1/*
2 * strcmp - compare two strings
3 *
4 * Copyright (c) 2012-2022, Arm Limited.
5 * SPDX-License-Identifier: MIT
6 */
7
8
9/* Assumptions:
10 *
11 * ARMv8-a, AArch64.
12 * MTE compatible.
13 */
14
15#include <machine/asm.h>
16
17#define L(l) .L ## l
18
19#define REP8_01 0x0101010101010101
20#define REP8_7f 0x7f7f7f7f7f7f7f7f
21
22#define src1		x0
23#define src2		x1
24#define result		x0
25
26#define data1		x2
27#define data1w		w2
28#define data2		x3
29#define data2w		w3
30#define has_nul		x4
31#define diff		x5
32#define off1		x5
33#define syndrome	x6
34#define tmp		x6
35#define data3		x7
36#define zeroones	x8
37#define shift		x9
38#define off2		x10
39
40/* On big-endian early bytes are at MSB and on little-endian LSB.
41   LS_FW means shifting towards early bytes.  */
42#ifdef __AARCH64EB__
43# define LS_FW lsl
44#else
45# define LS_FW lsr
46#endif
47
48/* NUL detection works on the principle that (X - 1) & (~X) & 0x80
49   (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
50   can be done in parallel across the entire word.
51   Since carry propagation makes 0x1 bytes before a NUL byte appear
52   NUL too in big-endian, byte-reverse the data before the NUL check.  */
53
54
55ENTRY (strcmp)
56	sub	off2, src2, src1
57	mov	zeroones, REP8_01
58	and	tmp, src1, 7
59	tst	off2, 7
60	b.ne	L(misaligned8)
61	cbnz	tmp, L(mutual_align)
62
63	.p2align 4
64
65L(loop_aligned):
66	ldr	data2, [src1, off2]
67	ldr	data1, [src1], 8
68L(start_realigned):
69#ifdef __AARCH64EB__
70	rev	tmp, data1
71	sub	has_nul, tmp, zeroones
72	orr	tmp, tmp, REP8_7f
73#else
74	sub	has_nul, data1, zeroones
75	orr	tmp, data1, REP8_7f
76#endif
77	bics	has_nul, has_nul, tmp	/* Non-zero if NUL terminator.  */
78	ccmp	data1, data2, 0, eq
79	b.eq	L(loop_aligned)
80#ifdef __AARCH64EB__
81	rev	has_nul, has_nul
82#endif
83	eor	diff, data1, data2
84	orr	syndrome, diff, has_nul
85L(end):
86#ifndef __AARCH64EB__
87	rev	syndrome, syndrome
88	rev	data1, data1
89	rev	data2, data2
90#endif
91	clz	shift, syndrome
92	/* The most-significant-non-zero bit of the syndrome marks either the
93	   first bit that is different, or the top bit of the first zero byte.
94	   Shifting left now will bring the critical information into the
95	   top bits.  */
96	lsl	data1, data1, shift
97	lsl	data2, data2, shift
98	/* But we need to zero-extend (char is unsigned) the value and then
99	   perform a signed 32-bit subtraction.  */
100	lsr	data1, data1, 56
101	sub	result, data1, data2, lsr 56
102	ret
103
104	.p2align 4
105
106L(mutual_align):
107	/* Sources are mutually aligned, but are not currently at an
108	   alignment boundary.  Round down the addresses and then mask off
109	   the bytes that precede the start point.  */
110	bic	src1, src1, 7
111	ldr	data2, [src1, off2]
112	ldr	data1, [src1], 8
113	neg	shift, src2, lsl 3	/* Bits to alignment -64.  */
114	mov	tmp, -1
115	LS_FW	tmp, tmp, shift
116	orr	data1, data1, tmp
117	orr	data2, data2, tmp
118	b	L(start_realigned)
119
120L(misaligned8):
121	/* Align SRC1 to 8 bytes and then compare 8 bytes at a time, always
122	   checking to make sure that we don't access beyond the end of SRC2.  */
123	cbz	tmp, L(src1_aligned)
124L(do_misaligned):
125	ldrb	data1w, [src1], 1
126	ldrb	data2w, [src2], 1
127	cmp	data1w, 0
128	ccmp	data1w, data2w, 0, ne	/* NZCV = 0b0000.  */
129	b.ne	L(done)
130	tst	src1, 7
131	b.ne	L(do_misaligned)
132
133L(src1_aligned):
134	neg	shift, src2, lsl 3
135	bic	src2, src2, 7
136	ldr	data3, [src2], 8
137#ifdef __AARCH64EB__
138	rev	data3, data3
139#endif
140	lsr	tmp, zeroones, shift
141	orr	data3, data3, tmp
142	sub	has_nul, data3, zeroones
143	orr	tmp, data3, REP8_7f
144	bics	has_nul, has_nul, tmp
145	b.ne	L(tail)
146
147	sub	off1, src2, src1
148
149	.p2align 4
150
151L(loop_unaligned):
152	ldr	data3, [src1, off1]
153	ldr	data2, [src1, off2]
154#ifdef __AARCH64EB__
155	rev	data3, data3
156#endif
157	sub	has_nul, data3, zeroones
158	orr	tmp, data3, REP8_7f
159	ldr	data1, [src1], 8
160	bics	has_nul, has_nul, tmp
161	ccmp	data1, data2, 0, eq
162	b.eq	L(loop_unaligned)
163
164	lsl	tmp, has_nul, shift
165#ifdef __AARCH64EB__
166	rev	tmp, tmp
167#endif
168	eor	diff, data1, data2
169	orr	syndrome, diff, tmp
170	cbnz	syndrome, L(end)
171L(tail):
172	ldr	data1, [src1]
173	neg	shift, shift
174	lsr	data2, data3, shift
175	lsr	has_nul, has_nul, shift
176#ifdef __AARCH64EB__
177	rev     data2, data2
178	rev	has_nul, has_nul
179#endif
180	eor	diff, data1, data2
181	orr	syndrome, diff, has_nul
182	b	L(end)
183
184L(done):
185	sub	result, data1, data2
186	ret
187
188END (strcmp)
189
190