arch/microblaze/lib/memcpy.c

   1 /*
   2  * Copyright (C) 2008-2009 Michal Simek <monstr@monstr.eu>
   3  * Copyright (C) 2008-2009 PetaLogix
   4  * Copyright (C) 2007 John Williams
   5  *
   6  * Reasonably optimised generic C-code for memcpy on Microblaze
   7  * This is generic C code to do efficient, alignment-aware memcpy.
   8  *
   9  * It is based on demo code originally Copyright 2001 by Intel Corp, taken from
  10  * http://www.embedded.com/showArticle.jhtml?articleID=19205567
  11  *
  12  * Attempts were made, unsuccessfully, to contact the original
  13  * author of this code (Michael Morrow, Intel).  Below is the original
  14  * copyright notice.
  15  *
  16  * This software has been developed by Intel Corporation.
  17  * Intel specifically disclaims all warranties, express or
  18  * implied, and all liability, including consequential and
  19  * other indirect damages, for the use of this program, including
  20  * liability for infringement of any proprietary rights,
  21  * and including the warranties of merchantability and fitness
  22  * for a particular purpose. Intel does not assume any
  23  * responsibility for and errors which may appear in this program
  24  * not any responsibility to update it.
  25  */
  26
  27 #include <linux/types.h>
  28 #include <linux/stddef.h>
  29 #include <linux/compiler.h>
  30 #include <linux/module.h>
  31
  32 #include <linux/string.h>
  33 #include <asm/system.h>
  34
  35 #ifdef __HAVE_ARCH_MEMCPY
  36 void *memcpy(void *v_dst, const void *v_src, __kernel_size_t c)
  37 {
  38         const char *src = v_src;
  39         char *dst = v_dst;
  40 #ifndef CONFIG_OPT_LIB_FUNCTION
  41         /* Simple, byte oriented memcpy. */
  42         while (c--)
  43                 *dst++ = *src++;
  44
  45         return v_dst;
  46 #else
  47         /* The following code tries to optimize the copy by using unsigned
  48          * alignment. This will work fine if both source and destination are
  49          * aligned on the same boundary. However, if they are aligned on
  50          * different boundaries shifts will be necessary. This might result in
  51          * bad performance on MicroBlaze systems without a barrel shifter.
  52          */
  53         const uint32_t *i_src;
  54         uint32_t *i_dst;
  55
  56         if (likely(c >= 4)) {
  57                 unsigned  value, buf_hold;
  58
  59                 /* Align the dstination to a word boundry. */
  60                 /* This is done in an endian independant manner. */
  61                 switch ((unsigned long)dst & 3) {
  62                 case 1:
  63                         *dst++ = *src++;
  64                         --c;
  65                 case 2:
  66                         *dst++ = *src++;
  67                         --c;
  68                 case 3:
  69                         *dst++ = *src++;
  70                         --c;
  71                 }
  72
  73                 i_dst = (void *)dst;
  74
  75                 /* Choose a copy scheme based on the source */
  76                 /* alignment relative to dstination. */
  77                 switch ((unsigned long)src & 3) {
  78                 case 0x0:       /* Both byte offsets are aligned */
  79                         i_src  = (const void *)src;
  80
  81                         for (; c >= 4; c -= 4)
  82                                 *i_dst++ = *i_src++;
  83
  84                         src  = (const void *)i_src;
  85                         break;
  86                 case 0x1:       /* Unaligned - Off by 1 */
  87                         /* Word align the source */
  88                         i_src = (const void *) ((unsigned)src & ~3);
  89
  90                         /* Load the holding buffer */
  91                         buf_hold = *i_src++ << 8;
  92
  93                         for (; c >= 4; c -= 4) {
  94                                 value = *i_src++;
  95                                 *i_dst++ = buf_hold | value >> 24;
  96                                 buf_hold = value << 8;
  97                         }
  98
  99                         /* Realign the source */
 100                         src = (const void *)i_src;
 101                         src -= 3;
 102                         break;
 103                 case 0x2:       /* Unaligned - Off by 2 */
 104                         /* Word align the source */
 105                         i_src = (const void *) ((unsigned)src & ~3);
 106
 107                         /* Load the holding buffer */
 108                         buf_hold = *i_src++ << 16;
 109
 110                         for (; c >= 4; c -= 4) {
 111                                 value = *i_src++;
 112                                 *i_dst++ = buf_hold | value >> 16;
 113                                 buf_hold = value << 16;
 114                         }
 115
 116                         /* Realign the source */
 117                         src = (const void *)i_src;
 118                         src -= 2;
 119                         break;
 120                 case 0x3:       /* Unaligned - Off by 3 */
 121                         /* Word align the source */
 122                         i_src = (const void *) ((unsigned)src & ~3);
 123
 124                         /* Load the holding buffer */
 125                         buf_hold = *i_src++ << 24;
 126
 127                         for (; c >= 4; c -= 4) {
 128                                 value = *i_src++;
 129                                 *i_dst++ = buf_hold | value >> 8;
 130                                 buf_hold = value << 24;
 131                         }
 132
 133                         /* Realign the source */
 134                         src = (const void *)i_src;
 135                         src -= 1;
 136                         break;
 137                 }
 138                 dst = (void *)i_dst;
 139         }
 140
 141         /* Finish off any remaining bytes */
 142         /* simple fast copy, ... unless a cache boundry is crossed */
 143         switch (c) {
 144         case 3:
 145                 *dst++ = *src++;
 146         case 2:
 147                 *dst++ = *src++;
 148         case 1:
 149                 *dst++ = *src++;
 150         }
 151
 152         return v_dst;
 153 #endif
 154 }
 155 EXPORT_SYMBOL(memcpy);
 156 #endif /* __HAVE_ARCH_MEMCPY */