]>
Commit | Line | Data |
---|---|---|
fe2f79db | 1 | /* Optimized strnlen implementation for PowerPC32/POWER7 using cmpb insn. |
dff8da6b | 2 | Copyright (C) 2010-2024 Free Software Foundation, Inc. |
fe2f79db LM |
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 | |
59ba27a6 | 16 | License along with the GNU C Library; if not, see |
5a82c748 | 17 | <https://www.gnu.org/licenses/>. */ |
fe2f79db LM |
18 | |
19 | #include <sysdep.h> | |
fe2f79db LM |
20 | |
21 | /* int [r3] strnlen (char *s [r3], int size [r4]) */ | |
22 | .machine power7 | |
b5510883 | 23 | ENTRY (__strnlen) |
fe2f79db LM |
24 | CALL_MCOUNT |
25 | dcbt 0,r3 | |
26 | clrrwi r8,r3,2 /* Align the address to word boundary. */ | |
27 | add r7,r3,r4 /* Calculate the last acceptable address. */ | |
28 | cmplwi r4,16 | |
29 | li r0,0 /* Word with null chars. */ | |
33ee81de | 30 | addi r7,r7,-1 |
fe2f79db LM |
31 | ble L(small_range) |
32 | ||
fe2f79db LM |
33 | rlwinm r6,r3,3,27,28 /* Calculate padding. */ |
34 | lwz r12,0(r8) /* Load word from memory. */ | |
35 | cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */ | |
33ee81de AM |
36 | #ifdef __LITTLE_ENDIAN__ |
37 | srw r10,r10,r6 | |
38 | slw r10,r10,r6 | |
39 | #else | |
fe2f79db LM |
40 | slw r10,r10,r6 |
41 | srw r10,r10,r6 | |
33ee81de | 42 | #endif |
fe2f79db LM |
43 | cmplwi cr7,r10,0 /* If r10 == 0, no null's have been found. */ |
44 | bne cr7,L(done) | |
45 | ||
33ee81de | 46 | clrrwi r7,r7,2 /* Address of last word. */ |
fe2f79db LM |
47 | mtcrf 0x01,r8 |
48 | /* Are we now aligned to a doubleword boundary? If so, skip to | |
49 | the main loop. Otherwise, go through the alignment code. */ | |
50 | ||
51 | bt 29,L(loop_setup) | |
52 | ||
33ee81de | 53 | /* Handle WORD2 of pair. */ |
fe2f79db LM |
54 | lwzu r12,4(r8) |
55 | cmpb r10,r12,r0 | |
56 | cmplwi cr7,r10,0 | |
57 | bne cr7,L(done) | |
58 | ||
fe2f79db | 59 | L(loop_setup): |
33ee81de AM |
60 | /* The last word we want to read in the loop below is the one |
61 | containing the last byte of the string, ie. the word at | |
62 | (s + size - 1) & ~3, or r7. The first word read is at | |
63 | r8 + 4, we read 2 * cnt words, so the last word read will | |
64 | be at r8 + 4 + 8 * cnt - 4. Solving for cnt gives | |
65 | cnt = (r7 - r8) / 8 */ | |
66 | sub r5,r7,r8 | |
fe2f79db LM |
67 | srwi r6,r5,3 /* Number of loop iterations. */ |
68 | mtctr r6 /* Setup the counter. */ | |
33ee81de AM |
69 | |
70 | /* Main loop to look for the null byte in the string. Since | |
fe2f79db LM |
71 | it's a small loop (< 8 instructions), align it to 32-bytes. */ |
72 | .p2align 5 | |
73 | L(loop): | |
74 | /* Load two words, compare and merge in a | |
75 | single register for speed. This is an attempt | |
76 | to speed up the null-checking process for bigger strings. */ | |
77 | ||
78 | lwz r12,4(r8) | |
79 | lwzu r11,8(r8) | |
80 | cmpb r10,r12,r0 | |
81 | cmpb r9,r11,r0 | |
82 | or r5,r9,r10 /* Merge everything in one word. */ | |
83 | cmplwi cr7,r5,0 | |
84 | bne cr7,L(found) | |
85 | bdnz L(loop) | |
33ee81de AM |
86 | |
87 | /* We may have one more word to read. */ | |
88 | cmplw cr6,r8,r7 | |
89 | beq cr6,L(end_max) | |
90 | ||
91 | lwzu r12,4(r8) | |
92 | cmpb r10,r12,r0 | |
93 | cmplwi cr6,r10,0 | |
94 | bne cr6,L(done) | |
fe2f79db LM |
95 | |
96 | L(end_max): | |
33ee81de | 97 | mr r3,r4 |
fe2f79db LM |
98 | blr |
99 | ||
100 | /* OK, one (or both) of the words contains a null byte. Check | |
101 | the first word and decrement the address in case the first | |
102 | word really contains a null byte. */ | |
103 | .align 4 | |
104 | L(found): | |
105 | cmplwi cr6,r10,0 | |
106 | addi r8,r8,-4 | |
107 | bne cr6,L(done) | |
108 | ||
109 | /* The null byte must be in the second word. Adjust the address | |
110 | again and move the result of cmpb to r10 so we can calculate the | |
111 | length. */ | |
112 | ||
113 | mr r10,r9 | |
114 | addi r8,r8,4 | |
115 | ||
116 | /* r10 has the output of the cmpb instruction, that is, it contains | |
117 | 0xff in the same position as the null byte in the original | |
118 | word from the string. Use that to calculate the length. | |
119 | We need to make sure the null char is *before* the end of the | |
120 | range. */ | |
121 | L(done): | |
33ee81de AM |
122 | #ifdef __LITTLE_ENDIAN__ |
123 | addi r0,r10,-1 | |
124 | andc r0,r0,r10 | |
125 | popcntw r0,r0 | |
126 | #else | |
127 | cntlzw r0,r10 /* Count leading zeros before the match. */ | |
128 | #endif | |
129 | sub r3,r8,r3 | |
130 | srwi r0,r0,3 /* Convert leading/trailing zeros to bytes. */ | |
131 | add r3,r3,r0 /* Length until the match. */ | |
132 | cmplw r3,r4 | |
133 | blelr | |
134 | mr r3,r4 | |
fe2f79db LM |
135 | blr |
136 | ||
33ee81de | 137 | /* Deals with size <= 16. */ |
fe2f79db LM |
138 | .align 4 |
139 | L(small_range): | |
140 | cmplwi r4,0 | |
33ee81de AM |
141 | beq L(end_max) |
142 | ||
143 | clrrwi r7,r7,2 /* Address of last word. */ | |
fe2f79db LM |
144 | |
145 | rlwinm r6,r3,3,27,28 /* Calculate padding. */ | |
146 | lwz r12,0(r8) /* Load word from memory. */ | |
147 | cmpb r10,r12,r0 /* Check for null bytes in WORD1. */ | |
33ee81de AM |
148 | #ifdef __LITTLE_ENDIAN__ |
149 | srw r10,r10,r6 | |
150 | slw r10,r10,r6 | |
151 | #else | |
fe2f79db LM |
152 | slw r10,r10,r6 |
153 | srw r10,r10,r6 | |
33ee81de | 154 | #endif |
fe2f79db LM |
155 | cmplwi cr7,r10,0 |
156 | bne cr7,L(done) | |
157 | ||
33ee81de AM |
158 | cmplw r8,r7 |
159 | beq L(end_max) | |
fe2f79db LM |
160 | |
161 | .p2align 5 | |
162 | L(loop_small): | |
163 | lwzu r12,4(r8) | |
164 | cmpb r10,r12,r0 | |
fe2f79db LM |
165 | cmplwi cr6,r10,0 |
166 | bne cr6,L(done) | |
33ee81de AM |
167 | cmplw r8,r7 |
168 | bne L(loop_small) | |
169 | mr r3,r4 | |
170 | blr | |
171 | ||
b5510883 | 172 | END (__strnlen) |
17696087 | 173 | libc_hidden_def (__strnlen) |
b5510883 | 174 | weak_alias (__strnlen, strnlen) |
fe2f79db | 175 | libc_hidden_builtin_def (strnlen) |