Michal Simek | 322ae8e | 2009-03-27 14:25:21 +0100 | [diff] [blame] | 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 memmove. |
| 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, unsuccesfully, 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 | #include <linux/string.h> |
| 32 | |
| 33 | #ifdef __HAVE_ARCH_MEMMOVE |
| 34 | void *memmove(void *v_dst, const void *v_src, __kernel_size_t c) |
| 35 | { |
| 36 | const char *src = v_src; |
| 37 | char *dst = v_dst; |
| 38 | |
| 39 | #ifdef CONFIG_OPT_LIB_FUNCTION |
| 40 | const uint32_t *i_src; |
| 41 | uint32_t *i_dst; |
| 42 | #endif |
| 43 | |
| 44 | if (!c) |
| 45 | return v_dst; |
| 46 | |
| 47 | /* Use memcpy when source is higher than dest */ |
| 48 | if (v_dst <= v_src) |
| 49 | return memcpy(v_dst, v_src, c); |
| 50 | |
| 51 | #ifndef CONFIG_OPT_LIB_FUNCTION |
| 52 | /* copy backwards, from end to beginning */ |
| 53 | src += c; |
| 54 | dst += c; |
| 55 | |
| 56 | /* Simple, byte oriented memmove. */ |
| 57 | while (c--) |
| 58 | *--dst = *--src; |
| 59 | |
| 60 | return v_dst; |
| 61 | #else |
| 62 | /* The following code tries to optimize the copy by using unsigned |
| 63 | * alignment. This will work fine if both source and destination are |
| 64 | * aligned on the same boundary. However, if they are aligned on |
| 65 | * different boundaries shifts will be necessary. This might result in |
| 66 | * bad performance on MicroBlaze systems without a barrel shifter. |
| 67 | */ |
| 68 | /* FIXME this part needs more test */ |
| 69 | /* Do a descending copy - this is a bit trickier! */ |
| 70 | dst += c; |
| 71 | src += c; |
| 72 | |
| 73 | if (c >= 4) { |
| 74 | unsigned value, buf_hold; |
| 75 | |
| 76 | /* Align the destination to a word boundry. */ |
| 77 | /* This is done in an endian independant manner. */ |
| 78 | |
| 79 | switch ((unsigned long)dst & 3) { |
| 80 | case 3: |
| 81 | *--dst = *--src; |
| 82 | --c; |
| 83 | case 2: |
| 84 | *--dst = *--src; |
| 85 | --c; |
| 86 | case 1: |
| 87 | *--dst = *--src; |
| 88 | --c; |
| 89 | } |
| 90 | |
| 91 | i_dst = (void *)dst; |
| 92 | /* Choose a copy scheme based on the source */ |
| 93 | /* alignment relative to dstination. */ |
| 94 | switch ((unsigned long)src & 3) { |
| 95 | case 0x0: /* Both byte offsets are aligned */ |
| 96 | |
| 97 | i_src = (const void *)src; |
| 98 | |
| 99 | for (; c >= 4; c -= 4) |
| 100 | *--i_dst = *--i_src; |
| 101 | |
| 102 | src = (const void *)i_src; |
| 103 | break; |
| 104 | case 0x1: /* Unaligned - Off by 1 */ |
| 105 | /* Word align the source */ |
| 106 | i_src = (const void *) (((unsigned)src + 4) & ~3); |
| 107 | |
| 108 | /* Load the holding buffer */ |
| 109 | buf_hold = *--i_src >> 24; |
| 110 | |
| 111 | for (; c >= 4; c -= 4) { |
| 112 | value = *--i_src; |
| 113 | *--i_dst = buf_hold << 8 | value; |
| 114 | buf_hold = value >> 24; |
| 115 | } |
| 116 | |
| 117 | /* Realign the source */ |
| 118 | src = (const void *)i_src; |
| 119 | src += 1; |
| 120 | break; |
| 121 | case 0x2: /* Unaligned - Off by 2 */ |
| 122 | /* Word align the source */ |
| 123 | i_src = (const void *) (((unsigned)src + 4) & ~3); |
| 124 | |
| 125 | /* Load the holding buffer */ |
| 126 | buf_hold = *--i_src >> 16; |
| 127 | |
| 128 | for (; c >= 4; c -= 4) { |
| 129 | value = *--i_src; |
| 130 | *--i_dst = buf_hold << 16 | value; |
| 131 | buf_hold = value >> 16; |
| 132 | } |
| 133 | |
| 134 | /* Realign the source */ |
| 135 | src = (const void *)i_src; |
| 136 | src += 2; |
| 137 | break; |
| 138 | case 0x3: /* Unaligned - Off by 3 */ |
| 139 | /* Word align the source */ |
| 140 | i_src = (const void *) (((unsigned)src + 4) & ~3); |
| 141 | |
| 142 | /* Load the holding buffer */ |
| 143 | buf_hold = *--i_src >> 8; |
| 144 | |
| 145 | for (; c >= 4; c -= 4) { |
| 146 | value = *--i_src; |
| 147 | *--i_dst = buf_hold << 24 | value; |
| 148 | buf_hold = value >> 8; |
| 149 | } |
| 150 | |
| 151 | /* Realign the source */ |
| 152 | src = (const void *)i_src; |
| 153 | src += 3; |
| 154 | break; |
| 155 | } |
| 156 | dst = (void *)i_dst; |
| 157 | } |
| 158 | |
| 159 | /* simple fast copy, ... unless a cache boundry is crossed */ |
| 160 | /* Finish off any remaining bytes */ |
| 161 | switch (c) { |
| 162 | case 4: |
| 163 | *--dst = *--src; |
| 164 | case 3: |
| 165 | *--dst = *--src; |
| 166 | case 2: |
| 167 | *--dst = *--src; |
| 168 | case 1: |
| 169 | *--dst = *--src; |
| 170 | } |
| 171 | return v_dst; |
| 172 | #endif |
| 173 | } |
| 174 | EXPORT_SYMBOL(memmove); |
| 175 | #endif /* __HAVE_ARCH_MEMMOVE */ |