yading@11: /*
yading@11:  * default memory allocator for libavutil
yading@11:  * Copyright (c) 2002 Fabrice Bellard
yading@11:  *
yading@11:  * This file is part of FFmpeg.
yading@11:  *
yading@11:  * FFmpeg is free software; you can redistribute it and/or
yading@11:  * modify it under the terms of the GNU Lesser General Public
yading@11:  * License as published by the Free Software Foundation; either
yading@11:  * version 2.1 of the License, or (at your option) any later version.
yading@11:  *
yading@11:  * FFmpeg is distributed in the hope that it will be useful,
yading@11:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@11:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@11:  * Lesser General Public License for more details.
yading@11:  *
yading@11:  * You should have received a copy of the GNU Lesser General Public
yading@11:  * License along with FFmpeg; if not, write to the Free Software
yading@11:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@11:  */
yading@11: 
yading@11: /**
yading@11:  * @file
yading@11:  * default memory allocator for libavutil
yading@11:  */
yading@11: 
yading@11: #define _XOPEN_SOURCE 600
yading@11: 
yading@11: #include "config.h"
yading@11: 
yading@11: #include <limits.h>
yading@11: #include <stdint.h>
yading@11: #include <stdlib.h>
yading@11: #include <string.h>
yading@11: #if HAVE_MALLOC_H
yading@11: #include <malloc.h>
yading@11: #endif
yading@11: 
yading@11: #include "avassert.h"
yading@11: #include "avutil.h"
yading@11: #include "intreadwrite.h"
yading@11: #include "mem.h"
yading@11: 
yading@11: #ifdef MALLOC_PREFIX
yading@11: 
yading@11: #define malloc         AV_JOIN(MALLOC_PREFIX, malloc)
yading@11: #define memalign       AV_JOIN(MALLOC_PREFIX, memalign)
yading@11: #define posix_memalign AV_JOIN(MALLOC_PREFIX, posix_memalign)
yading@11: #define realloc        AV_JOIN(MALLOC_PREFIX, realloc)
yading@11: #define free           AV_JOIN(MALLOC_PREFIX, free)
yading@11: 
yading@11: void *malloc(size_t size);
yading@11: void *memalign(size_t align, size_t size);
yading@11: int   posix_memalign(void **ptr, size_t align, size_t size);
yading@11: void *realloc(void *ptr, size_t size);
yading@11: void  free(void *ptr);
yading@11: 
yading@11: #endif /* MALLOC_PREFIX */
yading@11: 
yading@11: #define ALIGN (HAVE_AVX ? 32 : 16)
yading@11: 
yading@11: /* NOTE: if you want to override these functions with your own
yading@11:  * implementations (not recommended) you have to link libav* as
yading@11:  * dynamic libraries and remove -Wl,-Bsymbolic from the linker flags.
yading@11:  * Note that this will cost performance. */
yading@11: 
yading@11: static size_t max_alloc_size= INT_MAX;
yading@11: 
yading@11: void av_max_alloc(size_t max){
yading@11:     max_alloc_size = max;
yading@11: }
yading@11: 
yading@11: void *av_malloc(size_t size)
yading@11: {
yading@11:     void *ptr = NULL;
yading@11: #if CONFIG_MEMALIGN_HACK
yading@11:     long diff;
yading@11: #endif
yading@11: 
yading@11:     /* let's disallow possible ambiguous cases */
yading@11:     if (size > (max_alloc_size - 32))
yading@11:         return NULL;
yading@11: 
yading@11: #if CONFIG_MEMALIGN_HACK
yading@11:     ptr = malloc(size + ALIGN);
yading@11:     if (!ptr)
yading@11:         return ptr;
yading@11:     diff              = ((~(long)ptr)&(ALIGN - 1)) + 1;
yading@11:     ptr               = (char *)ptr + diff;
yading@11:     ((char *)ptr)[-1] = diff;
yading@11: #elif HAVE_POSIX_MEMALIGN
yading@11:     if (size) //OS X on SDK 10.6 has a broken posix_memalign implementation
yading@11:     if (posix_memalign(&ptr, ALIGN, size))
yading@11:         ptr = NULL;
yading@11: #elif HAVE_ALIGNED_MALLOC
yading@11:     ptr = _aligned_malloc(size, ALIGN);
yading@11: #elif HAVE_MEMALIGN
yading@11: #ifndef __DJGPP__
yading@11:     ptr = memalign(ALIGN, size);
yading@11: #else
yading@11:     ptr = memalign(size, ALIGN);
yading@11: #endif
yading@11:     /* Why 64?
yading@11:      * Indeed, we should align it:
yading@11:      *   on  4 for 386
yading@11:      *   on 16 for 486
yading@11:      *   on 32 for 586, PPro - K6-III
yading@11:      *   on 64 for K7 (maybe for P3 too).
yading@11:      * Because L1 and L2 caches are aligned on those values.
yading@11:      * But I don't want to code such logic here!
yading@11:      */
yading@11:     /* Why 32?
yading@11:      * For AVX ASM. SSE / NEON needs only 16.
yading@11:      * Why not larger? Because I did not see a difference in benchmarks ...
yading@11:      */
yading@11:     /* benchmarks with P3
yading@11:      * memalign(64) + 1          3071, 3051, 3032
yading@11:      * memalign(64) + 2          3051, 3032, 3041
yading@11:      * memalign(64) + 4          2911, 2896, 2915
yading@11:      * memalign(64) + 8          2545, 2554, 2550
yading@11:      * memalign(64) + 16         2543, 2572, 2563
yading@11:      * memalign(64) + 32         2546, 2545, 2571
yading@11:      * memalign(64) + 64         2570, 2533, 2558
yading@11:      *
yading@11:      * BTW, malloc seems to do 8-byte alignment by default here.
yading@11:      */
yading@11: #else
yading@11:     ptr = malloc(size);
yading@11: #endif
yading@11:     if(!ptr && !size) {
yading@11:         size = 1;
yading@11:         ptr= av_malloc(1);
yading@11:     }
yading@11: #if CONFIG_MEMORY_POISONING
yading@11:     if (ptr)
yading@11:         memset(ptr, 0x2a, size);
yading@11: #endif
yading@11:     return ptr;
yading@11: }
yading@11: 
yading@11: void *av_realloc(void *ptr, size_t size)
yading@11: {
yading@11: #if CONFIG_MEMALIGN_HACK
yading@11:     int diff;
yading@11: #endif
yading@11: 
yading@11:     /* let's disallow possible ambiguous cases */
yading@11:     if (size > (max_alloc_size - 32))
yading@11:         return NULL;
yading@11: 
yading@11: #if CONFIG_MEMALIGN_HACK
yading@11:     //FIXME this isn't aligned correctly, though it probably isn't needed
yading@11:     if (!ptr)
yading@11:         return av_malloc(size);
yading@11:     diff = ((char *)ptr)[-1];
yading@11:     av_assert0(diff>0 && diff<=ALIGN);
yading@11:     ptr = realloc((char *)ptr - diff, size + diff);
yading@11:     if (ptr)
yading@11:         ptr = (char *)ptr + diff;
yading@11:     return ptr;
yading@11: #elif HAVE_ALIGNED_MALLOC
yading@11:     return _aligned_realloc(ptr, size + !size, ALIGN);
yading@11: #else
yading@11:     return realloc(ptr, size + !size);
yading@11: #endif
yading@11: }
yading@11: 
yading@11: void *av_realloc_f(void *ptr, size_t nelem, size_t elsize)
yading@11: {
yading@11:     size_t size;
yading@11:     void *r;
yading@11: 
yading@11:     if (av_size_mult(elsize, nelem, &size)) {
yading@11:         av_free(ptr);
yading@11:         return NULL;
yading@11:     }
yading@11:     r = av_realloc(ptr, size);
yading@11:     if (!r && size)
yading@11:         av_free(ptr);
yading@11:     return r;
yading@11: }
yading@11: 
yading@11: void av_free(void *ptr)
yading@11: {
yading@11: #if CONFIG_MEMALIGN_HACK
yading@11:     if (ptr) {
yading@11:         int v= ((char *)ptr)[-1];
yading@11:         av_assert0(v>0 && v<=ALIGN);
yading@11:         free((char *)ptr - v);
yading@11:     }
yading@11: #elif HAVE_ALIGNED_MALLOC
yading@11:     _aligned_free(ptr);
yading@11: #else
yading@11:     free(ptr);
yading@11: #endif
yading@11: }
yading@11: 
yading@11: void av_freep(void *arg)
yading@11: {
yading@11:     void **ptr = (void **)arg;
yading@11:     av_free(*ptr);
yading@11:     *ptr = NULL;
yading@11: }
yading@11: 
yading@11: void *av_mallocz(size_t size)
yading@11: {
yading@11:     void *ptr = av_malloc(size);
yading@11:     if (ptr)
yading@11:         memset(ptr, 0, size);
yading@11:     return ptr;
yading@11: }
yading@11: 
yading@11: void *av_calloc(size_t nmemb, size_t size)
yading@11: {
yading@11:     if (size <= 0 || nmemb >= INT_MAX / size)
yading@11:         return NULL;
yading@11:     return av_mallocz(nmemb * size);
yading@11: }
yading@11: 
yading@11: char *av_strdup(const char *s)
yading@11: {
yading@11:     char *ptr = NULL;
yading@11:     if (s) {
yading@11:         int len = strlen(s) + 1;
yading@11:         ptr = av_malloc(len);
yading@11:         if (ptr)
yading@11:             memcpy(ptr, s, len);
yading@11:     }
yading@11:     return ptr;
yading@11: }
yading@11: 
yading@11: /* add one element to a dynamic array */
yading@11: void av_dynarray_add(void *tab_ptr, int *nb_ptr, void *elem)
yading@11: {
yading@11:     /* see similar ffmpeg.c:grow_array() */
yading@11:     int nb, nb_alloc;
yading@11:     intptr_t *tab;
yading@11: 
yading@11:     nb = *nb_ptr;
yading@11:     tab = *(intptr_t**)tab_ptr;
yading@11:     if ((nb & (nb - 1)) == 0) {
yading@11:         if (nb == 0)
yading@11:             nb_alloc = 1;
yading@11:         else
yading@11:             nb_alloc = nb * 2;
yading@11:         tab = av_realloc(tab, nb_alloc * sizeof(intptr_t));
yading@11:         *(intptr_t**)tab_ptr = tab;
yading@11:     }
yading@11:     tab[nb++] = (intptr_t)elem;
yading@11:     *nb_ptr = nb;
yading@11: }
yading@11: 
yading@11: static void fill16(uint8_t *dst, int len)
yading@11: {
yading@11:     uint32_t v = AV_RN16(dst - 2);
yading@11: 
yading@11:     v |= v << 16;
yading@11: 
yading@11:     while (len >= 4) {
yading@11:         AV_WN32(dst, v);
yading@11:         dst += 4;
yading@11:         len -= 4;
yading@11:     }
yading@11: 
yading@11:     while (len--) {
yading@11:         *dst = dst[-2];
yading@11:         dst++;
yading@11:     }
yading@11: }
yading@11: 
yading@11: static void fill24(uint8_t *dst, int len)
yading@11: {
yading@11: #if HAVE_BIGENDIAN
yading@11:     uint32_t v = AV_RB24(dst - 3);
yading@11:     uint32_t a = v << 8  | v >> 16;
yading@11:     uint32_t b = v << 16 | v >> 8;
yading@11:     uint32_t c = v << 24 | v;
yading@11: #else
yading@11:     uint32_t v = AV_RL24(dst - 3);
yading@11:     uint32_t a = v       | v << 24;
yading@11:     uint32_t b = v >> 8  | v << 16;
yading@11:     uint32_t c = v >> 16 | v << 8;
yading@11: #endif
yading@11: 
yading@11:     while (len >= 12) {
yading@11:         AV_WN32(dst,     a);
yading@11:         AV_WN32(dst + 4, b);
yading@11:         AV_WN32(dst + 8, c);
yading@11:         dst += 12;
yading@11:         len -= 12;
yading@11:     }
yading@11: 
yading@11:     if (len >= 4) {
yading@11:         AV_WN32(dst, a);
yading@11:         dst += 4;
yading@11:         len -= 4;
yading@11:     }
yading@11: 
yading@11:     if (len >= 4) {
yading@11:         AV_WN32(dst, b);
yading@11:         dst += 4;
yading@11:         len -= 4;
yading@11:     }
yading@11: 
yading@11:     while (len--) {
yading@11:         *dst = dst[-3];
yading@11:         dst++;
yading@11:     }
yading@11: }
yading@11: 
yading@11: static void fill32(uint8_t *dst, int len)
yading@11: {
yading@11:     uint32_t v = AV_RN32(dst - 4);
yading@11: 
yading@11:     while (len >= 4) {
yading@11:         AV_WN32(dst, v);
yading@11:         dst += 4;
yading@11:         len -= 4;
yading@11:     }
yading@11: 
yading@11:     while (len--) {
yading@11:         *dst = dst[-4];
yading@11:         dst++;
yading@11:     }
yading@11: }
yading@11: 
yading@11: void av_memcpy_backptr(uint8_t *dst, int back, int cnt)
yading@11: {
yading@11:     const uint8_t *src = &dst[-back];
yading@11:     if (!back)
yading@11:         return;
yading@11: 
yading@11:     if (back == 1) {
yading@11:         memset(dst, *src, cnt);
yading@11:     } else if (back == 2) {
yading@11:         fill16(dst, cnt);
yading@11:     } else if (back == 3) {
yading@11:         fill24(dst, cnt);
yading@11:     } else if (back == 4) {
yading@11:         fill32(dst, cnt);
yading@11:     } else {
yading@11:         if (cnt >= 16) {
yading@11:             int blocklen = back;
yading@11:             while (cnt > blocklen) {
yading@11:                 memcpy(dst, src, blocklen);
yading@11:                 dst       += blocklen;
yading@11:                 cnt       -= blocklen;
yading@11:                 blocklen <<= 1;
yading@11:             }
yading@11:             memcpy(dst, src, cnt);
yading@11:             return;
yading@11:         }
yading@11:         if (cnt >= 8) {
yading@11:             AV_COPY32U(dst,     src);
yading@11:             AV_COPY32U(dst + 4, src + 4);
yading@11:             src += 8;
yading@11:             dst += 8;
yading@11:             cnt -= 8;
yading@11:         }
yading@11:         if (cnt >= 4) {
yading@11:             AV_COPY32U(dst, src);
yading@11:             src += 4;
yading@11:             dst += 4;
yading@11:             cnt -= 4;
yading@11:         }
yading@11:         if (cnt >= 2) {
yading@11:             AV_COPY16U(dst, src);
yading@11:             src += 2;
yading@11:             dst += 2;
yading@11:             cnt -= 2;
yading@11:         }
yading@11:         if (cnt)
yading@11:             *dst = *src;
yading@11:     }
yading@11: }
yading@11: