Mercurial > libavcodec.hg
annotate common.c @ 212:80edc2aaff86 libavcodec
Added reverting support for -pp and -npp options
author | albeu |
---|---|
date | Sat, 19 Jan 2002 17:01:03 +0000 |
parents | b691dd3e9088 |
children | 5b88ee1abf97 |
rev | line source |
---|---|
0 | 1 /* |
2 * Common bit i/o utils | |
3 * Copyright (c) 2000, 2001 Gerard Lantau. | |
4 * | |
5 * This program is free software; you can redistribute it and/or modify | |
6 * it under the terms of the GNU General Public License as published by | |
7 * the Free Software Foundation; either version 2 of the License, or | |
8 * (at your option) any later version. | |
9 * | |
10 * This program is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 * GNU General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU General Public License | |
16 * along with this program; if not, write to the Free Software | |
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
18 * |
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
19 * alternative bitstream reader by Michael Niedermayer <michaelni@gmx.at> |
0 | 20 */ |
64 | 21 #include "common.h" |
0 | 22 #include <math.h> |
8 | 23 |
0 | 24 void init_put_bits(PutBitContext *s, |
25 UINT8 *buffer, int buffer_size, | |
26 void *opaque, | |
27 void (*write_data)(void *, UINT8 *, int)) | |
28 { | |
29 s->buf = buffer; | |
30 s->buf_ptr = s->buf; | |
31 s->buf_end = s->buf + buffer_size; | |
32 s->bit_cnt=0; | |
33 s->bit_buf=0; | |
34 s->data_out_size = 0; | |
35 s->write_data = write_data; | |
36 s->opaque = opaque; | |
37 } | |
38 | |
39 static void flush_buffer(PutBitContext *s) | |
40 { | |
41 int size; | |
42 if (s->write_data) { | |
43 size = s->buf_ptr - s->buf; | |
44 if (size > 0) | |
45 s->write_data(s->opaque, s->buf, size); | |
46 s->buf_ptr = s->buf; | |
47 s->data_out_size += size; | |
48 } | |
49 } | |
50 | |
51 void put_bits(PutBitContext *s, int n, unsigned int value) | |
52 { | |
53 unsigned int bit_buf; | |
54 int bit_cnt; | |
55 | |
56 #ifdef STATS | |
57 st_out_bit_counts[st_current_index] += n; | |
58 #endif | |
59 // printf("put_bits=%d %x\n", n, value); | |
60 assert(n == 32 || value < (1U << n)); | |
61 | |
62 bit_buf = s->bit_buf; | |
63 bit_cnt = s->bit_cnt; | |
64 | |
65 // printf("n=%d value=%x cnt=%d buf=%x\n", n, value, bit_cnt, bit_buf); | |
66 /* XXX: optimize */ | |
67 if (n < (32-bit_cnt)) { | |
68 bit_buf |= value << (32 - n - bit_cnt); | |
69 bit_cnt+=n; | |
70 } else { | |
71 bit_buf |= value >> (n + bit_cnt - 32); | |
64 | 72 *(UINT32 *)s->buf_ptr = be2me_32(bit_buf); |
0 | 73 //printf("bitbuf = %08x\n", bit_buf); |
74 s->buf_ptr+=4; | |
75 if (s->buf_ptr >= s->buf_end) | |
76 flush_buffer(s); | |
77 bit_cnt=bit_cnt + n - 32; | |
78 if (bit_cnt == 0) { | |
79 bit_buf = 0; | |
80 } else { | |
81 bit_buf = value << (32 - bit_cnt); | |
82 } | |
83 } | |
84 | |
85 s->bit_buf = bit_buf; | |
86 s->bit_cnt = bit_cnt; | |
87 } | |
88 | |
89 /* return the number of bits output */ | |
64 | 90 INT64 get_bit_count(PutBitContext *s) |
0 | 91 { |
64 | 92 return (s->buf_ptr - s->buf + s->data_out_size) * 8 + (INT64)s->bit_cnt; |
0 | 93 } |
94 | |
95 void align_put_bits(PutBitContext *s) | |
96 { | |
97 put_bits(s,(8 - s->bit_cnt) & 7,0); | |
98 } | |
99 | |
100 /* pad the end of the output stream with zeros */ | |
101 void flush_put_bits(PutBitContext *s) | |
102 { | |
103 while (s->bit_cnt > 0) { | |
104 /* XXX: should test end of buffer */ | |
105 *s->buf_ptr++=s->bit_buf >> 24; | |
106 s->bit_buf<<=8; | |
107 s->bit_cnt-=8; | |
108 } | |
109 flush_buffer(s); | |
110 s->bit_cnt=0; | |
111 s->bit_buf=0; | |
112 } | |
113 | |
24 | 114 /* for jpeg : escape 0xff with 0x00 after it */ |
0 | 115 void jput_bits(PutBitContext *s, int n, unsigned int value) |
116 { | |
117 unsigned int bit_buf, b; | |
118 int bit_cnt, i; | |
119 | |
120 assert(n == 32 || value < (1U << n)); | |
121 | |
122 bit_buf = s->bit_buf; | |
123 bit_cnt = s->bit_cnt; | |
124 | |
125 //printf("n=%d value=%x cnt=%d buf=%x\n", n, value, bit_cnt, bit_buf); | |
126 /* XXX: optimize */ | |
127 if (n < (32-bit_cnt)) { | |
128 bit_buf |= value << (32 - n - bit_cnt); | |
129 bit_cnt+=n; | |
130 } else { | |
131 bit_buf |= value >> (n + bit_cnt - 32); | |
132 /* handle escape */ | |
133 for(i=0;i<4;i++) { | |
134 b = (bit_buf >> 24); | |
135 *(s->buf_ptr++) = b; | |
136 if (b == 0xff) | |
137 *(s->buf_ptr++) = 0; | |
138 bit_buf <<= 8; | |
139 } | |
140 /* we flush the buffer sooner to handle worst case */ | |
141 if (s->buf_ptr >= (s->buf_end - 8)) | |
142 flush_buffer(s); | |
143 | |
144 bit_cnt=bit_cnt + n - 32; | |
145 if (bit_cnt == 0) { | |
146 bit_buf = 0; | |
147 } else { | |
148 bit_buf = value << (32 - bit_cnt); | |
149 } | |
150 } | |
151 | |
152 s->bit_buf = bit_buf; | |
153 s->bit_cnt = bit_cnt; | |
154 } | |
155 | |
156 /* pad the end of the output stream with zeros */ | |
157 void jflush_put_bits(PutBitContext *s) | |
158 { | |
159 unsigned int b; | |
160 | |
161 while (s->bit_cnt > 0) { | |
162 b = s->bit_buf >> 24; | |
163 *s->buf_ptr++ = b; | |
164 if (b == 0xff) | |
165 *s->buf_ptr++ = 0; | |
166 s->bit_buf<<=8; | |
167 s->bit_cnt-=8; | |
168 } | |
169 flush_buffer(s); | |
170 s->bit_cnt=0; | |
171 s->bit_buf=0; | |
172 } | |
173 | |
174 /* bit input functions */ | |
175 | |
176 void init_get_bits(GetBitContext *s, | |
177 UINT8 *buffer, int buffer_size) | |
178 { | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
179 #ifdef ALT_BITSTREAM_READER |
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
180 s->index=0; |
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
181 s->buffer= buffer; |
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
182 #else |
0 | 183 s->buf = buffer; |
184 s->buf_ptr = buffer; | |
185 s->buf_end = buffer + buffer_size; | |
186 s->bit_cnt = 0; | |
187 s->bit_buf = 0; | |
188 while (s->buf_ptr < s->buf_end && | |
189 s->bit_cnt < 32) { | |
190 s->bit_buf |= (*s->buf_ptr++ << (24 - s->bit_cnt)); | |
191 s->bit_cnt += 8; | |
192 } | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
193 #endif |
0 | 194 } |
195 | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
196 #ifndef ALT_BITSTREAM_READER |
0 | 197 /* n must be >= 1 and <= 32 */ |
20
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
198 /* also true: n > s->bit_cnt */ |
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
199 unsigned int get_bits_long(GetBitContext *s, int n) |
0 | 200 { |
201 unsigned int val; | |
202 int bit_cnt; | |
203 unsigned int bit_buf; | |
204 | |
205 #ifdef STATS | |
206 st_bit_counts[st_current_index] += n; | |
207 #endif | |
208 | |
209 bit_buf = s->bit_buf; | |
20
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
210 bit_cnt = s->bit_cnt - n; |
0 | 211 |
20
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
212 // if (bit_cnt >= 0) { |
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
213 // val = bit_buf >> (32 - n); |
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
214 // bit_buf <<= n; |
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
215 // } else |
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
216 { |
907b67420d84
inlineing common case of get_bits() -> gives 2speedup. more optim coming soon...
arpi_esp
parents:
8
diff
changeset
|
217 UINT8 *buf_ptr; |
0 | 218 val = bit_buf >> (32 - n); |
219 buf_ptr = s->buf_ptr; | |
220 buf_ptr += 4; | |
221 /* handle common case: we can read everything */ | |
222 if (buf_ptr <= s->buf_end) { | |
8 | 223 #if ARCH_X86 |
224 bit_buf = bswap_32(*((unsigned long*)(&buf_ptr[-4]))); | |
225 #else | |
226 bit_buf = (buf_ptr[-4] << 24) | | |
227 (buf_ptr[-3] << 16) | | |
0 | 228 (buf_ptr[-2] << 8) | |
8 | 229 (buf_ptr[-1]); |
230 #endif | |
0 | 231 } else { |
232 buf_ptr -= 4; | |
233 bit_buf = 0; | |
234 if (buf_ptr < s->buf_end) | |
235 bit_buf |= *buf_ptr++ << 24; | |
236 if (buf_ptr < s->buf_end) | |
237 bit_buf |= *buf_ptr++ << 16; | |
238 if (buf_ptr < s->buf_end) | |
239 bit_buf |= *buf_ptr++ << 8; | |
240 if (buf_ptr < s->buf_end) | |
241 bit_buf |= *buf_ptr++; | |
242 } | |
243 s->buf_ptr = buf_ptr; | |
244 val |= bit_buf >> (32 + bit_cnt); | |
245 bit_buf <<= - bit_cnt; | |
246 bit_cnt += 32; | |
247 } | |
248 s->bit_buf = bit_buf; | |
249 s->bit_cnt = bit_cnt; | |
250 return val; | |
251 } | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
252 #endif |
0 | 253 |
254 void align_get_bits(GetBitContext *s) | |
255 { | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
256 #ifdef ALT_BITSTREAM_READER |
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
257 s->index= (s->index + 7) & (~7); |
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
258 #else |
0 | 259 int n; |
260 n = s->bit_cnt & 7; | |
261 if (n > 0) { | |
262 get_bits(s, n); | |
263 } | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
264 #endif |
0 | 265 } |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
266 |
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
267 #ifndef ALT_BITSTREAM_READER |
144 | 268 /* This function is identical to get_bits_long(), the */ |
269 /* only diference is that it doesn't touch the buffer */ | |
270 /* it is usefull to see the buffer. */ | |
271 | |
272 unsigned int show_bits_long(GetBitContext *s, int n) | |
273 { | |
274 unsigned int val; | |
275 int bit_cnt; | |
276 unsigned int bit_buf; | |
277 UINT8 *buf_ptr; | |
278 | |
279 bit_buf = s->bit_buf; | |
280 bit_cnt = s->bit_cnt - n; | |
281 | |
282 val = bit_buf >> (32 - n); | |
283 buf_ptr = s->buf_ptr; | |
284 buf_ptr += 4; | |
285 | |
286 /* handle common case: we can read everything */ | |
287 if (buf_ptr <= s->buf_end) { | |
288 #ifdef ARCH_X86 | |
289 bit_buf = bswap_32(*((unsigned long*)(&buf_ptr[-4]))); | |
290 #else | |
291 bit_buf = (buf_ptr[-4] << 24) | | |
292 (buf_ptr[-3] << 16) | | |
293 (buf_ptr[-2] << 8) | | |
294 (buf_ptr[-1]); | |
295 #endif | |
296 } else { | |
297 buf_ptr -= 4; | |
298 bit_buf = 0; | |
299 if (buf_ptr < s->buf_end) | |
300 bit_buf |= *buf_ptr++ << 24; | |
301 if (buf_ptr < s->buf_end) | |
302 bit_buf |= *buf_ptr++ << 16; | |
303 if (buf_ptr < s->buf_end) | |
304 bit_buf |= *buf_ptr++ << 8; | |
305 if (buf_ptr < s->buf_end) | |
306 bit_buf |= *buf_ptr++; | |
307 } | |
308 val |= bit_buf >> (32 + bit_cnt); | |
309 bit_buf <<= - bit_cnt; | |
310 bit_cnt += 32; | |
311 | |
312 return val; | |
313 } | |
192
1e5f64be86fc
another bitstream reader code (faster on intel cpus) - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
144
diff
changeset
|
314 #endif |
0 | 315 |
316 /* VLC decoding */ | |
317 | |
318 //#define DEBUG_VLC | |
319 | |
320 #define GET_DATA(v, table, i, wrap, size) \ | |
321 {\ | |
322 UINT8 *ptr = (UINT8 *)table + i * wrap;\ | |
323 switch(size) {\ | |
324 case 1:\ | |
325 v = *(UINT8 *)ptr;\ | |
326 break;\ | |
327 case 2:\ | |
328 v = *(UINT16 *)ptr;\ | |
329 break;\ | |
330 default:\ | |
331 v = *(UINT32 *)ptr;\ | |
332 break;\ | |
333 }\ | |
334 } | |
335 | |
336 | |
337 static int alloc_table(VLC *vlc, int size) | |
338 { | |
339 int index; | |
340 index = vlc->table_size; | |
341 vlc->table_size += size; | |
342 if (vlc->table_size > vlc->table_allocated) { | |
343 vlc->table_allocated += (1 << vlc->bits); | |
344 vlc->table_bits = realloc(vlc->table_bits, | |
345 sizeof(INT8) * vlc->table_allocated); | |
346 vlc->table_codes = realloc(vlc->table_codes, | |
347 sizeof(INT16) * vlc->table_allocated); | |
348 if (!vlc->table_bits || | |
349 !vlc->table_codes) | |
350 return -1; | |
351 } | |
352 return index; | |
353 } | |
354 | |
355 static int build_table(VLC *vlc, int table_nb_bits, | |
356 int nb_codes, | |
357 const void *bits, int bits_wrap, int bits_size, | |
358 const void *codes, int codes_wrap, int codes_size, | |
359 UINT32 code_prefix, int n_prefix) | |
360 { | |
361 int i, j, k, n, table_size, table_index, nb, n1, index; | |
362 UINT32 code; | |
363 INT8 *table_bits; | |
364 INT16 *table_codes; | |
365 | |
366 table_size = 1 << table_nb_bits; | |
367 table_index = alloc_table(vlc, table_size); | |
368 #ifdef DEBUG_VLC | |
369 printf("new table index=%d size=%d code_prefix=%x n=%d\n", | |
370 table_index, table_size, code_prefix, n_prefix); | |
371 #endif | |
372 if (table_index < 0) | |
373 return -1; | |
374 table_bits = &vlc->table_bits[table_index]; | |
375 table_codes = &vlc->table_codes[table_index]; | |
376 | |
377 for(i=0;i<table_size;i++) { | |
378 table_bits[i] = 0; | |
379 table_codes[i] = -1; | |
380 } | |
381 | |
382 /* first pass: map codes and compute auxillary table sizes */ | |
383 for(i=0;i<nb_codes;i++) { | |
384 GET_DATA(n, bits, i, bits_wrap, bits_size); | |
385 GET_DATA(code, codes, i, codes_wrap, codes_size); | |
386 /* we accept tables with holes */ | |
387 if (n <= 0) | |
388 continue; | |
389 #if defined(DEBUG_VLC) && 0 | |
390 printf("i=%d n=%d code=0x%x\n", i, n, code); | |
391 #endif | |
392 /* if code matches the prefix, it is in the table */ | |
393 n -= n_prefix; | |
394 if (n > 0 && (code >> n) == code_prefix) { | |
395 if (n <= table_nb_bits) { | |
396 /* no need to add another table */ | |
397 j = (code << (table_nb_bits - n)) & (table_size - 1); | |
398 nb = 1 << (table_nb_bits - n); | |
399 for(k=0;k<nb;k++) { | |
400 #ifdef DEBUG_VLC | |
401 printf("%4x: code=%d n=%d\n", | |
402 j, i, n); | |
403 #endif | |
404 if (table_bits[j] != 0) { | |
405 fprintf(stderr, "incorrect codes\n"); | |
406 exit(1); | |
407 } | |
408 table_bits[j] = n; | |
409 table_codes[j] = i; | |
410 j++; | |
411 } | |
412 } else { | |
413 n -= table_nb_bits; | |
414 j = (code >> n) & ((1 << table_nb_bits) - 1); | |
415 #ifdef DEBUG_VLC | |
416 printf("%4x: n=%d (subtable)\n", | |
417 j, n); | |
418 #endif | |
419 /* compute table size */ | |
420 n1 = -table_bits[j]; | |
421 if (n > n1) | |
422 n1 = n; | |
423 table_bits[j] = -n1; | |
424 } | |
425 } | |
426 } | |
427 | |
428 /* second pass : fill auxillary tables recursively */ | |
429 for(i=0;i<table_size;i++) { | |
430 n = table_bits[i]; | |
431 if (n < 0) { | |
432 n = -n; | |
433 if (n > table_nb_bits) { | |
434 n = table_nb_bits; | |
435 table_bits[i] = -n; | |
436 } | |
437 index = build_table(vlc, n, nb_codes, | |
438 bits, bits_wrap, bits_size, | |
439 codes, codes_wrap, codes_size, | |
440 (code_prefix << table_nb_bits) | i, | |
441 n_prefix + table_nb_bits); | |
442 if (index < 0) | |
443 return -1; | |
444 /* note: realloc has been done, so reload tables */ | |
445 table_bits = &vlc->table_bits[table_index]; | |
446 table_codes = &vlc->table_codes[table_index]; | |
447 table_codes[i] = index; | |
448 } | |
449 } | |
450 return table_index; | |
451 } | |
452 | |
453 | |
24 | 454 /* Build VLC decoding tables suitable for use with get_vlc(). |
455 | |
456 'nb_bits' set thee decoding table size (2^nb_bits) entries. The | |
457 bigger it is, the faster is the decoding. But it should not be too | |
458 big to save memory and L1 cache. '9' is a good compromise. | |
459 | |
460 'nb_codes' : number of vlcs codes | |
461 | |
462 'bits' : table which gives the size (in bits) of each vlc code. | |
463 | |
464 'codes' : table which gives the bit pattern of of each vlc code. | |
465 | |
466 'xxx_wrap' : give the number of bytes between each entry of the | |
467 'bits' or 'codes' tables. | |
468 | |
469 'xxx_size' : gives the number of bytes of each entry of the 'bits' | |
470 or 'codes' tables. | |
471 | |
472 'wrap' and 'size' allows to use any memory configuration and types | |
473 (byte/word/long) to store the 'bits' and 'codes' tables. | |
474 */ | |
0 | 475 int init_vlc(VLC *vlc, int nb_bits, int nb_codes, |
476 const void *bits, int bits_wrap, int bits_size, | |
477 const void *codes, int codes_wrap, int codes_size) | |
478 { | |
479 vlc->bits = nb_bits; | |
480 vlc->table_bits = NULL; | |
481 vlc->table_codes = NULL; | |
482 vlc->table_allocated = 0; | |
483 vlc->table_size = 0; | |
484 #ifdef DEBUG_VLC | |
485 printf("build table nb_codes=%d\n", nb_codes); | |
486 #endif | |
487 | |
488 if (build_table(vlc, nb_bits, nb_codes, | |
489 bits, bits_wrap, bits_size, | |
490 codes, codes_wrap, codes_size, | |
491 0, 0) < 0) { | |
492 if (vlc->table_bits) | |
493 free(vlc->table_bits); | |
494 if (vlc->table_codes) | |
495 free(vlc->table_codes); | |
496 return -1; | |
497 } | |
498 return 0; | |
499 } | |
500 | |
501 | |
502 void free_vlc(VLC *vlc) | |
503 { | |
504 free(vlc->table_bits); | |
505 free(vlc->table_codes); | |
506 } | |
507 |