1 #include "usr_lib_define.h"
2 #include <string.h>
3 #include <stdint.h>
4
memset(void * dest,int c,size_t n)5 _LIBC_TEXT_SECTION void *memset(void *dest, int c, size_t n)
6 {
7 unsigned char *s = dest;
8 size_t k;
9
10 /* Fill head and tail with minimal branching. Each
11 * conditional ensures that all the subsequently used
12 * offsets are well-defined and in the dest region. */
13
14 if (!n) return dest;
15 s[0] = c;
16 s[n-1] = c;
17 if (n <= 2) return dest;
18 s[1] = c;
19 s[2] = c;
20 s[n-2] = c;
21 s[n-3] = c;
22 if (n <= 6) return dest;
23 s[3] = c;
24 s[n-4] = c;
25 if (n <= 8) return dest;
26
27 /* Advance pointer to align it at a 4-byte boundary,
28 * and truncate n to a multiple of 4. The previous code
29 * already took care of any head/tail that get cut off
30 * by the alignment. */
31
32 k = -(uintptr_t)s & 3;
33 s += k;
34 n -= k;
35 n &= -4;
36
37 #ifdef __GNUC__
38 typedef uint32_t __attribute__((__may_alias__)) u32;
39 typedef uint64_t __attribute__((__may_alias__)) u64;
40
41 u32 c32 = ((u32)-1)/255 * (unsigned char)c;
42
43 /* In preparation to copy 32 bytes at a time, aligned on
44 * an 8-byte bounary, fill head/tail up to 28 bytes each.
45 * As in the initial byte-based head/tail fill, each
46 * conditional below ensures that the subsequent offsets
47 * are valid (e.g. !(n<=24) implies n>=28). */
48
49 *(u32 *)(s+0) = c32;
50 *(u32 *)(s+n-4) = c32;
51 if (n <= 8) return dest;
52 *(u32 *)(s+4) = c32;
53 *(u32 *)(s+8) = c32;
54 *(u32 *)(s+n-12) = c32;
55 *(u32 *)(s+n-8) = c32;
56 if (n <= 24) return dest;
57 *(u32 *)(s+12) = c32;
58 *(u32 *)(s+16) = c32;
59 *(u32 *)(s+20) = c32;
60 *(u32 *)(s+24) = c32;
61 *(u32 *)(s+n-28) = c32;
62 *(u32 *)(s+n-24) = c32;
63 *(u32 *)(s+n-20) = c32;
64 *(u32 *)(s+n-16) = c32;
65
66 /* Align to a multiple of 8 so we can fill 64 bits at a time,
67 * and avoid writing the same bytes twice as much as is
68 * practical without introducing additional branching. */
69
70 k = 24 + ((uintptr_t)s & 4);
71 s += k;
72 n -= k;
73
74 /* If this loop is reached, 28 tail bytes have already been
75 * filled, so any remainder when n drops below 32 can be
76 * safely ignored. */
77
78 u64 c64 = c32 | ((u64)c32 << 32);
79 for (; n >= 32; n-=32, s+=32) {
80 *(u64 *)(s+0) = c64;
81 *(u64 *)(s+8) = c64;
82 *(u64 *)(s+16) = c64;
83 *(u64 *)(s+24) = c64;
84 }
85 #else
86 /* Pure C fallback with no aliasing violations. */
87 for (; n; n--, s++) *s = c;
88 #endif
89
90 return dest;
91 }
92