xref: /freebsd/contrib/arm-optimized-routines/string/aarch64/experimental/memcmp-sve.S (revision ee3960cba1068e12fb032a68c46d74841d9edab3)
1/*
2 * memcmp - compare memory
3 *
4 * Copyright (c) 2018-2022, Arm Limited.
5 * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
6 */
7
8#include "asmdefs.h"
9
10.arch armv8-a+sve
11
12/* Assumptions:
13 *
14 * ARMv8-a, AArch64
15 * SVE Available.
16 */
17
18ENTRY (__memcmp_aarch64_sve)
19	mov	x3, 0			/* initialize off */
20
210:	whilelo	p0.b, x3, x2		/* while off < max */
22	b.none	9f
23
24	ld1b	z0.b, p0/z, [x0, x3]	/* read vectors bounded by max.  */
25	ld1b	z1.b, p0/z, [x1, x3]
26
27	/* Increment for a whole vector, even if we've only read a partial.
28	   This is significantly cheaper than INCP, and since OFF is not
29	   used after the loop it is ok to increment OFF past MAX.  */
30	incb	x3
31
32	cmpne	p1.b, p0/z, z0.b, z1.b	/* while no inequalities */
33	b.none	0b
34
35	/* Found inequality.  */
361:	brkb	p1.b, p0/z, p1.b	/* find first such */
37	lasta	w0, p1, z0.b		/* extract each byte */
38	lasta	w1, p1, z1.b
39	sub	x0, x0, x1		/* return comparison */
40	ret
41
42	/* Found end-of-count.  */
439:	mov	x0, 0			/* return equality */
44	ret
45
46END (__memcmp_aarch64_sve)
47