FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
mem.c
Go to the documentation of this file.
1 /*
2  * default memory allocator for libavutil
3  * Copyright (c) 2002 Fabrice Bellard
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * default memory allocator for libavutil
25  */
26 
27 #define _XOPEN_SOURCE 600
28 
29 #include "config.h"
30 
31 #include <limits.h>
32 #include <stdint.h>
33 #include <stdlib.h>
34 #include <string.h>
35 #if HAVE_MALLOC_H
36 #include <malloc.h>
37 #endif
38 
39 #include "avassert.h"
40 #include "avutil.h"
41 #include "intreadwrite.h"
42 #include "mem.h"
43 
44 #ifdef MALLOC_PREFIX
45 
46 #define malloc AV_JOIN(MALLOC_PREFIX, malloc)
47 #define memalign AV_JOIN(MALLOC_PREFIX, memalign)
48 #define posix_memalign AV_JOIN(MALLOC_PREFIX, posix_memalign)
49 #define realloc AV_JOIN(MALLOC_PREFIX, realloc)
50 #define free AV_JOIN(MALLOC_PREFIX, free)
51 
52 void *malloc(size_t size);
53 void *memalign(size_t align, size_t size);
54 int posix_memalign(void **ptr, size_t align, size_t size);
55 void *realloc(void *ptr, size_t size);
56 void free(void *ptr);
57 
58 #endif /* MALLOC_PREFIX */
59 
60 #define ALIGN (HAVE_AVX ? 32 : 16)
61 
62 /* NOTE: if you want to override these functions with your own
63  * implementations (not recommended) you have to link libav* as
64  * dynamic libraries and remove -Wl,-Bsymbolic from the linker flags.
65  * Note that this will cost performance. */
66 
67 static size_t max_alloc_size= INT_MAX;
68 
69 void av_max_alloc(size_t max){
70  max_alloc_size = max;
71 }
72 
73 void *av_malloc(size_t size)
74 {
75  void *ptr = NULL;
76 #if CONFIG_MEMALIGN_HACK
77  long diff;
78 #endif
79 
80  /* let's disallow possible ambiguous cases */
81  if (size > (max_alloc_size - 32))
82  return NULL;
83 
84 #if CONFIG_MEMALIGN_HACK
85  ptr = malloc(size + ALIGN);
86  if (!ptr)
87  return ptr;
88  diff = ((~(long)ptr)&(ALIGN - 1)) + 1;
89  ptr = (char *)ptr + diff;
90  ((char *)ptr)[-1] = diff;
91 #elif HAVE_POSIX_MEMALIGN
92  if (size) //OS X on SDK 10.6 has a broken posix_memalign implementation
93  if (posix_memalign(&ptr, ALIGN, size))
94  ptr = NULL;
95 #elif HAVE_ALIGNED_MALLOC
96  ptr = _aligned_malloc(size, ALIGN);
97 #elif HAVE_MEMALIGN
98 #ifndef __DJGPP__
99  ptr = memalign(ALIGN, size);
100 #else
101  ptr = memalign(size, ALIGN);
102 #endif
103  /* Why 64?
104  * Indeed, we should align it:
105  * on 4 for 386
106  * on 16 for 486
107  * on 32 for 586, PPro - K6-III
108  * on 64 for K7 (maybe for P3 too).
109  * Because L1 and L2 caches are aligned on those values.
110  * But I don't want to code such logic here!
111  */
112  /* Why 32?
113  * For AVX ASM. SSE / NEON needs only 16.
114  * Why not larger? Because I did not see a difference in benchmarks ...
115  */
116  /* benchmarks with P3
117  * memalign(64) + 1 3071, 3051, 3032
118  * memalign(64) + 2 3051, 3032, 3041
119  * memalign(64) + 4 2911, 2896, 2915
120  * memalign(64) + 8 2545, 2554, 2550
121  * memalign(64) + 16 2543, 2572, 2563
122  * memalign(64) + 32 2546, 2545, 2571
123  * memalign(64) + 64 2570, 2533, 2558
124  *
125  * BTW, malloc seems to do 8-byte alignment by default here.
126  */
127 #else
128  ptr = malloc(size);
129 #endif
130  if(!ptr && !size) {
131  size = 1;
132  ptr= av_malloc(1);
133  }
134 #if CONFIG_MEMORY_POISONING
135  if (ptr)
136  memset(ptr, FF_MEMORY_POISON, size);
137 #endif
138  return ptr;
139 }
140 
141 void *av_realloc(void *ptr, size_t size)
142 {
143 #if CONFIG_MEMALIGN_HACK
144  int diff;
145 #endif
146 
147  /* let's disallow possible ambiguous cases */
148  if (size > (max_alloc_size - 32))
149  return NULL;
150 
151 #if CONFIG_MEMALIGN_HACK
152  //FIXME this isn't aligned correctly, though it probably isn't needed
153  if (!ptr)
154  return av_malloc(size);
155  diff = ((char *)ptr)[-1];
156  av_assert0(diff>0 && diff<=ALIGN);
157  ptr = realloc((char *)ptr - diff, size + diff);
158  if (ptr)
159  ptr = (char *)ptr + diff;
160  return ptr;
161 #elif HAVE_ALIGNED_MALLOC
162  return _aligned_realloc(ptr, size + !size, ALIGN);
163 #else
164  return realloc(ptr, size + !size);
165 #endif
166 }
167 
168 void *av_realloc_f(void *ptr, size_t nelem, size_t elsize)
169 {
170  size_t size;
171  void *r;
172 
173  if (av_size_mult(elsize, nelem, &size)) {
174  av_free(ptr);
175  return NULL;
176  }
177  r = av_realloc(ptr, size);
178  if (!r && size)
179  av_free(ptr);
180  return r;
181 }
182 
183 void *av_realloc_array(void *ptr, size_t nmemb, size_t size)
184 {
185  if (size <= 0 || nmemb >= INT_MAX / size)
186  return NULL;
187  return av_realloc(ptr, nmemb * size);
188 }
189 
190 int av_reallocp_array(void *ptr, size_t nmemb, size_t size)
191 {
192  void **ptrptr = ptr;
193  *ptrptr = av_realloc_f(*ptrptr, nmemb, size);
194  if (!*ptrptr && !(nmemb && size))
195  return AVERROR(ENOMEM);
196  return 0;
197 }
198 
199 void av_free(void *ptr)
200 {
201 #if CONFIG_MEMALIGN_HACK
202  if (ptr) {
203  int v= ((char *)ptr)[-1];
204  av_assert0(v>0 && v<=ALIGN);
205  free((char *)ptr - v);
206  }
207 #elif HAVE_ALIGNED_MALLOC
208  _aligned_free(ptr);
209 #else
210  free(ptr);
211 #endif
212 }
213 
214 void av_freep(void *arg)
215 {
216  void **ptr = (void **)arg;
217  av_free(*ptr);
218  *ptr = NULL;
219 }
220 
221 void *av_mallocz(size_t size)
222 {
223  void *ptr = av_malloc(size);
224  if (ptr)
225  memset(ptr, 0, size);
226  return ptr;
227 }
228 
229 void *av_calloc(size_t nmemb, size_t size)
230 {
231  if (size <= 0 || nmemb >= INT_MAX / size)
232  return NULL;
233  return av_mallocz(nmemb * size);
234 }
235 
236 char *av_strdup(const char *s)
237 {
238  char *ptr = NULL;
239  if (s) {
240  int len = strlen(s) + 1;
241  ptr = av_malloc(len);
242  if (ptr)
243  memcpy(ptr, s, len);
244  }
245  return ptr;
246 }
247 
248 void *av_memdup(const void *p, size_t size)
249 {
250  void *ptr = NULL;
251  if (p) {
252  ptr = av_malloc(size);
253  if (ptr)
254  memcpy(ptr, p, size);
255  }
256  return ptr;
257 }
258 
259 void av_dynarray_add(void *tab_ptr, int *nb_ptr, void *elem)
260 {
261  /* see similar ffmpeg.c:grow_array() */
262  int nb, nb_alloc;
263  intptr_t *tab;
264 
265  nb = *nb_ptr;
266  tab = *(intptr_t**)tab_ptr;
267  if ((nb & (nb - 1)) == 0) {
268  if (nb == 0) {
269  nb_alloc = 1;
270  } else {
271  if (nb > INT_MAX / (2 * sizeof(intptr_t)))
272  goto fail;
273  nb_alloc = nb * 2;
274  }
275  tab = av_realloc(tab, nb_alloc * sizeof(intptr_t));
276  if (!tab)
277  goto fail;
278  *(intptr_t**)tab_ptr = tab;
279  }
280  tab[nb++] = (intptr_t)elem;
281  *nb_ptr = nb;
282  return;
283 
284 fail:
285  av_freep(tab_ptr);
286  *nb_ptr = 0;
287 }
288 
289 void *av_dynarray2_add(void **tab_ptr, int *nb_ptr, size_t elem_size,
290  const uint8_t *elem_data)
291 {
292  int nb = *nb_ptr, nb_alloc;
293  uint8_t *tab = *tab_ptr, *tab_elem_data;
294 
295  if ((nb & (nb - 1)) == 0) {
296  if (nb == 0) {
297  nb_alloc = 1;
298  } else {
299  if (nb > INT_MAX / (2 * elem_size))
300  goto fail;
301  nb_alloc = nb * 2;
302  }
303  tab = av_realloc(tab, nb_alloc * elem_size);
304  if (!tab)
305  goto fail;
306  *tab_ptr = tab;
307  }
308  *nb_ptr = nb + 1;
309  tab_elem_data = tab + nb*elem_size;
310  if (elem_data)
311  memcpy(tab_elem_data, elem_data, elem_size);
312  else if (CONFIG_MEMORY_POISONING)
313  memset(tab_elem_data, FF_MEMORY_POISON, elem_size);
314  return tab_elem_data;
315 
316 fail:
317  av_freep(tab_ptr);
318  *nb_ptr = 0;
319  return NULL;
320 }
321 
322 static void fill16(uint8_t *dst, int len)
323 {
324  uint32_t v = AV_RN16(dst - 2);
325 
326  v |= v << 16;
327 
328  while (len >= 4) {
329  AV_WN32(dst, v);
330  dst += 4;
331  len -= 4;
332  }
333 
334  while (len--) {
335  *dst = dst[-2];
336  dst++;
337  }
338 }
339 
340 static void fill24(uint8_t *dst, int len)
341 {
342 #if HAVE_BIGENDIAN
343  uint32_t v = AV_RB24(dst - 3);
344  uint32_t a = v << 8 | v >> 16;
345  uint32_t b = v << 16 | v >> 8;
346  uint32_t c = v << 24 | v;
347 #else
348  uint32_t v = AV_RL24(dst - 3);
349  uint32_t a = v | v << 24;
350  uint32_t b = v >> 8 | v << 16;
351  uint32_t c = v >> 16 | v << 8;
352 #endif
353 
354  while (len >= 12) {
355  AV_WN32(dst, a);
356  AV_WN32(dst + 4, b);
357  AV_WN32(dst + 8, c);
358  dst += 12;
359  len -= 12;
360  }
361 
362  if (len >= 4) {
363  AV_WN32(dst, a);
364  dst += 4;
365  len -= 4;
366  }
367 
368  if (len >= 4) {
369  AV_WN32(dst, b);
370  dst += 4;
371  len -= 4;
372  }
373 
374  while (len--) {
375  *dst = dst[-3];
376  dst++;
377  }
378 }
379 
380 static void fill32(uint8_t *dst, int len)
381 {
382  uint32_t v = AV_RN32(dst - 4);
383 
384  while (len >= 4) {
385  AV_WN32(dst, v);
386  dst += 4;
387  len -= 4;
388  }
389 
390  while (len--) {
391  *dst = dst[-4];
392  dst++;
393  }
394 }
395 
396 void av_memcpy_backptr(uint8_t *dst, int back, int cnt)
397 {
398  const uint8_t *src = &dst[-back];
399  if (!back)
400  return;
401 
402  if (back == 1) {
403  memset(dst, *src, cnt);
404  } else if (back == 2) {
405  fill16(dst, cnt);
406  } else if (back == 3) {
407  fill24(dst, cnt);
408  } else if (back == 4) {
409  fill32(dst, cnt);
410  } else {
411  if (cnt >= 16) {
412  int blocklen = back;
413  while (cnt > blocklen) {
414  memcpy(dst, src, blocklen);
415  dst += blocklen;
416  cnt -= blocklen;
417  blocklen <<= 1;
418  }
419  memcpy(dst, src, cnt);
420  return;
421  }
422  if (cnt >= 8) {
423  AV_COPY32U(dst, src);
424  AV_COPY32U(dst + 4, src + 4);
425  src += 8;
426  dst += 8;
427  cnt -= 8;
428  }
429  if (cnt >= 4) {
430  AV_COPY32U(dst, src);
431  src += 4;
432  dst += 4;
433  cnt -= 4;
434  }
435  if (cnt >= 2) {
436  AV_COPY16U(dst, src);
437  src += 2;
438  dst += 2;
439  cnt -= 2;
440  }
441  if (cnt)
442  *dst = *src;
443  }
444 }
445