xref: /llvm-project/libc/AOR_v20.02/string/aarch64/strchr-sve.S (revision 0570de73c4822b6b0da4758eefca2a36f01c7445)
1/*
2 * strchr/strchrnul - find a character in a string
3 *
4 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
5 * See https://llvm.org/LICENSE.txt for license information.
6 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 */
8
9#if __ARM_FEATURE_SVE
10/* Assumptions:
11 *
12 * ARMv8-a, AArch64
13 * SVE Available.
14 */
15
16	.arch	armv8-a+sve
17	.text
18
19/* To build as strchrnul, define BUILD_STRCHRNUL before compiling this file.  */
20#ifdef BUILD_STRCHRNUL
21#define FUNC  __strchrnul_aarch64_sve
22#else
23#define FUNC  __strchr_aarch64_sve
24#endif
25
26	.globl	FUNC
27	.type	FUNC, %function
28	.p2align 4
29FUNC:
30	dup	z1.b, w1		/* replicate byte across vector */
31	setffr				/* initialize FFR */
32	ptrue	p1.b			/* all ones; loop invariant */
33
34	.p2align 4
35	/* Read a vector's worth of bytes, stopping on first fault.  */
360:	ldff1b	z0.b, p1/z, [x0, xzr]
37	rdffrs	p0.b, p1/z
38	b.nlast	2f
39
40	/* First fault did not fail: the whole vector is valid.
41	   Avoid depending on the contents of FFR beyond the branch.  */
42	incb	x0				/* speculate increment */
43	cmpeq	p2.b, p1/z, z0.b, z1.b		/* search for c */
44	cmpeq	p3.b, p1/z, z0.b, 0		/* search for 0 */
45	orrs	p4.b, p1/z, p2.b, p3.b		/* c | 0 */
46	b.none	0b
47	decb	x0				/* undo speculate */
48
49	/* Found C or 0.  */
501:	brka	p4.b, p1/z, p4.b	/* find first such */
51	sub	x0, x0, 1		/* adjust pointer for that byte */
52	incp	x0, p4.b
53#ifndef BUILD_STRCHRNUL
54	ptest	p4, p2.b		/* was first in c? */
55	csel	x0, xzr, x0, none	/* if there was no c, return null */
56#endif
57	ret
58
59	/* First fault failed: only some of the vector is valid.
60	   Perform the comparison only on the valid bytes.  */
612:	cmpeq	p2.b, p0/z, z0.b, z1.b		/* search for c */
62	cmpeq	p3.b, p0/z, z0.b, 0		/* search for 0 */
63	orrs	p4.b, p0/z, p2.b, p3.b		/* c | 0 */
64	b.any	1b
65
66	/* No C or 0 found.  Re-init FFR, increment, and loop.  */
67	setffr
68	incp	x0, p0.b
69	b	0b
70
71	.size	FUNC, . - FUNC
72#endif
73