1 | /* Optimized strchrnul implementation for PowerPC64/POWER7 using cmpb insn. |
2 | Copyright (C) 2010-2024 Free Software Foundation, Inc. |
3 | This file is part of the GNU C Library. |
4 | |
5 | The GNU C Library is free software; you can redistribute it and/or |
6 | modify it under the terms of the GNU Lesser General Public |
7 | License as published by the Free Software Foundation; either |
8 | version 2.1 of the License, or (at your option) any later version. |
9 | |
10 | The GNU C Library is distributed in the hope that it will be useful, |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 | Lesser General Public License for more details. |
14 | |
15 | You should have received a copy of the GNU Lesser General Public |
16 | License along with the GNU C Library; if not, see |
17 | <https://www.gnu.org/licenses/>. */ |
18 | |
19 | #include <sysdep.h> |
20 | |
21 | #ifndef STRCHRNUL |
22 | # define STRCHRNUL __strchrnul |
23 | #endif |
24 | /* int [r3] strchrnul (char *s [r3], int c [r4]) */ |
25 | .machine power7 |
26 | ENTRY_TOCLESS (STRCHRNUL) |
27 | CALL_MCOUNT 2 |
28 | dcbt 0,r3 |
29 | clrrdi r8,r3,3 /* Align the address to doubleword boundary. */ |
30 | |
31 | /* Replicate byte to doubleword. */ |
32 | insrdi r4,r4,8,48 |
33 | insrdi r4,r4,16,32 |
34 | insrdi r4,r4,32,0 |
35 | |
36 | rlwinm r6,r3,3,26,28 /* Calculate padding. */ |
37 | ld r12,0(r8) /* Load doubleword from memory. */ |
38 | li r0,0 /* Doubleword with null chars to use |
39 | with cmpb. */ |
40 | |
41 | /* Now r4 has a doubleword of c bytes and r0 has |
42 | a doubleword of null bytes. */ |
43 | |
44 | cmpb r10,r12,r0 /* Compare each byte against c byte. */ |
45 | cmpb r9,r12,r4 /* Compare each byte against null byte. */ |
46 | |
47 | /* Move the doublewords left and right to discard the bits that are |
48 | not part of the string and to bring them back as zeros. */ |
49 | #ifdef __LITTLE_ENDIAN__ |
50 | srd r10,r10,r6 |
51 | srd r9,r9,r6 |
52 | sld r10,r10,r6 |
53 | sld r9,r9,r6 |
54 | #else |
55 | sld r10,r10,r6 |
56 | sld r9,r9,r6 |
57 | srd r10,r10,r6 |
58 | srd r9,r9,r6 |
59 | #endif |
60 | or r5,r9,r10 /* OR the results to speed things up. */ |
61 | cmpdi cr7,r5,0 /* If r5 == 0, no c or null bytes |
62 | have been found. */ |
63 | bne cr7,L(done) |
64 | |
65 | mtcrf 0x01,r8 |
66 | |
67 | /* Are we now aligned to a quadword boundary? If so, skip to |
68 | the main loop. Otherwise, go through the alignment code. */ |
69 | |
70 | bt 28,L(loop) |
71 | |
72 | /* Handle DWORD2 of pair. */ |
73 | ldu r12,8(r8) |
74 | cmpb r10,r12,r0 |
75 | cmpb r9,r12,r4 |
76 | or r5,r9,r10 |
77 | cmpdi cr7,r5,0 |
78 | bne cr7,L(done) |
79 | b L(loop) /* We branch here (rather than falling through) |
80 | to skip the nops due to heavy alignment |
81 | of the loop below. */ |
82 | |
83 | .p2align 5 |
84 | L(loop): |
85 | /* Load two doublewords, compare and merge in a |
86 | single register for speed. This is an attempt |
87 | to speed up the null-checking process for bigger strings. */ |
88 | ld r12,8(r8) |
89 | ldu r11,16(r8) |
90 | cmpb r10,r12,r0 |
91 | cmpb r9,r12,r4 |
92 | cmpb r6,r11,r0 |
93 | cmpb r7,r11,r4 |
94 | or r5,r9,r10 |
95 | or r10,r6,r7 |
96 | or r11,r5,r10 |
97 | cmpdi cr7,r11,0 |
98 | beq cr7,L(loop) |
99 | |
100 | /* OK, one (or both) of the doublewords contains a c/null byte. Check |
101 | the first doubleword and decrement the address in case the first |
102 | doubleword really contains a c/null byte. */ |
103 | |
104 | cmpdi cr6,r5,0 |
105 | addi r8,r8,-8 |
106 | bne cr6,L(done) |
107 | |
108 | /* The c/null byte must be in the second doubleword. Adjust the |
109 | address again and move the result of cmpb to r5 so we can calculate |
110 | the pointer. */ |
111 | mr r5,r10 |
112 | addi r8,r8,8 |
113 | |
114 | /* r5 has the output of the cmpb instruction, that is, it contains |
115 | 0xff in the same position as the c/null byte in the original |
116 | doubleword from the string. Use that to calculate the pointer. */ |
117 | L(done): |
118 | #ifdef __LITTLE_ENDIAN__ |
119 | addi r0,r5,-1 |
120 | andc r0,r0,r5 |
121 | popcntd r0,r0 |
122 | #else |
123 | cntlzd r0,r5 /* Count leading zeros before the match. */ |
124 | #endif |
125 | srdi r0,r0,3 /* Convert leading zeros to bytes. */ |
126 | add r3,r8,r0 /* Return address of matching c/null byte. */ |
127 | blr |
128 | END (STRCHRNUL) |
129 | weak_alias (STRCHRNUL, strchrnul) |
130 | libc_hidden_builtin_def (STRCHRNUL) |
131 | |