]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2002 Mark Debbage (Mark.Debbage@superh.com) | |
3 | * | |
4 | * May be copied or modified under the terms of the GNU General Public | |
5 | * License. See linux/COPYING for more information. | |
6 | * | |
7 | */ | |
8 | ||
1da177e4 LT |
9 | #include <linux/types.h> |
10 | #include <asm/string.h> | |
11 | ||
12 | // This is a simplistic optimization of memcpy to increase the | |
13 | // granularity of access beyond one byte using aligned | |
14 | // loads and stores. This is not an optimal implementation | |
15 | // for SH-5 (especially with regard to prefetching and the cache), | |
16 | // and a better version should be provided later ... | |
17 | ||
18 | void *memcpy(void *dest, const void *src, size_t count) | |
19 | { | |
20 | char *d = (char *) dest, *s = (char *) src; | |
21 | ||
22 | if (count >= 32) { | |
23 | int i = 8 - (((unsigned long) d) & 0x7); | |
24 | ||
25 | if (i != 8) | |
26 | while (i-- && count--) { | |
27 | *d++ = *s++; | |
28 | } | |
29 | ||
30 | if (((((unsigned long) d) & 0x7) == 0) && | |
31 | ((((unsigned long) s) & 0x7) == 0)) { | |
32 | while (count >= 32) { | |
33 | unsigned long long t1, t2, t3, t4; | |
34 | t1 = *(unsigned long long *) (s); | |
35 | t2 = *(unsigned long long *) (s + 8); | |
36 | t3 = *(unsigned long long *) (s + 16); | |
37 | t4 = *(unsigned long long *) (s + 24); | |
38 | *(unsigned long long *) (d) = t1; | |
39 | *(unsigned long long *) (d + 8) = t2; | |
40 | *(unsigned long long *) (d + 16) = t3; | |
41 | *(unsigned long long *) (d + 24) = t4; | |
42 | d += 32; | |
43 | s += 32; | |
44 | count -= 32; | |
45 | } | |
46 | while (count >= 8) { | |
47 | *(unsigned long long *) d = | |
48 | *(unsigned long long *) s; | |
49 | d += 8; | |
50 | s += 8; | |
51 | count -= 8; | |
52 | } | |
53 | } | |
54 | ||
55 | if (((((unsigned long) d) & 0x3) == 0) && | |
56 | ((((unsigned long) s) & 0x3) == 0)) { | |
57 | while (count >= 4) { | |
58 | *(unsigned long *) d = *(unsigned long *) s; | |
59 | d += 4; | |
60 | s += 4; | |
61 | count -= 4; | |
62 | } | |
63 | } | |
64 | ||
65 | if (((((unsigned long) d) & 0x1) == 0) && | |
66 | ((((unsigned long) s) & 0x1) == 0)) { | |
67 | while (count >= 2) { | |
68 | *(unsigned short *) d = *(unsigned short *) s; | |
69 | d += 2; | |
70 | s += 2; | |
71 | count -= 2; | |
72 | } | |
73 | } | |
74 | } | |
75 | ||
76 | while (count--) { | |
77 | *d++ = *s++; | |
78 | } | |
79 | ||
80 | return d; | |
81 | } |