xref: /aosp_15_r20/external/musl/src/string/memset.c (revision c9945492fdd68bbe62686c5b452b4dc1be3f8453)
1*c9945492SAndroid Build Coastguard Worker #include <string.h>
2*c9945492SAndroid Build Coastguard Worker #include <stdint.h>
3*c9945492SAndroid Build Coastguard Worker 
memset(void * dest,int c,size_t n)4*c9945492SAndroid Build Coastguard Worker void *memset(void *dest, int c, size_t n)
5*c9945492SAndroid Build Coastguard Worker {
6*c9945492SAndroid Build Coastguard Worker 	unsigned char *s = dest;
7*c9945492SAndroid Build Coastguard Worker 	size_t k;
8*c9945492SAndroid Build Coastguard Worker 
9*c9945492SAndroid Build Coastguard Worker 	/* Fill head and tail with minimal branching. Each
10*c9945492SAndroid Build Coastguard Worker 	 * conditional ensures that all the subsequently used
11*c9945492SAndroid Build Coastguard Worker 	 * offsets are well-defined and in the dest region. */
12*c9945492SAndroid Build Coastguard Worker 
13*c9945492SAndroid Build Coastguard Worker 	if (!n) return dest;
14*c9945492SAndroid Build Coastguard Worker 	s[0] = c;
15*c9945492SAndroid Build Coastguard Worker 	s[n-1] = c;
16*c9945492SAndroid Build Coastguard Worker 	if (n <= 2) return dest;
17*c9945492SAndroid Build Coastguard Worker 	s[1] = c;
18*c9945492SAndroid Build Coastguard Worker 	s[2] = c;
19*c9945492SAndroid Build Coastguard Worker 	s[n-2] = c;
20*c9945492SAndroid Build Coastguard Worker 	s[n-3] = c;
21*c9945492SAndroid Build Coastguard Worker 	if (n <= 6) return dest;
22*c9945492SAndroid Build Coastguard Worker 	s[3] = c;
23*c9945492SAndroid Build Coastguard Worker 	s[n-4] = c;
24*c9945492SAndroid Build Coastguard Worker 	if (n <= 8) return dest;
25*c9945492SAndroid Build Coastguard Worker 
26*c9945492SAndroid Build Coastguard Worker 	/* Advance pointer to align it at a 4-byte boundary,
27*c9945492SAndroid Build Coastguard Worker 	 * and truncate n to a multiple of 4. The previous code
28*c9945492SAndroid Build Coastguard Worker 	 * already took care of any head/tail that get cut off
29*c9945492SAndroid Build Coastguard Worker 	 * by the alignment. */
30*c9945492SAndroid Build Coastguard Worker 
31*c9945492SAndroid Build Coastguard Worker 	k = -(uintptr_t)s & 3;
32*c9945492SAndroid Build Coastguard Worker 	s += k;
33*c9945492SAndroid Build Coastguard Worker 	n -= k;
34*c9945492SAndroid Build Coastguard Worker 	n &= -4;
35*c9945492SAndroid Build Coastguard Worker 
36*c9945492SAndroid Build Coastguard Worker #ifdef __GNUC__
37*c9945492SAndroid Build Coastguard Worker 	typedef uint32_t __attribute__((__may_alias__)) u32;
38*c9945492SAndroid Build Coastguard Worker 	typedef uint64_t __attribute__((__may_alias__)) u64;
39*c9945492SAndroid Build Coastguard Worker 
40*c9945492SAndroid Build Coastguard Worker 	u32 c32 = ((u32)-1)/255 * (unsigned char)c;
41*c9945492SAndroid Build Coastguard Worker 
42*c9945492SAndroid Build Coastguard Worker 	/* In preparation to copy 32 bytes at a time, aligned on
43*c9945492SAndroid Build Coastguard Worker 	 * an 8-byte bounary, fill head/tail up to 28 bytes each.
44*c9945492SAndroid Build Coastguard Worker 	 * As in the initial byte-based head/tail fill, each
45*c9945492SAndroid Build Coastguard Worker 	 * conditional below ensures that the subsequent offsets
46*c9945492SAndroid Build Coastguard Worker 	 * are valid (e.g. !(n<=24) implies n>=28). */
47*c9945492SAndroid Build Coastguard Worker 
48*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+0) = c32;
49*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+n-4) = c32;
50*c9945492SAndroid Build Coastguard Worker 	if (n <= 8) return dest;
51*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+4) = c32;
52*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+8) = c32;
53*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+n-12) = c32;
54*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+n-8) = c32;
55*c9945492SAndroid Build Coastguard Worker 	if (n <= 24) return dest;
56*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+12) = c32;
57*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+16) = c32;
58*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+20) = c32;
59*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+24) = c32;
60*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+n-28) = c32;
61*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+n-24) = c32;
62*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+n-20) = c32;
63*c9945492SAndroid Build Coastguard Worker 	*(u32 *)(s+n-16) = c32;
64*c9945492SAndroid Build Coastguard Worker 
65*c9945492SAndroid Build Coastguard Worker 	/* Align to a multiple of 8 so we can fill 64 bits at a time,
66*c9945492SAndroid Build Coastguard Worker 	 * and avoid writing the same bytes twice as much as is
67*c9945492SAndroid Build Coastguard Worker 	 * practical without introducing additional branching. */
68*c9945492SAndroid Build Coastguard Worker 
69*c9945492SAndroid Build Coastguard Worker 	k = 24 + ((uintptr_t)s & 4);
70*c9945492SAndroid Build Coastguard Worker 	s += k;
71*c9945492SAndroid Build Coastguard Worker 	n -= k;
72*c9945492SAndroid Build Coastguard Worker 
73*c9945492SAndroid Build Coastguard Worker 	/* If this loop is reached, 28 tail bytes have already been
74*c9945492SAndroid Build Coastguard Worker 	 * filled, so any remainder when n drops below 32 can be
75*c9945492SAndroid Build Coastguard Worker 	 * safely ignored. */
76*c9945492SAndroid Build Coastguard Worker 
77*c9945492SAndroid Build Coastguard Worker 	u64 c64 = c32 | ((u64)c32 << 32);
78*c9945492SAndroid Build Coastguard Worker 	for (; n >= 32; n-=32, s+=32) {
79*c9945492SAndroid Build Coastguard Worker 		*(u64 *)(s+0) = c64;
80*c9945492SAndroid Build Coastguard Worker 		*(u64 *)(s+8) = c64;
81*c9945492SAndroid Build Coastguard Worker 		*(u64 *)(s+16) = c64;
82*c9945492SAndroid Build Coastguard Worker 		*(u64 *)(s+24) = c64;
83*c9945492SAndroid Build Coastguard Worker 	}
84*c9945492SAndroid Build Coastguard Worker #else
85*c9945492SAndroid Build Coastguard Worker 	/* Pure C fallback with no aliasing violations. */
86*c9945492SAndroid Build Coastguard Worker 	for (; n; n--, s++) *s = c;
87*c9945492SAndroid Build Coastguard Worker #endif
88*c9945492SAndroid Build Coastguard Worker 
89*c9945492SAndroid Build Coastguard Worker 	return dest;
90*c9945492SAndroid Build Coastguard Worker }
91