xref: /freebsd/contrib/arm-optimized-routines/string/aarch64/memchr-mte.S (revision dd21556857e8d40f66bf5ad54754d9d52669ebf7)
1/*
2 * memchr - find a character in a memory zone
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 cntin		x2
19#define result		x0
20
21#define src		x3
22#define cntrem		x4
23#define synd		x5
24#define shift		x6
25#define	tmp		x7
26
27#define vrepchr		v0
28#define qdata		q1
29#define vdata		v1
30#define vhas_chr	v2
31#define vend		v3
32#define dend		d3
33
34/*
35   Core algorithm:
36   For each 16-byte chunk we calculate a 64-bit nibble mask value with four bits
37   per byte. We take 4 bits of every comparison byte with shift right and narrow
38   by 4 instruction. Since the bits in the nibble mask reflect the order in
39   which things occur in the original string, counting leading zeros identifies
40   exactly which byte matched.  */
41
42ENTRY (__memchr_aarch64_mte)
43	bic	src, srcin, 15
44	cbz	cntin, L(nomatch)
45	ld1	{vdata.16b}, [src]
46	dup	vrepchr.16b, chrin
47	cmeq	vhas_chr.16b, vdata.16b, vrepchr.16b
48	lsl	shift, srcin, 2
49	shrn	vend.8b, vhas_chr.8h, 4		/* 128->64 */
50	fmov	synd, dend
51	lsr	synd, synd, shift
52	cbz	synd, L(start_loop)
53
54	rbit	synd, synd
55	clz	synd, synd
56	cmp	cntin, synd, lsr 2
57	add	result, srcin, synd, lsr 2
58	csel	result, result, xzr, hi
59	ret
60
61	.p2align 3
62L(start_loop):
63	sub	tmp, src, srcin
64	add	tmp, tmp, 17
65	subs	cntrem, cntin, tmp
66	b.lo	L(nomatch)
67
68	/* Make sure that it won't overread by a 16-byte chunk */
69	tbz	cntrem, 4, L(loop32_2)
70	sub	src, src, 16
71	.p2align 4
72L(loop32):
73	ldr	qdata, [src, 32]!
74	cmeq	vhas_chr.16b, vdata.16b, vrepchr.16b
75	umaxp	vend.16b, vhas_chr.16b, vhas_chr.16b		/* 128->64 */
76	fmov	synd, dend
77	cbnz	synd, L(end)
78
79L(loop32_2):
80	ldr	qdata, [src, 16]
81	cmeq	vhas_chr.16b, vdata.16b, vrepchr.16b
82	subs	cntrem, cntrem, 32
83	b.lo	L(end_2)
84	umaxp	vend.16b, vhas_chr.16b, vhas_chr.16b		/* 128->64 */
85	fmov	synd, dend
86	cbz	synd, L(loop32)
87L(end_2):
88	add	src, src, 16
89L(end):
90	shrn	vend.8b, vhas_chr.8h, 4		/* 128->64 */
91	sub	cntrem, src, srcin
92	fmov	synd, dend
93	sub	cntrem, cntin, cntrem
94#ifndef __AARCH64EB__
95	rbit	synd, synd
96#endif
97	clz	synd, synd
98	cmp	cntrem, synd, lsr 2
99	add	result, src, synd, lsr 2
100	csel	result, result, xzr, hi
101	ret
102
103L(nomatch):
104	mov	result, 0
105	ret
106
107END (__memchr_aarch64_mte)
108
109