]>
Commit | Line | Data |
---|---|---|
95e431cc WD |
1 | /* memchr - find a character in a memory zone |
2 | ||
dff8da6b | 3 | Copyright (C) 2015-2024 Free Software Foundation, Inc. |
95e431cc WD |
4 | |
5 | This file is part of the GNU C Library. | |
6 | ||
7 | The GNU C Library is free software; you can redistribute it and/or | |
8 | modify it under the terms of the GNU Lesser General Public | |
9 | License as published by the Free Software Foundation; either | |
10 | version 2.1 of the License, or (at your option) any later version. | |
11 | ||
12 | The GNU C Library is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | Lesser General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU Lesser General Public | |
18 | License along with the GNU C Library. If not, see | |
5a82c748 | 19 | <https://www.gnu.org/licenses/>. */ |
95e431cc WD |
20 | |
21 | #include <sysdep.h> | |
22 | ||
23 | /* Assumptions: | |
24 | * | |
7ff89996 AB |
25 | * ARMv8-a, AArch64, Advanced SIMD. |
26 | * MTE compatible. | |
95e431cc WD |
27 | */ |
28 | ||
83d1cc42 FX |
29 | #ifndef MEMCHR |
30 | # define MEMCHR __memchr | |
31 | #endif | |
32 | ||
95e431cc WD |
33 | #define srcin x0 |
34 | #define chrin w1 | |
35 | #define cntin x2 | |
95e431cc WD |
36 | #define result x0 |
37 | ||
38 | #define src x3 | |
7ff89996 AB |
39 | #define cntrem x4 |
40 | #define synd x5 | |
41 | #define shift x6 | |
42 | #define tmp x7 | |
95e431cc WD |
43 | |
44 | #define vrepchr v0 | |
7ff89996 AB |
45 | #define qdata q1 |
46 | #define vdata v1 | |
47 | #define vhas_chr v2 | |
3c998069 DK |
48 | #define vend v3 |
49 | #define dend d3 | |
95e431cc WD |
50 | |
51 | /* | |
7ff89996 | 52 | Core algorithm: |
3c998069 DK |
53 | For each 16-byte chunk we calculate a 64-bit nibble mask value with four bits |
54 | per byte. We take 4 bits of every comparison byte with shift right and narrow | |
55 | by 4 instruction. Since the bits in the nibble mask reflect the order in | |
56 | which things occur in the original string, counting leading zeros identifies | |
57 | exactly which byte matched. */ | |
95e431cc | 58 | |
83d1cc42 | 59 | ENTRY (MEMCHR) |
45b1e17e SN |
60 | PTR_ARG (0) |
61 | SIZE_ARG (2) | |
7ff89996 AB |
62 | bic src, srcin, 15 |
63 | cbz cntin, L(nomatch) | |
64 | ld1 {vdata.16b}, [src] | |
95e431cc | 65 | dup vrepchr.16b, chrin |
7ff89996 AB |
66 | cmeq vhas_chr.16b, vdata.16b, vrepchr.16b |
67 | lsl shift, srcin, 2 | |
3c998069 | 68 | shrn vend.8b, vhas_chr.8h, 4 /* 128->64 */ |
7ff89996 AB |
69 | fmov synd, dend |
70 | lsr synd, synd, shift | |
71 | cbz synd, L(start_loop) | |
95e431cc | 72 | |
7ff89996 AB |
73 | rbit synd, synd |
74 | clz synd, synd | |
7ff89996 | 75 | cmp cntin, synd, lsr 2 |
ce758d4f | 76 | add result, srcin, synd, lsr 2 |
7ff89996 AB |
77 | csel result, result, xzr, hi |
78 | ret | |
79 | ||
ce758d4f | 80 | .p2align 3 |
7ff89996 AB |
81 | L(start_loop): |
82 | sub tmp, src, srcin | |
ce758d4f | 83 | add tmp, tmp, 17 |
7ff89996 | 84 | subs cntrem, cntin, tmp |
ce758d4f | 85 | b.lo L(nomatch) |
7ff89996 AB |
86 | |
87 | /* Make sure that it won't overread by a 16-byte chunk */ | |
ce758d4f WD |
88 | tbz cntrem, 4, L(loop32_2) |
89 | sub src, src, 16 | |
7ff89996 AB |
90 | .p2align 4 |
91 | L(loop32): | |
ce758d4f | 92 | ldr qdata, [src, 32]! |
7ff89996 AB |
93 | cmeq vhas_chr.16b, vdata.16b, vrepchr.16b |
94 | umaxp vend.16b, vhas_chr.16b, vhas_chr.16b /* 128->64 */ | |
95 | fmov synd, dend | |
96 | cbnz synd, L(end) | |
97 | ||
98 | L(loop32_2): | |
ce758d4f | 99 | ldr qdata, [src, 16] |
7ff89996 | 100 | cmeq vhas_chr.16b, vdata.16b, vrepchr.16b |
ce758d4f WD |
101 | subs cntrem, cntrem, 32 |
102 | b.lo L(end_2) | |
7ff89996 AB |
103 | umaxp vend.16b, vhas_chr.16b, vhas_chr.16b /* 128->64 */ |
104 | fmov synd, dend | |
105 | cbz synd, L(loop32) | |
ce758d4f WD |
106 | L(end_2): |
107 | add src, src, 16 | |
95e431cc | 108 | L(end): |
3c998069 | 109 | shrn vend.8b, vhas_chr.8h, 4 /* 128->64 */ |
ce758d4f | 110 | sub cntrem, src, srcin |
7ff89996 | 111 | fmov synd, dend |
ce758d4f | 112 | sub cntrem, cntin, cntrem |
7ff89996 | 113 | #ifndef __AARCH64EB__ |
95e431cc | 114 | rbit synd, synd |
7ff89996 | 115 | #endif |
95e431cc | 116 | clz synd, synd |
7ff89996 AB |
117 | cmp cntrem, synd, lsr 2 |
118 | add result, src, synd, lsr 2 | |
119 | csel result, result, xzr, hi | |
95e431cc WD |
120 | ret |
121 | ||
7ff89996 AB |
122 | L(nomatch): |
123 | mov result, 0 | |
95e431cc | 124 | ret |
7ff89996 | 125 | |
83d1cc42 FX |
126 | END (MEMCHR) |
127 | weak_alias (MEMCHR, memchr) | |
95e431cc | 128 | libc_hidden_builtin_def (memchr) |