1 | /* strrchr: find the last instance of a character in a string. |
2 | |
3 | Copyright (C) 2014-2022 Free Software Foundation, Inc. |
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 |
19 | <https://www.gnu.org/licenses/>. */ |
20 | |
21 | #include <sysdep.h> |
22 | |
23 | /* Assumptions: |
24 | * |
25 | * ARMv8-a, AArch64 |
26 | * Neon Available. |
27 | * MTE compatible. |
28 | */ |
29 | |
30 | /* Arguments and results. */ |
31 | #define srcin x0 |
32 | #define chrin w1 |
33 | #define result x0 |
34 | |
35 | #define src x2 |
36 | #define tmp x3 |
37 | #define wtmp w3 |
38 | #define synd x3 |
39 | #define shift x4 |
40 | #define src_match x4 |
41 | #define nul_match x5 |
42 | #define chr_match x6 |
43 | |
44 | #define vrepchr v0 |
45 | #define vdata v1 |
46 | #define vhas_nul v2 |
47 | #define vhas_chr v3 |
48 | #define vrepmask v4 |
49 | #define vrepmask2 v5 |
50 | #define vend v5 |
51 | #define dend d5 |
52 | |
53 | /* Core algorithm. |
54 | |
55 | For each 16-byte chunk we calculate a 64-bit syndrome value, with |
56 | four bits per byte (LSB is always in bits 0 and 1, for both big |
57 | and little-endian systems). For each tuple, bits 0-1 are set if |
58 | the relevant byte matched the requested character; bits 2-3 are set |
59 | if the relevant byte matched the NUL end of string. */ |
60 | |
61 | ENTRY(strrchr) |
62 | PTR_ARG (0) |
63 | bic src, srcin, 15 |
64 | dup vrepchr.16b, chrin |
65 | mov wtmp, 0x3003 |
66 | dup vrepmask.8h, wtmp |
67 | tst srcin, 15 |
68 | beq L(loop1) |
69 | |
70 | ld1 {vdata.16b}, [src], 16 |
71 | cmeq vhas_nul.16b, vdata.16b, 0 |
72 | cmeq vhas_chr.16b, vdata.16b, vrepchr.16b |
73 | mov wtmp, 0xf00f |
74 | dup vrepmask2.8h, wtmp |
75 | bit vhas_nul.16b, vhas_chr.16b, vrepmask.16b |
76 | and vhas_nul.16b, vhas_nul.16b, vrepmask2.16b |
77 | addp vend.16b, vhas_nul.16b, vhas_nul.16b |
78 | lsl shift, srcin, 2 |
79 | fmov synd, dend |
80 | lsr synd, synd, shift |
81 | lsl synd, synd, shift |
82 | ands nul_match, synd, 0xcccccccccccccccc |
83 | bne L(tail) |
84 | cbnz synd, L(loop2) |
85 | |
86 | .p2align 5 |
87 | L(loop1): |
88 | ld1 {vdata.16b}, [src], 16 |
89 | cmeq vhas_chr.16b, vdata.16b, vrepchr.16b |
90 | cmhs vhas_nul.16b, vhas_chr.16b, vdata.16b |
91 | umaxp vend.16b, vhas_nul.16b, vhas_nul.16b |
92 | fmov synd, dend |
93 | cbz synd, L(loop1) |
94 | |
95 | cmeq vhas_nul.16b, vdata.16b, 0 |
96 | bit vhas_nul.16b, vhas_chr.16b, vrepmask.16b |
97 | bic vhas_nul.8h, 0x0f, lsl 8 |
98 | addp vend.16b, vhas_nul.16b, vhas_nul.16b |
99 | fmov synd, dend |
100 | ands nul_match, synd, 0xcccccccccccccccc |
101 | beq L(loop2) |
102 | |
103 | L(tail): |
104 | sub nul_match, nul_match, 1 |
105 | and chr_match, synd, 0x3333333333333333 |
106 | ands chr_match, chr_match, nul_match |
107 | sub result, src, 1 |
108 | clz tmp, chr_match |
109 | sub result, result, tmp, lsr 2 |
110 | csel result, result, xzr, ne |
111 | ret |
112 | |
113 | .p2align 4 |
114 | L(loop2): |
115 | cmp synd, 0 |
116 | csel src_match, src, src_match, ne |
117 | csel chr_match, synd, chr_match, ne |
118 | ld1 {vdata.16b}, [src], 16 |
119 | cmeq vhas_nul.16b, vdata.16b, 0 |
120 | cmeq vhas_chr.16b, vdata.16b, vrepchr.16b |
121 | bit vhas_nul.16b, vhas_chr.16b, vrepmask.16b |
122 | umaxp vend.16b, vhas_nul.16b, vhas_nul.16b |
123 | fmov synd, dend |
124 | tst synd, 0xcccccccccccccccc |
125 | beq L(loop2) |
126 | |
127 | bic vhas_nul.8h, 0x0f, lsl 8 |
128 | addp vend.16b, vhas_nul.16b, vhas_nul.16b |
129 | fmov synd, dend |
130 | and nul_match, synd, 0xcccccccccccccccc |
131 | sub nul_match, nul_match, 1 |
132 | and tmp, synd, 0x3333333333333333 |
133 | ands tmp, tmp, nul_match |
134 | csel chr_match, tmp, chr_match, ne |
135 | csel src_match, src, src_match, ne |
136 | sub src_match, src_match, 1 |
137 | clz tmp, chr_match |
138 | sub result, src_match, tmp, lsr 2 |
139 | ret |
140 | |
141 | END(strrchr) |
142 | weak_alias (strrchr, rindex) |
143 | libc_hidden_builtin_def (strrchr) |
144 | |