2 * default memory allocator for libavutil
3 * Copyright (c) 2002 Fabrice Bellard
5 * This file is part of FFmpeg.
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.
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.
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
24 * default memory allocator for libavutil
27 #define _XOPEN_SOURCE 600
41 #include "intreadwrite.h"
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)
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);
58 #endif /* MALLOC_PREFIX */
60 #define ALIGN (HAVE_AVX ? 32 : 16)
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. */
67 static size_t max_alloc_size= INT_MAX;
69 void av_max_alloc(size_t max){
73 void *av_malloc(size_t size)
76 #if CONFIG_MEMALIGN_HACK
80 /* let's disallow possible ambiguous cases */
81 if (size > (max_alloc_size - 32))
84 #if CONFIG_MEMALIGN_HACK
85 ptr = malloc(size + ALIGN);
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))
95 #elif HAVE_ALIGNED_MALLOC
96 ptr = _aligned_malloc(size, ALIGN);
98 ptr = memalign(ALIGN, size);
100 * Indeed, we should align it:
103 * on 32 for 586, PPro - K6-III
104 * on 64 for K7 (maybe for P3 too).
105 * Because L1 and L2 caches are aligned on those values.
106 * But I don't want to code such logic here!
109 * For AVX ASM. SSE / NEON needs only 16.
110 * Why not larger? Because I did not see a difference in benchmarks ...
112 /* benchmarks with P3
113 * memalign(64) + 1 3071, 3051, 3032
114 * memalign(64) + 2 3051, 3032, 3041
115 * memalign(64) + 4 2911, 2896, 2915
116 * memalign(64) + 8 2545, 2554, 2550
117 * memalign(64) + 16 2543, 2572, 2563
118 * memalign(64) + 32 2546, 2545, 2571
119 * memalign(64) + 64 2570, 2533, 2558
121 * BTW, malloc seems to do 8-byte alignment by default here.
130 #if CONFIG_MEMORY_POISONING
132 memset(ptr, 0x2a, size);
137 void *av_realloc(void *ptr, size_t size)
139 #if CONFIG_MEMALIGN_HACK
143 /* let's disallow possible ambiguous cases */
144 if (size > (max_alloc_size - 32))
147 #if CONFIG_MEMALIGN_HACK
148 //FIXME this isn't aligned correctly, though it probably isn't needed
150 return av_malloc(size);
151 diff = ((char *)ptr)[-1];
152 av_assert0(diff>0 && diff<=ALIGN);
153 ptr = realloc((char *)ptr - diff, size + diff);
155 ptr = (char *)ptr + diff;
157 #elif HAVE_ALIGNED_MALLOC
158 return _aligned_realloc(ptr, size + !size, ALIGN);
160 return realloc(ptr, size + !size);
164 void *av_realloc_f(void *ptr, size_t nelem, size_t elsize)
169 if (av_size_mult(elsize, nelem, &size)) {
173 r = av_realloc(ptr, size);
179 void av_free(void *ptr)
181 #if CONFIG_MEMALIGN_HACK
183 int v= ((char *)ptr)[-1];
184 av_assert0(v>0 && v<=ALIGN);
185 free((char *)ptr - v);
187 #elif HAVE_ALIGNED_MALLOC
194 void av_freep(void *arg)
196 void **ptr = (void **)arg;
201 void *av_mallocz(size_t size)
203 void *ptr = av_malloc(size);
205 memset(ptr, 0, size);
209 void *av_calloc(size_t nmemb, size_t size)
211 if (size <= 0 || nmemb >= INT_MAX / size)
213 return av_mallocz(nmemb * size);
216 char *av_strdup(const char *s)
220 int len = strlen(s) + 1;
221 ptr = av_malloc(len);
228 /* add one element to a dynamic array */
229 void av_dynarray_add(void *tab_ptr, int *nb_ptr, void *elem)
231 /* see similar ffmpeg.c:grow_array() */
236 tab = *(intptr_t**)tab_ptr;
237 if ((nb & (nb - 1)) == 0) {
242 tab = av_realloc(tab, nb_alloc * sizeof(intptr_t));
243 *(intptr_t**)tab_ptr = tab;
245 tab[nb++] = (intptr_t)elem;
249 static void fill16(uint8_t *dst, int len)
251 uint32_t v = AV_RN16(dst - 2);
267 static void fill24(uint8_t *dst, int len)
270 uint32_t v = AV_RB24(dst - 3);
271 uint32_t a = v << 8 | v >> 16;
272 uint32_t b = v << 16 | v >> 8;
273 uint32_t c = v << 24 | v;
275 uint32_t v = AV_RL24(dst - 3);
276 uint32_t a = v | v << 24;
277 uint32_t b = v >> 8 | v << 16;
278 uint32_t c = v >> 16 | v << 8;
307 static void fill32(uint8_t *dst, int len)
309 uint32_t v = AV_RN32(dst - 4);
323 void av_memcpy_backptr(uint8_t *dst, int back, int cnt)
325 const uint8_t *src = &dst[-back];
330 memset(dst, *src, cnt);
331 } else if (back == 2) {
333 } else if (back == 3) {
335 } else if (back == 4) {
340 while (cnt > blocklen) {
341 memcpy(dst, src, blocklen);
346 memcpy(dst, src, cnt);
350 AV_COPY32U(dst, src);
351 AV_COPY32U(dst + 4, src + 4);
357 AV_COPY32U(dst, src);
363 AV_COPY16U(dst, src);