Mercurial > libavcodec.hg
annotate svq3.c @ 1252:67ee8bab0f28 libavcodec
optimizations
author | michaelni |
---|---|
date | Tue, 13 May 2003 13:38:09 +0000 |
parents | fa181d095027 |
children | 2fa34e615c76 |
rev | line source |
---|---|
1234 | 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 | |
1235
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
21 * have stsd atoms to describe media trak properties. A stsd atom for a |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
22 * video trak contains 1 or more ImageDescription atoms. These atoms begin |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
23 * with the 4-byte length of the atom followed by the codec fourcc. Some |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
24 * decoders need information in this atom to operate correctly. Such |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
25 * is the case with SVQ3. In order to get the best use out of this decoder, |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
26 * the calling app must make the SVQ3 ImageDescription atom available |
1234 | 27 * via the AVCodecContext's extradata[_size] field: |
28 * | |
1235
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
29 * AVCodecContext.extradata = pointer to ImageDescription, first characters |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
30 * are expected to be 'S', 'V', 'Q', and '3', NOT the 4-byte atom length |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
31 * AVCodecContext.extradata_size = size of ImageDescription atom memory |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
32 * buffer (which will be the same as the ImageDescription atom size field |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
33 * from the QT file, minus 4 bytes since the length is missing) |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
34 * |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
35 * You will know you have these parameters passed correctly when the decoder |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
36 * correctly decodes this file: |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
37 * ftp://ftp.mplayerhq.hu/MPlayer/samples/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov |
1234 | 38 * |
39 */ | |
40 | |
41 /** | |
42 * @file svq3.c | |
43 * svq3 decoder. | |
44 */ | |
45 | |
1252 | 46 /* dual scan (from some older h264 draft) |
47 o-->o-->o o | |
48 | /| | |
49 o o o / o | |
50 | / | |/ | | |
51 o o o o | |
52 / | |
53 o-->o-->o-->o | |
54 */ | |
1234 | 55 static const uint8_t svq3_scan[16]={ |
56 0+0*4, 1+0*4, 2+0*4, 2+1*4, | |
57 2+2*4, 3+0*4, 3+1*4, 3+2*4, | |
58 0+1*4, 0+2*4, 1+1*4, 1+2*4, | |
59 0+3*4, 1+3*4, 2+3*4, 3+3*4, | |
60 }; | |
61 | |
62 static const uint8_t svq3_pred_0[25][2] = { | |
63 { 0, 0 }, | |
64 { 1, 0 }, { 0, 1 }, | |
65 { 0, 2 }, { 1, 1 }, { 2, 0 }, | |
66 { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 }, | |
67 { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 }, | |
68 { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 }, | |
69 { 2, 4 }, { 3, 3 }, { 4, 2 }, | |
70 { 4, 3 }, { 3, 4 }, | |
71 { 4, 4 } | |
72 }; | |
73 | |
74 static const int8_t svq3_pred_1[6][6][5] = { | |
75 { { 2,-1,-1,-1,-1 }, { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 }, | |
76 { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 }, { 1, 2,-1,-1,-1 } }, | |
77 { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 4, 3 }, { 0, 1, 2, 4, 3 }, | |
78 { 0, 2, 1, 4, 3 }, { 2, 0, 1, 3, 4 }, { 0, 4, 2, 1, 3 } }, | |
79 { { 2, 0,-1,-1,-1 }, { 2, 1, 0, 4, 3 }, { 1, 2, 4, 0, 3 }, | |
80 { 2, 1, 0, 4, 3 }, { 2, 1, 4, 3, 0 }, { 1, 2, 4, 0, 3 } }, | |
81 { { 2, 0,-1,-1,-1 }, { 2, 0, 1, 4, 3 }, { 1, 2, 0, 4, 3 }, | |
82 { 2, 1, 0, 4, 3 }, { 2, 1, 3, 4, 0 }, { 2, 4, 1, 0, 3 } }, | |
83 { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 3, 4 }, { 1, 2, 3, 0, 4 }, | |
84 { 2, 0, 1, 3, 4 }, { 2, 1, 3, 0, 4 }, { 2, 0, 4, 3, 1 } }, | |
85 { { 0, 2,-1,-1,-1 }, { 0, 2, 4, 1, 3 }, { 1, 4, 2, 0, 3 }, | |
86 { 4, 2, 0, 1, 3 }, { 2, 0, 1, 4, 3 }, { 4, 2, 1, 0, 3 } }, | |
87 }; | |
88 | |
89 static const struct { uint8_t run; uint8_t level; } svq3_dct_tables[2][16] = { | |
90 { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 }, | |
91 { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } }, | |
92 { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 }, | |
93 { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } } | |
94 }; | |
95 | |
96 static const uint32_t svq3_dequant_coeff[32] = { | |
97 3881, 4351, 4890, 5481, 6154, 6914, 7761, 8718, | |
98 9781, 10987, 12339, 13828, 15523, 17435, 19561, 21873, | |
99 24552, 27656, 30847, 34870, 38807, 43747, 49103, 54683, | |
100 61694, 68745, 77615, 89113,100253,109366,126635,141533 | |
101 }; | |
102 | |
103 | |
104 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp){ | |
105 const int qmul= svq3_dequant_coeff[qp]; | |
106 #define stride 16 | |
107 int i; | |
108 int temp[16]; | |
109 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
110 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
111 | |
112 for(i=0; i<4; i++){ | |
113 const int offset= y_offset[i]; | |
114 const int z0= 13*(block[offset+stride*0] + block[offset+stride*4]); | |
115 const int z1= 13*(block[offset+stride*0] - block[offset+stride*4]); | |
116 const int z2= 7* block[offset+stride*1] - 17*block[offset+stride*5]; | |
117 const int z3= 17* block[offset+stride*1] + 7*block[offset+stride*5]; | |
118 | |
119 temp[4*i+0]= z0+z3; | |
120 temp[4*i+1]= z1+z2; | |
121 temp[4*i+2]= z1-z2; | |
122 temp[4*i+3]= z0-z3; | |
123 } | |
124 | |
125 for(i=0; i<4; i++){ | |
126 const int offset= x_offset[i]; | |
127 const int z0= 13*(temp[4*0+i] + temp[4*2+i]); | |
128 const int z1= 13*(temp[4*0+i] - temp[4*2+i]); | |
129 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i]; | |
130 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i]; | |
131 | |
132 block[stride*0 +offset]= ((z0 + z3)*qmul + 0x80000)>>20; | |
133 block[stride*2 +offset]= ((z1 + z2)*qmul + 0x80000)>>20; | |
134 block[stride*8 +offset]= ((z1 - z2)*qmul + 0x80000)>>20; | |
135 block[stride*10+offset]= ((z0 - z3)*qmul + 0x80000)>>20; | |
136 } | |
137 } | |
138 #undef stride | |
139 | |
140 static void svq3_add_idct_c (uint8_t *dst, DCTELEM *block, int stride, int qp, int dc){ | |
141 const int qmul= svq3_dequant_coeff[qp]; | |
142 int i; | |
143 uint8_t *cm = cropTbl + MAX_NEG_CROP; | |
144 | |
145 if (dc) { | |
146 dc = 13*13*((dc == 1) ? 1538*block[0] : ((qmul*(block[0] >> 3)) / 2)); | |
147 block[0] = 0; | |
148 } | |
149 | |
150 for (i=0; i < 4; i++) { | |
151 const int z0= 13*(block[0 + 4*i] + block[2 + 4*i]); | |
152 const int z1= 13*(block[0 + 4*i] - block[2 + 4*i]); | |
153 const int z2= 7* block[1 + 4*i] - 17*block[3 + 4*i]; | |
154 const int z3= 17* block[1 + 4*i] + 7*block[3 + 4*i]; | |
155 | |
156 block[0 + 4*i]= z0 + z3; | |
157 block[1 + 4*i]= z1 + z2; | |
158 block[2 + 4*i]= z1 - z2; | |
159 block[3 + 4*i]= z0 - z3; | |
160 } | |
161 | |
162 for (i=0; i < 4; i++) { | |
163 const int z0= 13*(block[i + 4*0] + block[i + 4*2]); | |
164 const int z1= 13*(block[i + 4*0] - block[i + 4*2]); | |
165 const int z2= 7* block[i + 4*1] - 17*block[i + 4*3]; | |
166 const int z3= 17* block[i + 4*1] + 7*block[i + 4*3]; | |
167 const int rr= (dc + 0x80000); | |
168 | |
169 dst[i + stride*0]= cm[ dst[i + stride*0] + (((z0 + z3)*qmul + rr) >> 20) ]; | |
170 dst[i + stride*1]= cm[ dst[i + stride*1] + (((z1 + z2)*qmul + rr) >> 20) ]; | |
171 dst[i + stride*2]= cm[ dst[i + stride*2] + (((z1 - z2)*qmul + rr) >> 20) ]; | |
172 dst[i + stride*3]= cm[ dst[i + stride*3] + (((z0 - z3)*qmul + rr) >> 20) ]; | |
173 } | |
174 } | |
175 | |
176 static void pred4x4_down_left_svq3_c(uint8_t *src, uint8_t *topright, int stride){ | |
177 LOAD_TOP_EDGE | |
178 LOAD_LEFT_EDGE | |
179 const __attribute__((unused)) int unu0= t0; | |
180 const __attribute__((unused)) int unu1= l0; | |
181 | |
182 src[0+0*stride]=(l1 + t1)>>1; | |
183 src[1+0*stride]= | |
184 src[0+1*stride]=(l2 + t2)>>1; | |
185 src[2+0*stride]= | |
186 src[1+1*stride]= | |
187 src[0+2*stride]= | |
188 src[3+0*stride]= | |
189 src[2+1*stride]= | |
190 src[1+2*stride]= | |
191 src[0+3*stride]= | |
192 src[3+1*stride]= | |
193 src[2+2*stride]= | |
194 src[1+3*stride]= | |
195 src[3+2*stride]= | |
196 src[2+3*stride]= | |
197 src[3+3*stride]=(l3 + t3)>>1; | |
198 }; | |
199 | |
200 static void pred16x16_plane_svq3_c(uint8_t *src, int stride){ | |
201 pred16x16_plane_compat_c(src, stride, 1); | |
202 } | |
203 | |
204 static inline int svq3_decode_block (GetBitContext *gb, DCTELEM *block, | |
205 int index, const int type) { | |
206 | |
207 static const uint8_t *const scan_patterns[4] = | |
208 { luma_dc_zigzag_scan, zigzag_scan, svq3_scan, chroma_dc_scan }; | |
209 | |
210 int run, level, sign, vlc, limit; | |
211 const int intra = (3 * type) >> 2; | |
212 const uint8_t *const scan = scan_patterns[type]; | |
213 | |
214 for (limit=(16 >> intra); index < 16; index=limit, limit+=8) { | |
215 for (; (vlc = svq3_get_ue_golomb (gb)) != 0; index++) { | |
216 | |
217 if (vlc == INVALID_VLC) | |
218 return -1; | |
219 | |
220 sign = (vlc & 0x1) - 1; | |
221 vlc = (vlc + 1) >> 1; | |
222 | |
223 if (type == 3) { | |
224 if (vlc < 3) { | |
225 run = 0; | |
226 level = vlc; | |
227 } else if (vlc < 4) { | |
228 run = 1; | |
229 level = 1; | |
230 } else { | |
231 run = (vlc & 0x3); | |
232 level = ((vlc + 9) >> 2) - run; | |
233 } | |
234 } else { | |
235 if (vlc < 16) { | |
236 run = svq3_dct_tables[intra][vlc].run; | |
237 level = svq3_dct_tables[intra][vlc].level; | |
238 } else if (intra) { | |
239 run = (vlc & 0x7); | |
240 level = (vlc >> 3) + ((run == 0) ? 8 : ((run < 2) ? 2 : ((run < 5) ? 0 : -1))); | |
241 } else { | |
242 run = (vlc & 0xF); | |
243 level = (vlc >> 4) + ((run == 0) ? 4 : ((run < 3) ? 2 : ((run < 10) ? 1 : 0))); | |
244 } | |
245 } | |
246 | |
247 if ((index += run) >= limit) | |
248 return -1; | |
249 | |
250 block[scan[index]] = (level ^ sign) - sign; | |
251 } | |
252 | |
253 if (type != 2) { | |
254 break; | |
255 } | |
256 } | |
257 | |
258 return 0; | |
259 } | |
260 | |
261 static void sixpel_mc_put (MpegEncContext *s, | |
262 uint8_t *src, uint8_t *dst, int stride, | |
263 int dxy, int width, int height) { | |
264 int i, j; | |
265 | |
266 switch (dxy) { | |
267 case 6*0+0: | |
268 for (i=0; i < height; i++) { | |
269 memcpy (dst, src, width); | |
270 src += stride; | |
271 dst += stride; | |
272 } | |
273 break; | |
274 case 6*0+2: | |
275 for (i=0; i < height; i++) { | |
276 for (j=0; j < width; j++) { | |
277 dst[j] = (683*(2*src[j] + src[j+1] + 1)) >> 11; | |
278 } | |
279 src += stride; | |
280 dst += stride; | |
281 } | |
282 break; | |
283 case 6*0+3: | |
284 for (i=0; i < height; i++) { | |
285 for (j=0; j < width; j++) { | |
286 dst[j] = (src[j] + src[j+1] + 1) >> 1; | |
287 } | |
288 src += stride; | |
289 dst += stride; | |
290 } | |
291 break; | |
292 case 6*0+4: | |
293 for (i=0; i < height; i++) { | |
294 for (j=0; j < width; j++) { | |
295 dst[j] = (683*(src[j] + 2*src[j+1] + 1)) >> 11; | |
296 } | |
297 src += stride; | |
298 dst += stride; | |
299 } | |
300 break; | |
301 case 6*2+0: | |
302 for (i=0; i < height; i++) { | |
303 for (j=0; j < width; j++) { | |
304 dst[j] = (683*(2*src[j] + src[j+stride] + 1)) >> 11; | |
305 } | |
306 src += stride; | |
307 dst += stride; | |
308 } | |
309 break; | |
310 case 6*2+2: | |
311 for (i=0; i < height; i++) { | |
312 for (j=0; j < width; j++) { | |
313 dst[j] = (2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15; | |
314 } | |
315 src += stride; | |
316 dst += stride; | |
317 } | |
318 break; | |
319 case 6*2+4: | |
320 for (i=0; i < height; i++) { | |
321 for (j=0; j < width; j++) { | |
322 dst[j] = (2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15; | |
323 } | |
324 src += stride; | |
325 dst += stride; | |
326 } | |
327 break; | |
328 case 6*3+0: | |
329 for (i=0; i < height; i++) { | |
330 for (j=0; j < width; j++) { | |
331 dst[j] = (src[j] + src[j+stride]+1) >> 1; | |
332 } | |
333 src += stride; | |
334 dst += stride; | |
335 } | |
336 break; | |
337 case 6*3+3: | |
338 for (i=0; i < height; i++) { | |
339 for (j=0; j < width; j++) { | |
340 dst[j] = (src[j] + src[j+1] + src[j+stride] + src[j+stride+1] + 2) >> 2; | |
341 } | |
342 src += stride; | |
343 dst += stride; | |
344 } | |
345 break; | |
346 case 6*4+0: | |
347 for (i=0; i < height; i++) { | |
348 for (j=0; j < width; j++) { | |
349 dst[j] = (683*(src[j] + 2*src[j+stride] + 1)) >> 11; | |
350 } | |
351 src += stride; | |
352 dst += stride; | |
353 } | |
354 break; | |
355 case 6*4+2: | |
356 for (i=0; i < height; i++) { | |
357 for (j=0; j < width; j++) { | |
358 dst[j] = (2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15; | |
359 } | |
360 src += stride; | |
361 dst += stride; | |
362 } | |
363 break; | |
364 case 6*4+4: | |
365 for (i=0; i < height; i++) { | |
366 for (j=0; j < width; j++) { | |
367 dst[j] = (2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15; | |
368 } | |
369 src += stride; | |
370 dst += stride; | |
371 } | |
372 break; | |
373 } | |
374 } | |
375 | |
376 static inline void svq3_mc_dir_part (MpegEncContext *s, int x, int y, | |
377 int width, int height, int mx, int my) { | |
378 uint8_t *src, *dest; | |
379 int i, emu = 0; | |
380 const int sx = ((unsigned) (mx + 0x7FFFFFFE)) % 6; | |
381 const int sy = ((unsigned) (my + 0x7FFFFFFE)) % 6; | |
382 const int dxy= 6*sy + sx; | |
383 | |
384 /* decode and clip motion vector to frame border (+16) */ | |
385 mx = x + (mx - sx) / 6; | |
386 my = y + (my - sy) / 6; | |
387 | |
388 if (mx < 0 || mx >= (s->width - width - 1) || | |
389 my < 0 || my >= (s->height - height - 1)) { | |
390 | |
391 if ((s->flags & CODEC_FLAG_EMU_EDGE)) { | |
392 emu = 1; | |
393 } | |
394 | |
395 mx = clip (mx, -16, (s->width - width + 15)); | |
396 my = clip (my, -16, (s->height - height + 15)); | |
397 } | |
398 | |
399 /* form component predictions */ | |
400 dest = s->current_picture.data[0] + x + y*s->linesize; | |
401 src = s->last_picture.data[0] + mx + my*s->linesize; | |
402 | |
403 if (emu) { | |
404 ff_emulated_edge_mc (s, src, s->linesize, (width + 1), (height + 1), | |
405 mx, my, s->width, s->height); | |
406 src = s->edge_emu_buffer; | |
407 } | |
408 sixpel_mc_put (s, src, dest, s->linesize, dxy, width, height); | |
409 | |
410 if (!(s->flags & CODEC_FLAG_GRAY)) { | |
411 mx = (mx + (mx < (int) x)) >> 1; | |
412 my = (my + (my < (int) y)) >> 1; | |
413 width = (width >> 1); | |
414 height = (height >> 1); | |
415 | |
416 for (i=1; i < 3; i++) { | |
417 dest = s->current_picture.data[i] + (x >> 1) + (y >> 1)*s->uvlinesize; | |
418 src = s->last_picture.data[i] + mx + my*s->uvlinesize; | |
419 | |
420 if (emu) { | |
421 ff_emulated_edge_mc (s, src, s->uvlinesize, (width + 1), (height + 1), | |
422 mx, my, (s->width >> 1), (s->height >> 1)); | |
423 src = s->edge_emu_buffer; | |
424 } | |
425 sixpel_mc_put (s, src, dest, s->uvlinesize, dxy, width, height); | |
426 } | |
427 } | |
428 } | |
429 | |
430 static int svq3_decode_mb (H264Context *h, unsigned int mb_type) { | |
431 int cbp, dir, mode, mx, my, dx, dy, x, y, part_width, part_height; | |
432 int i, j, k, l, m; | |
433 uint32_t vlc; | |
434 int8_t *top, *left; | |
435 MpegEncContext *const s = (MpegEncContext *) h; | |
436 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride; | |
437 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
438 | |
439 h->top_samples_available = (s->mb_y == 0) ? 0x33FF : 0xFFFF; | |
440 h->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF; | |
441 h->topright_samples_available = 0xFFFF; | |
442 | |
443 if (mb_type == 0) { /* SKIP */ | |
444 svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0); | |
445 | |
446 cbp = 0; | |
447 mb_type = MB_TYPE_SKIP; | |
448 } else if (mb_type < 8) { /* INTER */ | |
449 if (h->thirdpel_flag && h->halfpel_flag == !get_bits (&s->gb, 1)) { | |
450 mode = 3; /* thirdpel */ | |
451 } else if (h->halfpel_flag && h->thirdpel_flag == !get_bits (&s->gb, 1)) { | |
452 mode = 2; /* halfpel */ | |
453 } else { | |
454 mode = 1; /* fullpel */ | |
455 } | |
456 | |
457 /* fill caches */ | |
1252 | 458 /* note ref_cache[0] should contain here: |
459 ???????? | |
460 ???11111 | |
461 N??11111 | |
462 N??11111 | |
463 N??11111 | |
464 N | |
465 */ | |
466 | |
1234 | 467 if (s->mb_x > 0) { |
468 for (i=0; i < 4; i++) { | |
469 *(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]; | |
470 } | |
471 } else { | |
472 for (i=0; i < 4; i++) { | |
473 *(uint32_t *) h->mv_cache[0][scan8[0] - 1 + i*8] = 0; | |
474 } | |
475 } | |
476 if (s->mb_y > 0) { | |
477 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)); | |
478 memset (&h->ref_cache[0][scan8[0] - 1*8], 1, 4); | |
479 | |
480 if (s->mb_x < (s->mb_width - 1)) { | |
481 *(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]; | |
482 h->ref_cache[0][scan8[0] + 4 - 1*8] = 1; | |
1252 | 483 }else |
484 h->ref_cache[0][scan8[0] + 4 - 1*8] = PART_NOT_AVAILABLE; | |
1234 | 485 if (s->mb_x > 0) { |
486 *(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]; | |
487 h->ref_cache[0][scan8[0] - 1 - 1*8] = 1; | |
1252 | 488 }else |
489 h->ref_cache[0][scan8[0] - 1 - 1*8] = PART_NOT_AVAILABLE; | |
490 }else | |
491 memset (&h->ref_cache[0][scan8[0] - 1*8 - 1], PART_NOT_AVAILABLE, 8); | |
1234 | 492 |
493 /* decode motion vector(s) and form prediction(s) */ | |
494 part_width = ((mb_type & 5) == 5) ? 4 : 8 << (mb_type & 1); | |
495 part_height = 16 >> ((unsigned) mb_type / 3); | |
496 | |
497 for (i=0; i < 16; i+=part_height) { | |
498 for (j=0; j < 16; j+=part_width) { | |
499 x = 16*s->mb_x + j; | |
500 y = 16*s->mb_y + i; | |
501 k = ((j>>2)&1) + ((i>>1)&2) + ((j>>1)&4) + (i&8); | |
502 | |
503 pred_motion (h, k, (part_width >> 2), 0, 1, &mx, &my); | |
504 | |
505 /* clip motion vector prediction to frame border */ | |
506 mx = clip (mx, -6*x, 6*(s->width - part_width - x)); | |
507 my = clip (my, -6*y, 6*(s->height - part_height - y)); | |
508 | |
509 /* get motion vector differential */ | |
510 dy = svq3_get_se_golomb (&s->gb); | |
511 dx = svq3_get_se_golomb (&s->gb); | |
512 | |
513 if (dx == INVALID_VLC || dy == INVALID_VLC) { | |
514 return -1; | |
515 } | |
516 | |
517 /* compute motion vector */ | |
518 if (mode == 3) { | |
519 mx = ((mx + 1) & ~0x1) + 2*dx; | |
520 my = ((my + 1) & ~0x1) + 2*dy; | |
521 } else if (mode == 2) { | |
522 mx = (mx + 1) - ((unsigned) (0x7FFFFFFF + mx) % 3) + 3*dx; | |
523 my = (my + 1) - ((unsigned) (0x7FFFFFFF + my) % 3) + 3*dy; | |
524 } else if (mode == 1) { | |
525 mx = (mx + 3) - ((unsigned) (0x7FFFFFFB + mx) % 6) + 6*dx; | |
526 my = (my + 3) - ((unsigned) (0x7FFFFFFB + my) % 6) + 6*dy; | |
527 } | |
528 | |
529 /* update mv_cache */ | |
1252 | 530 fill_rectangle(h->mv_cache[0][scan8[k]], part_width>>2, part_height>>2, 8, (mx&0xFFFF)+(my<<16), 4); |
1234 | 531 |
532 svq3_mc_dir_part (s, x, y, part_width, part_height, mx, my); | |
533 } | |
534 } | |
535 | |
536 for (i=0; i < 4; i++) { | |
537 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)); | |
538 } | |
539 | |
540 if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48) | |
541 return -1; | |
542 | |
543 cbp = golomb_to_inter_cbp[vlc]; | |
544 mb_type = MB_TYPE_16x16; | |
545 } else if (mb_type == 8) { /* INTRA4x4 */ | |
546 memset (h->intra4x4_pred_mode_cache, -1, 8*5*sizeof(int8_t)); | |
547 | |
548 if (s->mb_x > 0) { | |
549 for (i=0; i < 4; i++) { | |
550 h->intra4x4_pred_mode_cache[scan8[0] - 1 + i*8] = h->intra4x4_pred_mode[mb_xy - 1][i]; | |
551 } | |
552 } | |
553 if (s->mb_y > 0) { | |
554 h->intra4x4_pred_mode_cache[4+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][4]; | |
555 h->intra4x4_pred_mode_cache[5+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][5]; | |
556 h->intra4x4_pred_mode_cache[6+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][6]; | |
557 h->intra4x4_pred_mode_cache[7+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][3]; | |
558 } | |
559 | |
560 /* decode prediction codes for luma blocks */ | |
561 for (i=0; i < 16; i+=2) { | |
562 vlc = svq3_get_ue_golomb (&s->gb); | |
563 | |
564 if (vlc >= 25) | |
565 return -1; | |
566 | |
567 left = &h->intra4x4_pred_mode_cache[scan8[i] - 1]; | |
568 top = &h->intra4x4_pred_mode_cache[scan8[i] - 8]; | |
569 | |
570 left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]]; | |
571 left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]]; | |
572 | |
573 if (left[1] == -1 || left[2] == -1) | |
574 return -1; | |
575 } | |
576 | |
577 write_back_intra_pred_mode (h); | |
578 check_intra4x4_pred_mode (h); | |
579 | |
580 if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48) | |
581 return -1; | |
582 | |
583 cbp = golomb_to_intra4x4_cbp[vlc]; | |
584 mb_type = MB_TYPE_INTRA4x4; | |
585 } else { /* INTRA16x16 */ | |
586 dir = i_mb_type_info[mb_type - 8].pred_mode; | |
587 dir = (dir >> 1) ^ 3*(dir & 1) ^ 1; | |
588 | |
589 if ((h->intra16x16_pred_mode = check_intra_pred_mode (h, dir)) == -1) | |
590 return -1; | |
591 | |
592 cbp = i_mb_type_info[mb_type - 8].cbp; | |
593 mb_type = MB_TYPE_INTRA16x16; | |
594 } | |
595 | |
596 if (!IS_INTER(mb_type) && s->pict_type != I_TYPE) { | |
597 for (i=0; i < 4; i++) { | |
598 memset (s->current_picture.motion_val[0][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t)); | |
599 } | |
600 } | |
601 if (!IS_INTRA4x4(mb_type)) { | |
602 memset (h->intra4x4_pred_mode[mb_xy], DC_PRED, 8); | |
603 } | |
604 if (!IS_SKIP(mb_type)) { | |
1252 | 605 memset (h->non_zero_count_cache + 8, 0, 4*9*sizeof(uint8_t)); |
606 s->dsp.clear_blocks(h->mb); | |
1234 | 607 } |
608 | |
609 if (IS_INTRA16x16(mb_type) || (s->pict_type != I_TYPE && s->adaptive_quant && cbp)) { | |
610 s->qscale += svq3_get_se_golomb (&s->gb); | |
611 | |
612 if (s->qscale > 31) | |
613 return -1; | |
614 } | |
615 if (IS_INTRA16x16(mb_type)) { | |
616 if (svq3_decode_block (&s->gb, h->mb, 0, 0)) | |
617 return -1; | |
618 } | |
619 | |
620 if (!IS_SKIP(mb_type) && cbp) { | |
621 l = IS_INTRA16x16(mb_type) ? 1 : 0; | |
622 m = ((s->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1); | |
623 | |
624 for (i=0; i < 4; i++) { | |
625 if ((cbp & (1 << i))) { | |
626 for (j=0; j < 4; j++) { | |
627 k = l ? ((j&1) + 2*(i&1) + 2*(j&2) + 4*(i&2)) : (4*i + j); | |
628 h->non_zero_count_cache[ scan8[k] ] = 1; | |
629 | |
630 if (svq3_decode_block (&s->gb, &h->mb[16*k], l, m)) | |
631 return -1; | |
632 } | |
633 } | |
634 } | |
635 | |
636 if ((cbp & 0x30)) { | |
637 for (i=0; i < 2; ++i) { | |
638 if (svq3_decode_block (&s->gb, &h->mb[16*(16 + 4*i)], 0, 3)) | |
639 return -1; | |
640 } | |
641 | |
642 if ((cbp & 0x20)) { | |
643 for (i=0; i < 8; i++) { | |
644 h->non_zero_count_cache[ scan8[16+i] ] = 1; | |
645 | |
646 if (svq3_decode_block (&s->gb, &h->mb[16*(16 + i)], 1, 1)) | |
647 return -1; | |
648 } | |
649 } | |
650 } | |
651 } | |
652 | |
653 s->current_picture.mb_type[mb_xy] = mb_type; | |
654 | |
655 if (IS_INTRA(mb_type)) { | |
656 h->chroma_pred_mode = check_intra_pred_mode (h, DC_PRED8x8); | |
657 } | |
658 | |
659 return 0; | |
660 } | |
661 | |
662 static int svq3_decode_frame (AVCodecContext *avctx, | |
663 void *data, int *data_size, | |
664 uint8_t *buf, int buf_size) { | |
665 MpegEncContext *const s = avctx->priv_data; | |
666 H264Context *const h = avctx->priv_data; | |
667 int i; | |
668 | |
669 s->flags = avctx->flags; | |
670 | |
671 if (!s->context_initialized) { | |
672 s->width = (avctx->width + 15) & ~15; | |
673 s->height = (avctx->height + 15) & ~15; | |
674 h->b_stride = (s->width >> 2); | |
675 h->pred4x4[DIAG_DOWN_LEFT_PRED] = pred4x4_down_left_svq3_c; | |
676 h->pred16x16[PLANE_PRED8x8] = pred16x16_plane_svq3_c; | |
677 h->halfpel_flag = 1; | |
678 h->thirdpel_flag = 1; | |
679 h->chroma_qp = 4; | |
680 | |
681 if (MPV_common_init (s) < 0) | |
682 return -1; | |
683 | |
684 alloc_tables (h); | |
685 } | |
1235
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
686 if (avctx->extradata && avctx->extradata_size >= 0x63 |
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
687 && !memcmp (avctx->extradata, "SVQ3", 4)) { |
1234 | 688 |
1235
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
689 uint8_t *stsd = (uint8_t *) avctx->extradata + 0x62; |
1234 | 690 |
1235
8bb75c3c2f21
change the way the ImageDescription is passed to the decoder
tmmm
parents:
1234
diff
changeset
|
691 if ((*stsd >> 5) != 7 || avctx->extradata_size >= 0x66) { |
1234 | 692 |
693 if ((*stsd >> 5) == 7) { | |
694 stsd += 3; /* skip width, height (12 bits each) */ | |
695 } | |
696 | |
697 h->halfpel_flag = (*stsd >> 4) & 1; | |
698 h->thirdpel_flag = (*stsd >> 3) & 1; | |
699 } | |
700 } | |
701 | |
702 if ((buf[0] & 0x9F) != 1) { | |
703 /* TODO: what? */ | |
704 fprintf (stderr, "unsupported header (%02X)\n", buf[0]); | |
705 return -1; | |
706 } else { | |
707 int length = (buf[0] >> 5) & 3; | |
708 int offset = 0; | |
709 | |
710 for (i=0; i < length; i++) { | |
711 offset = (offset << 8) | buf[i + 1]; | |
712 } | |
713 | |
714 if (buf_size < (offset + length + 1) || length == 0) | |
715 return -1; | |
716 | |
717 memcpy (&buf[2], &buf[offset + 2], (length - 1)); | |
718 } | |
719 | |
720 init_get_bits (&s->gb, &buf[2], 8*(buf_size - 2)); | |
721 | |
722 if ((i = svq3_get_ue_golomb (&s->gb)) == INVALID_VLC || i >= 3) | |
723 return -1; | |
724 | |
725 s->pict_type = golomb_to_pict_type[i]; | |
726 | |
727 /* unknown fields */ | |
728 get_bits (&s->gb, 1); | |
729 get_bits (&s->gb, 8); | |
730 | |
731 s->qscale = get_bits (&s->gb, 5); | |
732 s->adaptive_quant = get_bits (&s->gb, 1); | |
733 | |
734 /* unknown fields */ | |
735 get_bits (&s->gb, 1); | |
736 get_bits (&s->gb, 1); | |
737 get_bits (&s->gb, 2); | |
738 | |
739 while (get_bits (&s->gb, 1)) { | |
740 get_bits (&s->gb, 8); | |
741 } | |
1250 | 742 |
743 if(avctx->debug&FF_DEBUG_PICT_INFO){ | |
744 printf("%c hpel:%d, tpel:%d aqp:%d qp:%d\n", | |
745 ff_get_pict_type_char(s->pict_type), h->halfpel_flag, h->thirdpel_flag, | |
746 s->adaptive_quant, s->qscale | |
747 ); | |
748 } | |
1234 | 749 |
750 /* B-frames are not supported */ | |
751 if (s->pict_type == B_TYPE/* && avctx->hurry_up*/) | |
752 return buf_size; | |
753 | |
754 frame_start (h); | |
755 | |
1252 | 756 for(i=0; i<4; i++){ |
757 int j; | |
758 for(j=-1; j<4; j++) | |
759 h->ref_cache[0][scan8[0] + 8*i + j]= 1; | |
760 h->ref_cache[0][scan8[0] + 8*i + j]= PART_NOT_AVAILABLE; | |
761 } | |
762 | |
1234 | 763 for (s->mb_y=0; s->mb_y < s->mb_height; s->mb_y++) { |
764 for (s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) { | |
765 int mb_type = svq3_get_ue_golomb (&s->gb); | |
766 | |
767 if (s->pict_type == I_TYPE) { | |
768 mb_type += 8; | |
769 } | |
770 if (mb_type > 32 || svq3_decode_mb (h, mb_type)) { | |
771 fprintf (stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); | |
772 return -1; | |
773 } | |
774 | |
775 if (mb_type != 0) { | |
776 hl_decode_mb (h); | |
777 } | |
778 } | |
779 } | |
780 | |
781 *(AVFrame *) data = *(AVFrame *) &s->current_picture; | |
782 *data_size = sizeof(AVFrame); | |
783 | |
784 MPV_frame_end(s); | |
785 | |
786 return buf_size; | |
787 } | |
788 | |
789 | |
790 AVCodec svq3_decoder = { | |
791 "svq3", | |
792 CODEC_TYPE_VIDEO, | |
793 CODEC_ID_SVQ3, | |
794 sizeof(H264Context), | |
795 decode_init, | |
796 NULL, | |
797 decode_end, | |
798 svq3_decode_frame, | |
799 CODEC_CAP_DR1, | |
800 }; |