xref: /freebsd/contrib/arm-optimized-routines/string/aarch64/memcmp.S (revision 5956d97f4b3204318ceb6aa9c77bd0bc6ea87a41)
1/* memcmp - compare memory
2 *
3 * Copyright (c) 2013-2020, Arm Limited.
4 * SPDX-License-Identifier: MIT
5 */
6
7/* Assumptions:
8 *
9 * ARMv8-a, AArch64, unaligned accesses.
10 */
11
12#include "../asmdefs.h"
13
14/* Parameters and result.  */
15#define src1		x0
16#define src2		x1
17#define limit		x2
18#define result		w0
19
20/* Internal variables.  */
21#define data1		x3
22#define data1w		w3
23#define data1h		x4
24#define data2		x5
25#define data2w		w5
26#define data2h		x6
27#define tmp1		x7
28#define tmp2		x8
29
30ENTRY (__memcmp_aarch64)
31	PTR_ARG (0)
32	PTR_ARG (1)
33	SIZE_ARG (2)
34	subs	limit, limit, 8
35	b.lo	L(less8)
36
37	ldr	data1, [src1], 8
38	ldr	data2, [src2], 8
39	cmp	data1, data2
40	b.ne	L(return)
41
42	subs	limit, limit, 8
43	b.gt	L(more16)
44
45	ldr	data1, [src1, limit]
46	ldr	data2, [src2, limit]
47	b	L(return)
48
49L(more16):
50	ldr	data1, [src1], 8
51	ldr	data2, [src2], 8
52	cmp	data1, data2
53	bne	L(return)
54
55	/* Jump directly to comparing the last 16 bytes for 32 byte (or less)
56	   strings.  */
57	subs	limit, limit, 16
58	b.ls	L(last_bytes)
59
60	/* We overlap loads between 0-32 bytes at either side of SRC1 when we
61	   try to align, so limit it only to strings larger than 128 bytes.  */
62	cmp	limit, 96
63	b.ls	L(loop16)
64
65	/* Align src1 and adjust src2 with bytes not yet done.  */
66	and	tmp1, src1, 15
67	add	limit, limit, tmp1
68	sub	src1, src1, tmp1
69	sub	src2, src2, tmp1
70
71	/* Loop performing 16 bytes per iteration using aligned src1.
72	   Limit is pre-decremented by 16 and must be larger than zero.
73	   Exit if <= 16 bytes left to do or if the data is not equal.  */
74	.p2align 4
75L(loop16):
76	ldp	data1, data1h, [src1], 16
77	ldp	data2, data2h, [src2], 16
78	subs	limit, limit, 16
79	ccmp	data1, data2, 0, hi
80	ccmp	data1h, data2h, 0, eq
81	b.eq	L(loop16)
82
83	cmp	data1, data2
84	bne	L(return)
85	mov	data1, data1h
86	mov	data2, data2h
87	cmp	data1, data2
88	bne	L(return)
89
90	/* Compare last 1-16 bytes using unaligned access.  */
91L(last_bytes):
92	add	src1, src1, limit
93	add	src2, src2, limit
94	ldp	data1, data1h, [src1]
95	ldp	data2, data2h, [src2]
96	cmp     data1, data2
97	bne	L(return)
98	mov	data1, data1h
99	mov	data2, data2h
100	cmp	data1, data2
101
102	/* Compare data bytes and set return value to 0, -1 or 1.  */
103L(return):
104#ifndef __AARCH64EB__
105	rev	data1, data1
106	rev	data2, data2
107#endif
108	cmp     data1, data2
109L(ret_eq):
110	cset	result, ne
111	cneg	result, result, lo
112	ret
113
114	.p2align 4
115	/* Compare up to 8 bytes.  Limit is [-8..-1].  */
116L(less8):
117	adds	limit, limit, 4
118	b.lo	L(less4)
119	ldr	data1w, [src1], 4
120	ldr	data2w, [src2], 4
121	cmp	data1w, data2w
122	b.ne	L(return)
123	sub	limit, limit, 4
124L(less4):
125	adds	limit, limit, 4
126	beq	L(ret_eq)
127L(byte_loop):
128	ldrb	data1w, [src1], 1
129	ldrb	data2w, [src2], 1
130	subs	limit, limit, 1
131	ccmp	data1w, data2w, 0, ne	/* NZCV = 0b0000.  */
132	b.eq	L(byte_loop)
133	sub	result, data1w, data2w
134	ret
135
136END (__memcmp_aarch64)
137
138