s->uvlinesize, dxy, width, height);
411 }
412 }
413}
414
415static int svq3_decode_mb (H264Context *h, unsigned int mb_type) {
416 int cbp, dir, mode, mx, my, dx, dy, x, y, part_width, part_height;
417 int i, j, k, l, m;
418 uint32_t vlc;
419 int8_t *top, *left;
420 MpegEncContext *const s = (MpegEncContext *) h;
421 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
422 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
423
424 h->top_samples_available = (s->mb_y == 0) ? 0x33FF : 0xFFFF;
425 h->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF;
426 h->topright_samples_available = 0xFFFF;
427
428 if (mb_type == 0) { /* SKIP */
429 svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0);
430
431 cbp = 0;
432 mb_type = MB_TYPE_SKIP;
433 } else if (mb_type < 8) { /* INTER */
434 if (h->thirdpel_flag && h->halfpel_flag == !get_bits (&s->gb, 1)) {
435 mode = 3; /* thirdpel */
436 } else if (h->halfpel_flag && h->thirdpel_flag == !get_bits (&s->gb, 1)) {
437 mode = 2; /* halfpel */
438 } else {
439 mode = 1; /* fullpel */
440 }
441
442 /* fill caches */
443 memset (h->ref_cache[0], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
444
445 if (s->mb_x > 0) {
446 for (i=0; i < 4; i++) {
447 *(uint32_t *) h->mv_cache[0][scan8[0] - 1 + i*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - 1 + i*h->b_stride];
448 h->ref_cache[0][scan8[0] - 1 + i*8] = 1;
449 }
450 } else {
451 for (i=0; i < 4; i++) {
452 *(uint32_t *) h->mv_cache[0][scan8[0] - 1 + i*8] = 0;
453 h->ref_cache[0][scan8[0] - 1 + i*8] = 1;
454 }
455 }
456 if (s->mb_y > 0) {
457 memcpy (h->mv_cache[0][scan8[0] - 1*8], s->current_picture.motion_val[0][b_xy - h->b_stride], 4*2*sizeof(int16_t));
458 memset (&h->ref_cache[0][scan8[0] - 1*8], 1, 4);
459
460 if (s->mb_x < (s->mb_width - 1)) {
461 *(uint32_t *) h->mv_cache[0][scan8[0] + 4 - 1*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - h->b_stride + 4];
462 h->ref_cache[0][scan8[0] + 4 - 1*8] = 1;
463 }
464 if (s->mb_x > 0) {
465 *(uint32_t *) h->mv_cache[0][scan8[0] - 1 - 1*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - h->b_stride - 1];
466 h->ref_cache[0][scan8[0] - 1 - 1*8] = 1;
467 }
468 }
469
470 /* decode motion vector(s) and form prediction(s) */
471 part_width = ((mb_type & 5) == 5) ? 4 : 8 << (mb_type & 1);
472 part_height = 16 >> ((unsigned) mb_type / 3);
473
474 for (i=0; i < 16; i+=part_height) {
475 for (j=0; j < 16; j+=part_width) {
476 x = 16*s->mb_x + j;
477 y = 16*s->mb_y + i;
478 k = ((j>>2)&1) + ((i>>1)&2) + ((j>>1)&4) + (i&8);
479
480 pred_motion (h, k, (part_width >> 2), 0, 1, &mx, &my);
481
482 /* clip motion vector prediction to frame border */
483 mx = clip (mx, -6*x, 6*(s->width - part_width - x));
484 my = clip (my, -6*y, 6*(s->height - part_height - y));
485
486 /* get motion vector differential */
487 dy = svq3_get_se_golomb (&s->gb);
488 dx = svq3_get_se_golomb (&s->gb);
489
490 if (dx == INVALID_VLC || dy == INVALID_VLC) {
491 return -1;
492 }
493
494 /* compute motion vector */
495 if (mode == 3) {
496 mx = ((mx + 1) & ~0x1) + 2*dx;
497 my = ((my + 1) & ~0x1) + 2*dy;
498 } else if (mode == 2) {
499 mx = (mx + 1) - ((unsigned) (0x7FFFFFFF + mx) % 3) + 3*dx;
500 my = (my + 1) - ((unsigned) (0x7FFFFFFF + my) % 3) + 3*dy;
501 } else if (mode == 1) {
502 mx = (mx + 3) - ((unsigned) (0x7FFFFFFB + mx) % 6) + 6*dx;
503 my = (my + 3) - ((unsigned) (0x7FFFFFFB + my) % 6) + 6*dy;
504 }
505
506 /* update mv_cache */
507 for (l=0; l < part_height; l+=4) {
508 for (m=0; m < part_width; m+=4) {
509 k = scan8[0] + ((m + j) >> 2) + ((l + i) << 1);
510 h->mv_cache [0][k][0] = mx;
511 h->mv_cache [0][k][1] = my;
512 h->ref_cache[0][k] = 1;
513 }
514 }
515
516 svq3_mc_dir_part (s, x, y, part_width, part_height, mx, my);
517 }
518 }
519
520 for (i=0; i < 4; i++) {
521 memcpy (s->current_picture.motion_val[0][b_xy + i*h->b_stride], h->mv_cache[0][scan8[0] + 8*i], 4*2*sizeof(int16_t));
522 }
523
524 if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48)
525 return -1;
526
527 cbp = golomb_to_inter_cbp[vlc];
528 mb_type = MB_TYPE_16x16;
529 } else if (mb_type == 8) { /* INTRA4x4 */
530 memset (h->intra4x4_pred_mode_cache, -1, 8*5*sizeof(int8_t));
531
532 if (s->mb_x > 0) {
533 for (i=0; i < 4; i++) {
534 h->intra4x4_pred_mode_cache[scan8[0] - 1 + i*8] = h->intra4x4_pred_mode[mb_xy - 1][i];
535 }
536 }
537 if (s->mb_y > 0) {
538 h->intra4x4_pred_mode_cache[4+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][4];
539 h->intra4x4_pred_mode_cache[5+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][5];
540 h->intra4x4_pred_mode_cache[6+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][6];
541 h->intra4x4_pred_mode_cache[7+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][3];
542 }
543
544 /* decode prediction codes for luma blocks */
545 for (i=0; i < 16; i+=2) {
546 vlc = svq3_get_ue_golomb (&s->gb);
547
548 if (vlc >= 25)
549 return -1;
550
551 left = &h->intra4x4_pred_mode_cache[scan8[i] - 1];
552 top = &h->intra4x4_pred_mode_cache[scan8[i] - 8];
553
554 left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]];
555 left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]];
556
557 if (left[1] == -1 || left[2] == -1)
558 return -1;
559 }
560
561 write_back_intra_pred_mode (h);
562 check_intra4x4_pred_mode (h);
563
564 if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48)
565 return -1;
566
567 cbp = golomb_to_intra4x4_cbp[vlc];
568 mb_type = MB_TYPE_INTRA4x4;
569 } else { /* INTRA16x16 */
570 dir = i_mb_type_info[mb_type - 8].pred_mode;
571 dir = (dir >> 1) ^ 3*(dir & 1) ^ 1;
572
573 if ((h->intra16x16_pred_mode = check_intra_pred_mode (h, dir)) == -1)
574 return -1;
575
576 cbp = i_mb_type_info[mb_type - 8].cbp;
577 mb_type = MB_TYPE_INTRA16x16;
578 }
579
580 if (!IS_INTER(mb_type) && s->pict_type != I_TYPE) {
581 for (i=0; i < 4; i++) {
582 memset (s->current_picture.motion_val[0][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
583 }
584 }
585 if (!IS_INTRA4x4(mb_type)) {
586 memset (h->intra4x4_pred_mode[mb_xy], DC_PRED, 8);
587 }
588 if (!IS_SKIP(mb_type)) {
589 memset (h->mb, 0, 24*16*sizeof(DCTELEM));
590 memset (h->non_zero_count_cache, 0, 8*6*sizeof(uint8_t));
591 }
592
593 if (IS_INTRA16x16(mb_type) || (s->pict_type != I_TYPE && s->adaptive_quant && cbp)) {
594 s->qscale += svq3_get_se_golomb (&s->gb);
595
596 if (s->qscale > 31)
597 return -1;
598 }
599 if (IS_INTRA16x16(mb_type)) {
600 if (svq3_decode_block (&s->gb, h->mb, 0, 0))
601 return -1;
602 }
603
604 if (!IS_SKIP(mb_type) && cbp) {
605 l = IS_INTRA16x16(mb_type) ? 1 : 0;
606 m = ((s->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1);
607
608 for (i=0; i < 4; i++) {
609 if ((cbp & (1 << i))) {
610 for (j=0; j < 4; j++) {
611 k = l ? ((j&1) + 2*(i&1) + 2*(j&2) + 4*(i&2)) : (4*i + j);
612 h->non_zero_count_cache[ scan8[k] ] = 1;
613
614 if (svq3_decode_block (&s->gb, &h->mb[16*k], l, m))
615 return -1;
616 }
617 }
618 }
619
620 if ((cbp & 0x30)) {
621 for (i=0; i < 2; ++i) {
622 if (svq3_decode_block (&s->gb, &h->mb[16*(16 + 4*i)], 0, 3))
623 return -1;
624 }
625
626 if ((cbp & 0x20)) {
627 for (i=0; i < 8; i++) {
628 h->non_zero_count_cache[ scan8[16+i] ] = 1;
629
630 if (svq3_decode_block (&s->gb, &h->mb[16*(16 + i)], 1, 1))
631 return -1;
632 }
633 }
634 }
635 }
636
637 s->current_picture.mb_type[mb_xy] = mb_type;
638
639 if (IS_INTRA(mb_type)) {
640 h->chroma_pred_mode = check_intra_pred_mode (h, DC_PRED8x8);
641 }
642
643 return 0;
644}
645
646static int svq3_decode_frame (AVCodecContext *avctx,
647 void *data, int *data_size,
648 uint8_t *buf, int buf_size) {
649 MpegEncContext *const s = avctx->priv_data;
650 H264Context *const h = avctx->priv_data;
651 int i;
652
653 s->flags = avctx->flags;
654
655 if (!s->context_initialized) {
656 s->width = (avctx->width + 15) & ~15;
657 s->height = (avctx->height + 15) & ~15;
658 h->b_stride = (s->width >> 2);
659 h->pred4x4[DIAG_DOWN_LEFT_PRED] = pred4x4_down_left_svq3_c;
660 h->pred16x16[PLANE_PRED8x8] = pred16x16_plane_svq3_c;
661 h->halfpel_flag = 1;
662 h->thirdpel_flag = 1;
663 h->chroma_qp = 4;
664
665 if (MPV_common_init (s) < 0)
666 return -1;
667
668 alloc_tables (h);
669 }
670 if (avctx->extradata && avctx->extradata_size >= 115
671 && !memcmp (avctx->extradata, "stsd", 4)) {
672
673 uint8_t *stsd = (uint8_t *) avctx->extradata + 114;
674
675 if ((*stsd >> 5) != 7 || avctx->extradata_size >= 118) {
676
677 if ((*stsd >> 5) == 7) {
678 stsd += 3; /* skip width, height (12 bits each) */
679 }
680
681 h->halfpel_flag = (*stsd >> 4) & 1;
682 h->thirdpel_flag = (*stsd >> 3) & 1;
683 }
684 }
685
686 if ((buf[0] & 0x9F) != 1) {
687 /* TODO: what? */
688 fprintf (stderr, "unsupported header (%02X)\n", buf[0]);
689 return -1;
690 } else {
691 int length = (buf[0] >> 5) & 3;
692 int offset = 0;
693
694 for (i=0; i < length; i++) {
695 offset = (offset << 8) | buf[i + 1];
696 }
697
698 if (buf_size < (offset + length + 1) || length == 0)
699 return -1;
700
701 memcpy (&buf[2], &buf[offset + 2], (length - 1));
702 }
703
704 init_get_bits (&s->gb, &buf[2], 8*(buf_size - 2));
705
706 if ((i = svq3_get_ue_golomb (&s->gb)) == INVALID_VLC || i >= 3)
707 return -1;
708
709 s->pict_type = golomb_to_pict_type[i];
710
711 /* unknown fields */
712 get_bits (&s->gb, 1);
713 get_bits (&s->gb, 8);
714
715 s->qscale = get_bits (&s->gb, 5);
716 s->adaptive_quant = get_bits (&s->gb, 1);
717
718 /* unknown fields */
719 get_bits (&s->gb, 1);
720 get_bits (&s->gb, 1);
721 get_bits (&s->gb, 2);
722
723 while (get_bits (&s->gb, 1)) {
724 get_bits (&s->gb, 8);
725 }
726
727 /* B-frames are not supported */
728 if (s->pict_type == B_TYPE/* && avctx->hurry_up*/)
729 return buf_size;
730
731 frame_start (h);
732
733 for (s->mb_y=0; s->mb_y < s->mb_height; s->mb_y++) {
734 for (s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
735 int mb_type = svq3_get_ue_golomb (&s->gb);
736
737 if (s->pict_type == I_TYPE) {
738 mb_type += 8;
739 }
740 if (mb_type > 32 || svq3_decode_mb (h, mb_type)) {
741 fprintf (stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
742 return -1;
743 }
744
745 if (mb_type != 0) {
746 hl_decode_mb (h);
747 }
748 }
749 }
750
751 *(AVFrame *) data = *(AVFrame *) &s->current_picture;
752 *data_size = sizeof(AVFrame);
753
754 MPV_frame_end(s);
755
756 return buf_size;
757}
758
759
760AVCodec svq3_decoder = {
761 "svq3",
762 CODEC_TYPE_VIDEO,
763 CODEC_ID_SVQ3,
764 sizeof(H264Context),
765 decode_init,
766 NULL,
767 decode_end,
768 svq3_decode_frame,
769 CODEC_CAP_DR1,
770};