]>
Commit | Line | Data |
---|---|---|
fe2f79db | 1 | /* Optimized strnlen implementation for PowerPC64/POWER7 using cmpb insn. |
2b778ceb | 2 | Copyright (C) 2010-2021 Free Software Foundation, Inc. |
fe2f79db LM |
3 | Contributed by Luis Machado <luisgpm@br.ibm.com>. |
4 | This file is part of the GNU C Library. | |
5 | ||
6 | The GNU C Library is free software; you can redistribute it and/or | |
7 | modify it under the terms of the GNU Lesser General Public | |
8 | License as published by the Free Software Foundation; either | |
9 | version 2.1 of the License, or (at your option) any later version. | |
10 | ||
11 | The GNU C Library is distributed in the hope that it will be useful, | |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | Lesser General Public License for more details. | |
15 | ||
16 | You should have received a copy of the GNU Lesser General Public | |
59ba27a6 | 17 | License along with the GNU C Library; if not, see |
5a82c748 | 18 | <https://www.gnu.org/licenses/>. */ |
fe2f79db LM |
19 | |
20 | #include <sysdep.h> | |
fe2f79db | 21 | |
001649fd WSM |
22 | #ifndef STRNLEN |
23 | # define STRNLEN __strnlen | |
24 | #endif | |
25 | ||
fe2f79db LM |
26 | /* int [r3] strnlen (char *s [r3], int size [r4]) */ |
27 | .machine power7 | |
d5b41185 | 28 | ENTRY_TOCLESS (STRNLEN) |
fe2f79db LM |
29 | CALL_MCOUNT 2 |
30 | dcbt 0,r3 | |
33ee81de | 31 | clrrdi r8,r3,3 |
fe2f79db LM |
32 | add r7,r3,r4 /* Calculate the last acceptable address. */ |
33 | cmpldi r4,32 | |
34 | li r0,0 /* Doubleword with null chars. */ | |
33ee81de AM |
35 | addi r7,r7,-1 |
36 | ||
fe2f79db LM |
37 | /* If we have less than 33 bytes to search, skip to a faster code. */ |
38 | ble L(small_range) | |
39 | ||
fe2f79db LM |
40 | rlwinm r6,r3,3,26,28 /* Calculate padding. */ |
41 | ld r12,0(r8) /* Load doubleword from memory. */ | |
42 | cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */ | |
33ee81de AM |
43 | #ifdef __LITTLE_ENDIAN__ |
44 | srd r10,r10,r6 | |
45 | sld r10,r10,r6 | |
46 | #else | |
fe2f79db LM |
47 | sld r10,r10,r6 |
48 | srd r10,r10,r6 | |
33ee81de | 49 | #endif |
fe2f79db LM |
50 | cmpldi cr7,r10,0 /* If r10 == 0, no null's have been found. */ |
51 | bne cr7,L(done) | |
52 | ||
33ee81de | 53 | clrrdi r7,r7,3 /* Address of last doubleword. */ |
fe2f79db LM |
54 | mtcrf 0x01,r8 |
55 | /* Are we now aligned to a quadword boundary? If so, skip to | |
56 | the main loop. Otherwise, go through the alignment code. */ | |
57 | ||
58 | bt 28,L(loop_setup) | |
59 | ||
60 | /* Handle DWORD2 of pair. */ | |
61 | ldu r12,8(r8) | |
62 | cmpb r10,r12,r0 | |
63 | cmpldi cr7,r10,0 | |
64 | bne cr7,L(done) | |
65 | ||
fe2f79db | 66 | L(loop_setup): |
33ee81de AM |
67 | /* The last dword we want to read in the loop below is the one |
68 | containing the last byte of the string, ie. the dword at | |
69 | (s + size - 1) & ~7, or r7. The first dword read is at | |
70 | r8 + 8, we read 2 * cnt dwords, so the last dword read will | |
71 | be at r8 + 8 + 16 * cnt - 8. Solving for cnt gives | |
72 | cnt = (r7 - r8) / 16 */ | |
73 | sub r5,r7,r8 | |
fe2f79db LM |
74 | srdi r6,r5,4 /* Number of loop iterations. */ |
75 | mtctr r6 /* Setup the counter. */ | |
33ee81de AM |
76 | |
77 | /* Main loop to look for the null byte in the string. Since | |
fe2f79db LM |
78 | it's a small loop (< 8 instructions), align it to 32-bytes. */ |
79 | .p2align 5 | |
80 | L(loop): | |
81 | /* Load two doublewords, compare and merge in a | |
82 | single register for speed. This is an attempt | |
83 | to speed up the null-checking process for bigger strings. */ | |
84 | ||
85 | ld r12,8(r8) | |
86 | ldu r11,16(r8) | |
87 | cmpb r10,r12,r0 | |
88 | cmpb r9,r11,r0 | |
89 | or r5,r9,r10 /* Merge everything in one doubleword. */ | |
90 | cmpldi cr7,r5,0 | |
91 | bne cr7,L(found) | |
92 | bdnz L(loop) | |
33ee81de AM |
93 | |
94 | /* We may have one more dword to read. */ | |
95 | cmpld cr6,r8,r7 | |
96 | beq cr6,L(end_max) | |
97 | ||
98 | ldu r12,8(r8) | |
99 | cmpb r10,r12,r0 | |
100 | cmpldi cr6,r10,0 | |
101 | bne cr6,L(done) | |
fe2f79db LM |
102 | |
103 | L(end_max): | |
33ee81de | 104 | mr r3,r4 |
fe2f79db LM |
105 | blr |
106 | ||
107 | /* OK, one (or both) of the doublewords contains a null byte. Check | |
108 | the first doubleword and decrement the address in case the first | |
109 | doubleword really contains a null byte. */ | |
110 | .align 4 | |
111 | L(found): | |
112 | cmpldi cr6,r10,0 | |
113 | addi r8,r8,-8 | |
114 | bne cr6,L(done) | |
115 | ||
116 | /* The null byte must be in the second doubleword. Adjust the address | |
117 | again and move the result of cmpb to r10 so we can calculate the | |
118 | length. */ | |
119 | ||
120 | mr r10,r9 | |
121 | addi r8,r8,8 | |
122 | ||
123 | /* r10 has the output of the cmpb instruction, that is, it contains | |
124 | 0xff in the same position as the null byte in the original | |
125 | doubleword from the string. Use that to calculate the length. | |
33ee81de AM |
126 | We need to make sure the null char is *before* the end of the |
127 | range. */ | |
fe2f79db | 128 | L(done): |
33ee81de AM |
129 | #ifdef __LITTLE_ENDIAN__ |
130 | addi r0,r10,-1 | |
131 | andc r0,r0,r10 | |
132 | popcntd r0,r0 | |
133 | #else | |
134 | cntlzd r0,r10 /* Count leading zeros before the match. */ | |
135 | #endif | |
136 | sub r3,r8,r3 | |
137 | srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */ | |
138 | add r3,r3,r0 /* Length until the match. */ | |
139 | cmpld r3,r4 | |
140 | blelr | |
141 | mr r3,r4 | |
fe2f79db LM |
142 | blr |
143 | ||
144 | /* Deals with size <= 32. */ | |
145 | .align 4 | |
146 | L(small_range): | |
147 | cmpldi r4,0 | |
33ee81de AM |
148 | beq L(end_max) |
149 | ||
150 | clrrdi r7,r7,3 /* Address of last doubleword. */ | |
fe2f79db LM |
151 | |
152 | rlwinm r6,r3,3,26,28 /* Calculate padding. */ | |
33ee81de | 153 | ld r12,0(r8) /* Load doubleword from memory. */ |
fe2f79db | 154 | cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */ |
33ee81de AM |
155 | #ifdef __LITTLE_ENDIAN__ |
156 | srd r10,r10,r6 | |
157 | sld r10,r10,r6 | |
158 | #else | |
fe2f79db LM |
159 | sld r10,r10,r6 |
160 | srd r10,r10,r6 | |
33ee81de | 161 | #endif |
fe2f79db LM |
162 | cmpldi cr7,r10,0 |
163 | bne cr7,L(done) | |
164 | ||
33ee81de AM |
165 | cmpld r8,r7 |
166 | beq L(end_max) | |
fe2f79db LM |
167 | |
168 | .p2align 5 | |
169 | L(loop_small): | |
170 | ldu r12,8(r8) | |
171 | cmpb r10,r12,r0 | |
fe2f79db LM |
172 | cmpldi cr6,r10,0 |
173 | bne cr6,L(done) | |
33ee81de AM |
174 | cmpld r8,r7 |
175 | bne L(loop_small) | |
176 | mr r3,r4 | |
177 | blr | |
178 | ||
001649fd | 179 | END (STRNLEN) |
14362ef1 | 180 | libc_hidden_def (__strnlen) |
2d67d91a | 181 | weak_alias (__strnlen, strnlen) |
14362ef1 | 182 | libc_hidden_def (strnlen) |