1/*
2 * strchr - find a character in a string
3 *
4 * Copyright (c) 2020-2022, Arm Limited.
5 * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
6 */
7
8/* Assumptions:
9 *
10 * ARMv8-a, AArch64, Advanced SIMD.
11 * MTE compatible.
12 */
13
14#include "asmdefs.h"
15
16#define srcin		x0
17#define chrin		w1
18#define result		x0
19
20#define src		x2
21#define tmp1		x1
22#define tmp2		x3
23
24#define vrepchr		v0
25#define vdata		v1
26#define qdata		q1
27#define vhas_nul	v2
28#define vhas_chr	v3
29#define vrepmask	v4
30#define vend		v5
31#define dend		d5
32
33/* Core algorithm.
34
35   For each 16-byte chunk we calculate a 64-bit syndrome value with four bits
36   per byte. Bits 0-1 are set if the relevant byte matched the requested
37   character, bits 2-3 are set if the byte is NUL or matched. Count trailing
38   zeroes gives the position of the matching byte if it is a multiple of 4.
39   If it is not a multiple of 4, there was no match.  */
40
41ENTRY (__strchr_aarch64_mte)
42	PTR_ARG (0)
43	bic	src, srcin, 15
44	dup	vrepchr.16b, chrin
45	ld1	{vdata.16b}, [src]
46	movi	vrepmask.16b, 0x33
47	cmeq	vhas_nul.16b, vdata.16b, 0
48	cmeq	vhas_chr.16b, vdata.16b, vrepchr.16b
49	bit	vhas_nul.16b, vhas_chr.16b, vrepmask.16b
50	lsl	tmp2, srcin, 2
51	shrn	vend.8b, vhas_nul.8h, 4		/* 128->64 */
52	fmov	tmp1, dend
53	lsr	tmp1, tmp1, tmp2
54	cbz	tmp1, L(loop)
55
56	rbit	tmp1, tmp1
57	clz	tmp1, tmp1
58	/* Tmp1 is an even multiple of 2 if the target character was
59	   found first. Otherwise we've found the end of string.  */
60	tst	tmp1, 2
61	add	result, srcin, tmp1, lsr 2
62	csel	result, result, xzr, eq
63	ret
64
65	.p2align 4
66L(loop):
67	ldr	qdata, [src, 16]
68	cmeq	vhas_chr.16b, vdata.16b, vrepchr.16b
69	cmhs	vhas_nul.16b, vhas_chr.16b, vdata.16b
70	umaxp	vend.16b, vhas_nul.16b, vhas_nul.16b
71	fmov	tmp1, dend
72	cbnz	tmp1, L(end)
73	ldr	qdata, [src, 32]!
74	cmeq	vhas_chr.16b, vdata.16b, vrepchr.16b
75	cmhs	vhas_nul.16b, vhas_chr.16b, vdata.16b
76	umaxp	vend.16b, vhas_nul.16b, vhas_nul.16b
77	fmov	tmp1, dend
78	cbz	tmp1, L(loop)
79	sub	src, src, 16
80L(end):
81
82#ifdef __AARCH64EB__
83	bif	vhas_nul.16b, vhas_chr.16b, vrepmask.16b
84	shrn	vend.8b, vhas_nul.8h, 4		/* 128->64 */
85	fmov	tmp1, dend
86#else
87	bit	vhas_nul.16b, vhas_chr.16b, vrepmask.16b
88	shrn	vend.8b, vhas_nul.8h, 4		/* 128->64 */
89	fmov	tmp1, dend
90	rbit	tmp1, tmp1
91#endif
92	add	src, src, 16
93	clz	tmp1, tmp1
94	/* Tmp1 is a multiple of 4 if the target character was found.  */
95	tst	tmp1, 2
96	add	result, src, tmp1, lsr 2
97	csel	result, result, xzr, eq
98	ret
99
100END (__strchr_aarch64_mte)
101
102