FFmpeg
 All Data Structures 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 "avutil.h"
40 #include "intreadwrite.h"
41 #include "mem.h"
42 
43 #ifdef MALLOC_PREFIX
44 
45 #define malloc AV_JOIN(MALLOC_PREFIX, malloc)
46 #define memalign AV_JOIN(MALLOC_PREFIX, memalign)
47 #define posix_memalign AV_JOIN(MALLOC_PREFIX, posix_memalign)
48 #define realloc AV_JOIN(MALLOC_PREFIX, realloc)
49 #define free AV_JOIN(MALLOC_PREFIX, free)
50 
51 void *malloc(size_t size);
52 void *memalign(size_t align, size_t size);
53 int posix_memalign(void **ptr, size_t align, size_t size);
54 void *realloc(void *ptr, size_t size);
55 void free(void *ptr);
56 
57 #endif /* MALLOC_PREFIX */
58 
59 #define ALIGN (HAVE_AVX ? 32 : 16)
60 
61 /* NOTE: if you want to override these functions with your own
62  * implementations (not recommended) you have to link libav* as
63  * dynamic libraries and remove -Wl,-Bsymbolic from the linker flags.
64  * Note that this will cost performance. */
65 
66 static size_t max_alloc_size= INT_MAX;
67 
68 void av_max_alloc(size_t max){
69  max_alloc_size = max;
70 }
71 
72 void *av_malloc(size_t size)
73 {
74  void *ptr = NULL;
75 #if CONFIG_MEMALIGN_HACK
76  long diff;
77 #endif
78 
79  /* let's disallow possible ambiguous cases */
80  if (size > (max_alloc_size - 32))
81  return NULL;
82 
83 #if CONFIG_MEMALIGN_HACK
84  ptr = malloc(size + ALIGN);
85  if (!ptr)
86  return ptr;
87  diff = ((~(long)ptr)&(ALIGN - 1)) + 1;
88  ptr = (char *)ptr + diff;
89  ((char *)ptr)[-1] = diff;
90 #elif HAVE_POSIX_MEMALIGN
91  if (size) //OS X on SDK 10.6 has a broken posix_memalign implementation
92  if (posix_memalign(&ptr, ALIGN, size))
93  ptr = NULL;
94 #elif HAVE_ALIGNED_MALLOC
95  ptr = _aligned_malloc(size, ALIGN);
96 #elif HAVE_MEMALIGN
97  ptr = memalign(ALIGN, size);
98  /* Why 64?
99  * Indeed, we should align it:
100  * on 4 for 386
101  * on 16 for 486
102  * on 32 for 586, PPro - K6-III
103  * on 64 for K7 (maybe for P3 too).
104  * Because L1 and L2 caches are aligned on those values.
105  * But I don't want to code such logic here!
106  */
107  /* Why 32?
108  * For AVX ASM. SSE / NEON needs only 16.
109  * Why not larger? Because I did not see a difference in benchmarks ...
110  */
111  /* benchmarks with P3
112  * memalign(64) + 1 3071, 3051, 3032
113  * memalign(64) + 2 3051, 3032, 3041
114  * memalign(64) + 4 2911, 2896, 2915
115  * memalign(64) + 8 2545, 2554, 2550
116  * memalign(64) + 16 2543, 2572, 2563
117  * memalign(64) + 32 2546, 2545, 2571
118  * memalign(64) + 64 2570, 2533, 2558
119  *
120  * BTW, malloc seems to do 8-byte alignment by default here.
121  */
122 #else
123  ptr = malloc(size);
124 #endif
125  if(!ptr && !size) {
126  size = 1;
127  ptr= av_malloc(1);
128  }
129 #if CONFIG_MEMORY_POISONING
130  if (ptr)
131  memset(ptr, 0x2a, size);
132 #endif
133  return ptr;
134 }
135 
136 void *av_realloc(void *ptr, size_t size)
137 {
138 #if CONFIG_MEMALIGN_HACK
139  int diff;
140 #endif
141 
142  /* let's disallow possible ambiguous cases */
143  if (size > (max_alloc_size - 32))
144  return NULL;
145 
146 #if CONFIG_MEMALIGN_HACK
147  //FIXME this isn't aligned correctly, though it probably isn't needed
148  if (!ptr)
149  return av_malloc(size);
150  diff = ((char *)ptr)[-1];
151  ptr = realloc((char *)ptr - diff, size + diff);
152  if (ptr)
153  ptr = (char *)ptr + diff;
154  return ptr;
155 #elif HAVE_ALIGNED_MALLOC
156  return _aligned_realloc(ptr, size + !size, ALIGN);
157 #else
158  return realloc(ptr, size + !size);
159 #endif
160 }
161 
162 void *av_realloc_f(void *ptr, size_t nelem, size_t elsize)
163 {
164  size_t size;
165  void *r;
166 
167  if (av_size_mult(elsize, nelem, &size)) {
168  av_free(ptr);
169  return NULL;
170  }
171  r = av_realloc(ptr, size);
172  if (!r && size)
173  av_free(ptr);
174  return r;
175 }
176 
177 void av_free(void *ptr)
178 {
179 #if CONFIG_MEMALIGN_HACK
180  if (ptr)
181  free((char *)ptr - ((char *)ptr)[-1]);
182 #elif HAVE_ALIGNED_MALLOC
183  _aligned_free(ptr);
184 #else
185  free(ptr);
186 #endif
187 }
188 
189 void av_freep(void *arg)
190 {
191  void **ptr = (void **)arg;
192  av_free(*ptr);
193  *ptr = NULL;
194 }
195 
196 void *av_mallocz(size_t size)
197 {
198  void *ptr = av_malloc(size);
199  if (ptr)
200  memset(ptr, 0, size);
201  return ptr;
202 }
203 
204 void *av_calloc(size_t nmemb, size_t size)
205 {
206  if (size <= 0 || nmemb >= INT_MAX / size)
207  return NULL;
208  return av_mallocz(nmemb * size);
209 }
210 
211 char *av_strdup(const char *s)
212 {
213  char *ptr = NULL;
214  if (s) {
215  int len = strlen(s) + 1;
216  ptr = av_malloc(len);
217  if (ptr)
218  memcpy(ptr, s, len);
219  }
220  return ptr;
221 }
222 
223 /* add one element to a dynamic array */
224 void av_dynarray_add(void *tab_ptr, int *nb_ptr, void *elem)
225 {
226  /* see similar ffmpeg.c:grow_array() */
227  int nb, nb_alloc;
228  intptr_t *tab;
229 
230  nb = *nb_ptr;
231  tab = *(intptr_t**)tab_ptr;
232  if ((nb & (nb - 1)) == 0) {
233  if (nb == 0)
234  nb_alloc = 1;
235  else
236  nb_alloc = nb * 2;
237  tab = av_realloc(tab, nb_alloc * sizeof(intptr_t));
238  *(intptr_t**)tab_ptr = tab;
239  }
240  tab[nb++] = (intptr_t)elem;
241  *nb_ptr = nb;
242 }
243 
244 static void fill16(uint8_t *dst, int len)
245 {
246  uint32_t v = AV_RN16(dst - 2);
247 
248  v |= v << 16;
249 
250  while (len >= 4) {
251  AV_WN32(dst, v);
252  dst += 4;
253  len -= 4;
254  }
255 
256  while (len--) {
257  *dst = dst[-2];
258  dst++;
259  }
260 }
261 
262 static void fill24(uint8_t *dst, int len)
263 {
264 #if HAVE_BIGENDIAN
265  uint32_t v = AV_RB24(dst - 3);
266  uint32_t a = v << 8 | v >> 16;
267  uint32_t b = v << 16 | v >> 8;
268  uint32_t c = v << 24 | v;
269 #else
270  uint32_t v = AV_RL24(dst - 3);
271  uint32_t a = v | v << 24;
272  uint32_t b = v >> 8 | v << 16;
273  uint32_t c = v >> 16 | v << 8;
274 #endif
275 
276  while (len >= 12) {
277  AV_WN32(dst, a);
278  AV_WN32(dst + 4, b);
279  AV_WN32(dst + 8, c);
280  dst += 12;
281  len -= 12;
282  }
283 
284  if (len >= 4) {
285  AV_WN32(dst, a);
286  dst += 4;
287  len -= 4;
288  }
289 
290  if (len >= 4) {
291  AV_WN32(dst, b);
292  dst += 4;
293  len -= 4;
294  }
295 
296  while (len--) {
297  *dst = dst[-3];
298  dst++;
299  }
300 }
301 
302 static void fill32(uint8_t *dst, int len)
303 {
304  uint32_t v = AV_RN32(dst - 4);
305 
306  while (len >= 4) {
307  AV_WN32(dst, v);
308  dst += 4;
309  len -= 4;
310  }
311 
312  while (len--) {
313  *dst = dst[-4];
314  dst++;
315  }
316 }
317 
318 void av_memcpy_backptr(uint8_t *dst, int back, int cnt)
319 {
320  const uint8_t *src = &dst[-back];
321  if (back <= 1) {
322  memset(dst, *src, cnt);
323  } else if (back == 2) {
324  fill16(dst, cnt);
325  } else if (back == 3) {
326  fill24(dst, cnt);
327  } else if (back == 4) {
328  fill32(dst, cnt);
329  } else {
330  if (cnt >= 16) {
331  int blocklen = back;
332  while (cnt > blocklen) {
333  memcpy(dst, src, blocklen);
334  dst += blocklen;
335  cnt -= blocklen;
336  blocklen <<= 1;
337  }
338  memcpy(dst, src, cnt);
339  return;
340  }
341  if (cnt >= 8) {
342  AV_COPY32U(dst, src);
343  AV_COPY32U(dst + 4, src + 4);
344  src += 8;
345  dst += 8;
346  cnt -= 8;
347  }
348  if (cnt >= 4) {
349  AV_COPY32U(dst, src);
350  src += 4;
351  dst += 4;
352  cnt -= 4;
353  }
354  if (cnt >= 2) {
355  AV_COPY16U(dst, src);
356  src += 2;
357  dst += 2;
358  cnt -= 2;
359  }
360  if (cnt)
361  *dst = *src;
362  }
363 }
364