]>
Commit | Line | Data |
---|---|---|
322ae8eb MS |
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 | * | |
af901ca1 | 12 | * Attempts were made, unsuccessfully, to contact the original |
322ae8eb MS |
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 | |
93e2e851 | 34 | #ifndef CONFIG_OPT_LIB_FUNCTION |
322ae8eb MS |
35 | void *memmove(void *v_dst, const void *v_src, __kernel_size_t c) |
36 | { | |
37 | const char *src = v_src; | |
38 | char *dst = v_dst; | |
39 | ||
322ae8eb MS |
40 | if (!c) |
41 | return v_dst; | |
42 | ||
43 | /* Use memcpy when source is higher than dest */ | |
44 | if (v_dst <= v_src) | |
45 | return memcpy(v_dst, v_src, c); | |
46 | ||
322ae8eb MS |
47 | /* copy backwards, from end to beginning */ |
48 | src += c; | |
49 | dst += c; | |
50 | ||
51 | /* Simple, byte oriented memmove. */ | |
52 | while (c--) | |
53 | *--dst = *--src; | |
54 | ||
55 | return v_dst; | |
93e2e851 MS |
56 | } |
57 | #else /* CONFIG_OPT_LIB_FUNCTION */ | |
58 | void *memmove(void *v_dst, const void *v_src, __kernel_size_t c) | |
59 | { | |
60 | const char *src = v_src; | |
61 | char *dst = v_dst; | |
62 | const uint32_t *i_src; | |
63 | uint32_t *i_dst; | |
64 | ||
65 | if (!c) | |
66 | return v_dst; | |
67 | ||
68 | /* Use memcpy when source is higher than dest */ | |
69 | if (v_dst <= v_src) | |
70 | return memcpy(v_dst, v_src, c); | |
71 | ||
322ae8eb MS |
72 | /* The following code tries to optimize the copy by using unsigned |
73 | * alignment. This will work fine if both source and destination are | |
74 | * aligned on the same boundary. However, if they are aligned on | |
75 | * different boundaries shifts will be necessary. This might result in | |
76 | * bad performance on MicroBlaze systems without a barrel shifter. | |
77 | */ | |
78 | /* FIXME this part needs more test */ | |
79 | /* Do a descending copy - this is a bit trickier! */ | |
80 | dst += c; | |
81 | src += c; | |
82 | ||
83 | if (c >= 4) { | |
84 | unsigned value, buf_hold; | |
85 | ||
25985edc LDM |
86 | /* Align the destination to a word boundary. */ |
87 | /* This is done in an endian independent manner. */ | |
322ae8eb MS |
88 | |
89 | switch ((unsigned long)dst & 3) { | |
90 | case 3: | |
91 | *--dst = *--src; | |
92 | --c; | |
93 | case 2: | |
94 | *--dst = *--src; | |
95 | --c; | |
96 | case 1: | |
97 | *--dst = *--src; | |
98 | --c; | |
99 | } | |
100 | ||
101 | i_dst = (void *)dst; | |
102 | /* Choose a copy scheme based on the source */ | |
103 | /* alignment relative to dstination. */ | |
104 | switch ((unsigned long)src & 3) { | |
105 | case 0x0: /* Both byte offsets are aligned */ | |
106 | ||
107 | i_src = (const void *)src; | |
108 | ||
109 | for (; c >= 4; c -= 4) | |
110 | *--i_dst = *--i_src; | |
111 | ||
112 | src = (const void *)i_src; | |
113 | break; | |
114 | case 0x1: /* Unaligned - Off by 1 */ | |
115 | /* Word align the source */ | |
116 | i_src = (const void *) (((unsigned)src + 4) & ~3); | |
1180b28c | 117 | #ifndef __MICROBLAZEEL__ |
322ae8eb MS |
118 | /* Load the holding buffer */ |
119 | buf_hold = *--i_src >> 24; | |
120 | ||
121 | for (; c >= 4; c -= 4) { | |
122 | value = *--i_src; | |
123 | *--i_dst = buf_hold << 8 | value; | |
124 | buf_hold = value >> 24; | |
125 | } | |
1180b28c MS |
126 | #else |
127 | /* Load the holding buffer */ | |
128 | buf_hold = (*--i_src & 0xFF) << 24; | |
322ae8eb | 129 | |
1180b28c MS |
130 | for (; c >= 4; c -= 4) { |
131 | value = *--i_src; | |
132 | *--i_dst = buf_hold | ((value & 0xFFFFFF00)>>8); | |
133 | buf_hold = (value & 0xFF) << 24; | |
134 | } | |
135 | #endif | |
322ae8eb MS |
136 | /* Realign the source */ |
137 | src = (const void *)i_src; | |
138 | src += 1; | |
139 | break; | |
140 | case 0x2: /* Unaligned - Off by 2 */ | |
141 | /* Word align the source */ | |
142 | i_src = (const void *) (((unsigned)src + 4) & ~3); | |
1180b28c | 143 | #ifndef __MICROBLAZEEL__ |
322ae8eb MS |
144 | /* Load the holding buffer */ |
145 | buf_hold = *--i_src >> 16; | |
146 | ||
147 | for (; c >= 4; c -= 4) { | |
148 | value = *--i_src; | |
149 | *--i_dst = buf_hold << 16 | value; | |
150 | buf_hold = value >> 16; | |
151 | } | |
1180b28c MS |
152 | #else |
153 | /* Load the holding buffer */ | |
154 | buf_hold = (*--i_src & 0xFFFF) << 16; | |
322ae8eb | 155 | |
1180b28c MS |
156 | for (; c >= 4; c -= 4) { |
157 | value = *--i_src; | |
158 | *--i_dst = buf_hold | ((value & 0xFFFF0000)>>16); | |
159 | buf_hold = (value & 0xFFFF) << 16; | |
160 | } | |
161 | #endif | |
322ae8eb MS |
162 | /* Realign the source */ |
163 | src = (const void *)i_src; | |
164 | src += 2; | |
165 | break; | |
166 | case 0x3: /* Unaligned - Off by 3 */ | |
167 | /* Word align the source */ | |
168 | i_src = (const void *) (((unsigned)src + 4) & ~3); | |
1180b28c | 169 | #ifndef __MICROBLAZEEL__ |
322ae8eb MS |
170 | /* Load the holding buffer */ |
171 | buf_hold = *--i_src >> 8; | |
172 | ||
173 | for (; c >= 4; c -= 4) { | |
174 | value = *--i_src; | |
175 | *--i_dst = buf_hold << 24 | value; | |
176 | buf_hold = value >> 8; | |
177 | } | |
1180b28c MS |
178 | #else |
179 | /* Load the holding buffer */ | |
180 | buf_hold = (*--i_src & 0xFFFFFF) << 8; | |
322ae8eb | 181 | |
1180b28c MS |
182 | for (; c >= 4; c -= 4) { |
183 | value = *--i_src; | |
184 | *--i_dst = buf_hold | ((value & 0xFF000000)>> 24); | |
473ff660 | 185 | buf_hold = (value & 0xFFFFFF) << 8; |
1180b28c MS |
186 | } |
187 | #endif | |
322ae8eb MS |
188 | /* Realign the source */ |
189 | src = (const void *)i_src; | |
190 | src += 3; | |
191 | break; | |
192 | } | |
193 | dst = (void *)i_dst; | |
194 | } | |
195 | ||
25985edc | 196 | /* simple fast copy, ... unless a cache boundary is crossed */ |
322ae8eb MS |
197 | /* Finish off any remaining bytes */ |
198 | switch (c) { | |
199 | case 4: | |
200 | *--dst = *--src; | |
201 | case 3: | |
202 | *--dst = *--src; | |
203 | case 2: | |
204 | *--dst = *--src; | |
205 | case 1: | |
206 | *--dst = *--src; | |
207 | } | |
208 | return v_dst; | |
322ae8eb | 209 | } |
93e2e851 | 210 | #endif /* CONFIG_OPT_LIB_FUNCTION */ |
322ae8eb MS |
211 | EXPORT_SYMBOL(memmove); |
212 | #endif /* __HAVE_ARCH_MEMMOVE */ |