1#include <string.h>
 2#include <stdint.h>
 3#include <limits.h>
 4
 5#ifdef __wasm_simd128__
 6#include <wasm_simd128.h>
 7#endif
 8
 9#define ALIGN (sizeof(size_t))
10#define ONES ((size_t)-1/UCHAR_MAX)
11#define HIGHS (ONES * (UCHAR_MAX/2+1))
12#define HASZERO(x) ((x)-ONES & ~(x) & HIGHS)
13
14char *__strchrnul(const char *s, int c)
15{
16	c = (unsigned char)c;
17	if (!c) return (char *)s + strlen(s);
18
19#if defined(__wasm_simd128__) && defined(__wasilibc_simd_string)
20	// Skip Clang 19 and Clang 20 which have a bug (llvm/llvm-project#146574)
21	// which results in an ICE when inline assembly is used with a vector result.
22#if __clang_major__ != 19 && __clang_major__ != 20
23	// Note that reading before/after the allocation of a pointer is UB in
24	// C, so inline assembly is used to generate the exact machine
25	// instruction we want with opaque semantics to the compiler to avoid
26	// the UB.
27	uintptr_t align = (uintptr_t)s % sizeof(v128_t);
28	uintptr_t addr = (uintptr_t)s - align;
29	v128_t vc = wasm_i8x16_splat(c);
30
31	for (;;) {
32		v128_t v;
33		__asm__ (
34			"local.get %1\n"
35			"v128.load 0\n"
36			"local.set %0\n"
37			: "=r"(v)
38			: "r"(addr)
39			: "memory");
40		const v128_t cmp = wasm_i8x16_eq(v, (v128_t){}) | wasm_i8x16_eq(v, vc);
41		// Bitmask is slow on AArch64, any_true is much faster.
42		if (wasm_v128_any_true(cmp)) {
43			// Clear the bits corresponding to align (little-endian)
44			// so we can count trailing zeros.
45			int mask = wasm_i8x16_bitmask(cmp) >> align << align;
46			// At least one bit will be set, unless align cleared them.
47			// Knowing this helps the compiler if it unrolls the loop.
48			__builtin_assume(mask || align);
49			// If the mask became zero because of align,
50			// it's as if we didn't find anything.
51			if (mask) {
52				// Find the offset of the first one bit (little-endian).
53				return (char *)s + (addr - (uintptr_t)s + __builtin_ctz(mask));
54			}
55		}
56		align = 0;
57		addr += sizeof(v128_t);
58	}
59#endif
60#endif
61
62#ifdef __GNUC__
63	typedef size_t __attribute__((__may_alias__)) word;
64	const word *w;
65	for (; (uintptr_t)s % ALIGN; s++)
66		if (!*s || *(unsigned char *)s == c) return (char *)s;
67	size_t k = ONES * c;
68	for (w = (void *)s; !HASZERO(*w) && !HASZERO(*w^k); w++);
69	s = (void *)w;
70#endif
71	for (; *s && *(unsigned char *)s != c; s++);
72	return (char *)s;
73}
74
75weak_alias(__strchrnul, strchrnul);