unscaled memcpy variant bugfix (negative strides)
[libav.git] / libavcodec / common.c
CommitLineData
de6d9b64
FB
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.
27a3e2c5
MN
18 *
19 * alternative bitstream reader by Michael Niedermayer <michaelni@gmx.at>
de6d9b64 20 */
de6d9b64 21#include "common.h"
1a565432 22#include <math.h>
a9b3f630 23
de6d9b64
FB
24void 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
39static 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
51void 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);
1a565432 72 *(UINT32 *)s->buf_ptr = be2me_32(bit_buf);
de6d9b64
FB
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 */
1a565432 90INT64 get_bit_count(PutBitContext *s)
de6d9b64 91{
1a565432 92 return (s->buf_ptr - s->buf + s->data_out_size) * 8 + (INT64)s->bit_cnt;
de6d9b64
FB
93}
94
95void 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 */
101void 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
4e66ab3b 114/* for jpeg : escape 0xff with 0x00 after it */
de6d9b64
FB
115void 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 */
157void jflush_put_bits(PutBitContext *s)
158{
159 unsigned int b;
580b82fa 160 s->bit_buf |= ~1U >> s->bit_cnt; /* set all the unused bits to one */
de6d9b64
FB
161
162 while (s->bit_cnt > 0) {
163 b = s->bit_buf >> 24;
164 *s->buf_ptr++ = b;
165 if (b == 0xff)
166 *s->buf_ptr++ = 0;
167 s->bit_buf<<=8;
168 s->bit_cnt-=8;
169 }
170 flush_buffer(s);
171 s->bit_cnt=0;
172 s->bit_buf=0;
173}
174
175/* bit input functions */
176
177void init_get_bits(GetBitContext *s,
178 UINT8 *buffer, int buffer_size)
179{
27a3e2c5
MN
180#ifdef ALT_BITSTREAM_READER
181 s->index=0;
182 s->buffer= buffer;
183#else
de6d9b64
FB
184 s->buf = buffer;
185 s->buf_ptr = buffer;
186 s->buf_end = buffer + buffer_size;
187 s->bit_cnt = 0;
188 s->bit_buf = 0;
189 while (s->buf_ptr < s->buf_end &&
190 s->bit_cnt < 32) {
191 s->bit_buf |= (*s->buf_ptr++ << (24 - s->bit_cnt));
192 s->bit_cnt += 8;
193 }
27a3e2c5 194#endif
de6d9b64
FB
195}
196
27a3e2c5 197#ifndef ALT_BITSTREAM_READER
de6d9b64 198/* n must be >= 1 and <= 32 */
2931ecb9
A
199/* also true: n > s->bit_cnt */
200unsigned int get_bits_long(GetBitContext *s, int n)
de6d9b64
FB
201{
202 unsigned int val;
203 int bit_cnt;
204 unsigned int bit_buf;
de6d9b64
FB
205
206#ifdef STATS
207 st_bit_counts[st_current_index] += n;
208#endif
209
de6d9b64 210 bit_buf = s->bit_buf;
2931ecb9 211 bit_cnt = s->bit_cnt - n;
de6d9b64 212
2931ecb9
A
213// if (bit_cnt >= 0) {
214// val = bit_buf >> (32 - n);
215// bit_buf <<= n;
216// } else
217 {
218 UINT8 *buf_ptr;
de6d9b64
FB
219 val = bit_buf >> (32 - n);
220 buf_ptr = s->buf_ptr;
221 buf_ptr += 4;
222 /* handle common case: we can read everything */
223 if (buf_ptr <= s->buf_end) {
a9b3f630
NK
224#if ARCH_X86
225 bit_buf = bswap_32(*((unsigned long*)(&buf_ptr[-4])));
226#else
227 bit_buf = (buf_ptr[-4] << 24) |
228 (buf_ptr[-3] << 16) |
de6d9b64 229 (buf_ptr[-2] << 8) |
a9b3f630
NK
230 (buf_ptr[-1]);
231#endif
de6d9b64
FB
232 } else {
233 buf_ptr -= 4;
234 bit_buf = 0;
235 if (buf_ptr < s->buf_end)
236 bit_buf |= *buf_ptr++ << 24;
237 if (buf_ptr < s->buf_end)
238 bit_buf |= *buf_ptr++ << 16;
239 if (buf_ptr < s->buf_end)
240 bit_buf |= *buf_ptr++ << 8;
241 if (buf_ptr < s->buf_end)
242 bit_buf |= *buf_ptr++;
243 }
244 s->buf_ptr = buf_ptr;
245 val |= bit_buf >> (32 + bit_cnt);
246 bit_buf <<= - bit_cnt;
247 bit_cnt += 32;
248 }
249 s->bit_buf = bit_buf;
250 s->bit_cnt = bit_cnt;
251 return val;
252}
27a3e2c5 253#endif
de6d9b64
FB
254
255void align_get_bits(GetBitContext *s)
256{
27a3e2c5
MN
257#ifdef ALT_BITSTREAM_READER
258 s->index= (s->index + 7) & (~7);
259#else
de6d9b64
FB
260 int n;
261 n = s->bit_cnt & 7;
262 if (n > 0) {
263 get_bits(s, n);
264 }
27a3e2c5 265#endif
de6d9b64 266}
27a3e2c5
MN
267
268#ifndef ALT_BITSTREAM_READER
4949028f
J
269/* This function is identical to get_bits_long(), the */
270/* only diference is that it doesn't touch the buffer */
271/* it is usefull to see the buffer. */
272
273unsigned int show_bits_long(GetBitContext *s, int n)
274{
275 unsigned int val;
276 int bit_cnt;
277 unsigned int bit_buf;
278 UINT8 *buf_ptr;
279
280 bit_buf = s->bit_buf;
281 bit_cnt = s->bit_cnt - n;
282
283 val = bit_buf >> (32 - n);
284 buf_ptr = s->buf_ptr;
285 buf_ptr += 4;
286
287 /* handle common case: we can read everything */
288 if (buf_ptr <= s->buf_end) {
289#ifdef ARCH_X86
290 bit_buf = bswap_32(*((unsigned long*)(&buf_ptr[-4])));
291#else
292 bit_buf = (buf_ptr[-4] << 24) |
293 (buf_ptr[-3] << 16) |
294 (buf_ptr[-2] << 8) |
295 (buf_ptr[-1]);
296#endif
297 } else {
298 buf_ptr -= 4;
299 bit_buf = 0;
300 if (buf_ptr < s->buf_end)
301 bit_buf |= *buf_ptr++ << 24;
302 if (buf_ptr < s->buf_end)
303 bit_buf |= *buf_ptr++ << 16;
304 if (buf_ptr < s->buf_end)
305 bit_buf |= *buf_ptr++ << 8;
306 if (buf_ptr < s->buf_end)
307 bit_buf |= *buf_ptr++;
308 }
309 val |= bit_buf >> (32 + bit_cnt);
310 bit_buf <<= - bit_cnt;
311 bit_cnt += 32;
312
313 return val;
314}
27a3e2c5 315#endif
de6d9b64
FB
316
317/* VLC decoding */
318
319//#define DEBUG_VLC
320
321#define GET_DATA(v, table, i, wrap, size) \
322{\
323 UINT8 *ptr = (UINT8 *)table + i * wrap;\
324 switch(size) {\
325 case 1:\
326 v = *(UINT8 *)ptr;\
327 break;\
328 case 2:\
329 v = *(UINT16 *)ptr;\
330 break;\
331 default:\
332 v = *(UINT32 *)ptr;\
333 break;\
334 }\
335}
336
337
338static int alloc_table(VLC *vlc, int size)
339{
340 int index;
341 index = vlc->table_size;
342 vlc->table_size += size;
343 if (vlc->table_size > vlc->table_allocated) {
344 vlc->table_allocated += (1 << vlc->bits);
345 vlc->table_bits = realloc(vlc->table_bits,
346 sizeof(INT8) * vlc->table_allocated);
347 vlc->table_codes = realloc(vlc->table_codes,
348 sizeof(INT16) * vlc->table_allocated);
349 if (!vlc->table_bits ||
350 !vlc->table_codes)
351 return -1;
352 }
353 return index;
354}
355
356static int build_table(VLC *vlc, int table_nb_bits,
357 int nb_codes,
358 const void *bits, int bits_wrap, int bits_size,
359 const void *codes, int codes_wrap, int codes_size,
360 UINT32 code_prefix, int n_prefix)
361{
362 int i, j, k, n, table_size, table_index, nb, n1, index;
363 UINT32 code;
364 INT8 *table_bits;
365 INT16 *table_codes;
366
367 table_size = 1 << table_nb_bits;
368 table_index = alloc_table(vlc, table_size);
369#ifdef DEBUG_VLC
370 printf("new table index=%d size=%d code_prefix=%x n=%d\n",
371 table_index, table_size, code_prefix, n_prefix);
372#endif
373 if (table_index < 0)
374 return -1;
375 table_bits = &vlc->table_bits[table_index];
376 table_codes = &vlc->table_codes[table_index];
377
378 for(i=0;i<table_size;i++) {
379 table_bits[i] = 0;
380 table_codes[i] = -1;
381 }
382
383 /* first pass: map codes and compute auxillary table sizes */
384 for(i=0;i<nb_codes;i++) {
385 GET_DATA(n, bits, i, bits_wrap, bits_size);
386 GET_DATA(code, codes, i, codes_wrap, codes_size);
387 /* we accept tables with holes */
388 if (n <= 0)
389 continue;
390#if defined(DEBUG_VLC) && 0
391 printf("i=%d n=%d code=0x%x\n", i, n, code);
392#endif
393 /* if code matches the prefix, it is in the table */
394 n -= n_prefix;
395 if (n > 0 && (code >> n) == code_prefix) {
396 if (n <= table_nb_bits) {
397 /* no need to add another table */
398 j = (code << (table_nb_bits - n)) & (table_size - 1);
399 nb = 1 << (table_nb_bits - n);
400 for(k=0;k<nb;k++) {
401#ifdef DEBUG_VLC
402 printf("%4x: code=%d n=%d\n",
403 j, i, n);
404#endif
405 if (table_bits[j] != 0) {
406 fprintf(stderr, "incorrect codes\n");
407 exit(1);
408 }
409 table_bits[j] = n;
410 table_codes[j] = i;
411 j++;
412 }
413 } else {
414 n -= table_nb_bits;
415 j = (code >> n) & ((1 << table_nb_bits) - 1);
416#ifdef DEBUG_VLC
417 printf("%4x: n=%d (subtable)\n",
418 j, n);
419#endif
420 /* compute table size */
421 n1 = -table_bits[j];
422 if (n > n1)
423 n1 = n;
424 table_bits[j] = -n1;
425 }
426 }
427 }
428
429 /* second pass : fill auxillary tables recursively */
430 for(i=0;i<table_size;i++) {
431 n = table_bits[i];
432 if (n < 0) {
433 n = -n;
434 if (n > table_nb_bits) {
435 n = table_nb_bits;
436 table_bits[i] = -n;
437 }
438 index = build_table(vlc, n, nb_codes,
439 bits, bits_wrap, bits_size,
440 codes, codes_wrap, codes_size,
441 (code_prefix << table_nb_bits) | i,
442 n_prefix + table_nb_bits);
443 if (index < 0)
444 return -1;
445 /* note: realloc has been done, so reload tables */
446 table_bits = &vlc->table_bits[table_index];
447 table_codes = &vlc->table_codes[table_index];
448 table_codes[i] = index;
449 }
450 }
451 return table_index;
452}
453
454
4e66ab3b
FB
455/* Build VLC decoding tables suitable for use with get_vlc().
456
457 'nb_bits' set thee decoding table size (2^nb_bits) entries. The
458 bigger it is, the faster is the decoding. But it should not be too
459 big to save memory and L1 cache. '9' is a good compromise.
460
461 'nb_codes' : number of vlcs codes
462
463 'bits' : table which gives the size (in bits) of each vlc code.
464
465 'codes' : table which gives the bit pattern of of each vlc code.
466
467 'xxx_wrap' : give the number of bytes between each entry of the
468 'bits' or 'codes' tables.
469
470 'xxx_size' : gives the number of bytes of each entry of the 'bits'
471 or 'codes' tables.
472
473 'wrap' and 'size' allows to use any memory configuration and types
474 (byte/word/long) to store the 'bits' and 'codes' tables.
475*/
de6d9b64
FB
476int init_vlc(VLC *vlc, int nb_bits, int nb_codes,
477 const void *bits, int bits_wrap, int bits_size,
478 const void *codes, int codes_wrap, int codes_size)
479{
480 vlc->bits = nb_bits;
481 vlc->table_bits = NULL;
482 vlc->table_codes = NULL;
483 vlc->table_allocated = 0;
484 vlc->table_size = 0;
485#ifdef DEBUG_VLC
486 printf("build table nb_codes=%d\n", nb_codes);
487#endif
488
489 if (build_table(vlc, nb_bits, nb_codes,
490 bits, bits_wrap, bits_size,
491 codes, codes_wrap, codes_size,
492 0, 0) < 0) {
493 if (vlc->table_bits)
494 free(vlc->table_bits);
495 if (vlc->table_codes)
496 free(vlc->table_codes);
497 return -1;
498 }
499 return 0;
500}
501
502
503void free_vlc(VLC *vlc)
504{
505 free(vlc->table_bits);
506 free(vlc->table_codes);
507}
508