]>
Commit | Line | Data |
---|---|---|
fe2f79db | 1 | /* Optimized strchrnul implementation for PowerPC64/POWER7 using cmpb insn. |
b168057a | 2 | Copyright (C) 2010-2015 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 PE |
17 | License along with the GNU C Library; if not, see |
18 | <http://www.gnu.org/licenses/>. */ | |
fe2f79db LM |
19 | |
20 | #include <sysdep.h> | |
fe2f79db LM |
21 | |
22 | /* int [r3] strchrnul (char *s [r3], int c [r4]) */ | |
23 | .machine power7 | |
2d67d91a | 24 | ENTRY (__strchrnul) |
fe2f79db LM |
25 | CALL_MCOUNT 2 |
26 | dcbt 0,r3 | |
27 | clrrdi r8,r3,3 /* Align the address to doubleword boundary. */ | |
28 | ||
29 | /* Replicate byte to doubleword. */ | |
664318c3 AM |
30 | insrdi r4,r4,8,48 |
31 | insrdi r4,r4,16,32 | |
fe2f79db LM |
32 | insrdi r4,r4,32,0 |
33 | ||
34 | rlwinm r6,r3,3,26,28 /* Calculate padding. */ | |
35 | ld r12,0(r8) /* Load doubleword from memory. */ | |
36 | li r0,0 /* Doubleword with null chars to use | |
37 | with cmpb. */ | |
38 | ||
39 | /* Now r4 has a doubleword of c bytes and r0 has | |
40 | a doubleword of null bytes. */ | |
41 | ||
42 | cmpb r10,r12,r0 /* Compare each byte against c byte. */ | |
43 | cmpb r9,r12,r4 /* Compare each byte against null byte. */ | |
44 | ||
45 | /* Move the doublewords left and right to discard the bits that are | |
46 | not part of the string and to bring them back as zeros. */ | |
664318c3 AM |
47 | #ifdef __LITTLE_ENDIAN__ |
48 | srd r10,r10,r6 | |
49 | srd r9,r9,r6 | |
50 | sld r10,r10,r6 | |
51 | sld r9,r9,r6 | |
52 | #else | |
fe2f79db LM |
53 | sld r10,r10,r6 |
54 | sld r9,r9,r6 | |
55 | srd r10,r10,r6 | |
56 | srd r9,r9,r6 | |
664318c3 | 57 | #endif |
fe2f79db LM |
58 | or r5,r9,r10 /* OR the results to speed things up. */ |
59 | cmpdi cr7,r5,0 /* If r5 == 0, no c or null bytes | |
60 | have been found. */ | |
61 | bne cr7,L(done) | |
62 | ||
63 | mtcrf 0x01,r8 | |
64 | ||
65 | /* Are we now aligned to a quadword boundary? If so, skip to | |
66 | the main loop. Otherwise, go through the alignment code. */ | |
67 | ||
68 | bt 28,L(loop) | |
69 | ||
70 | /* Handle DWORD2 of pair. */ | |
71 | ldu r12,8(r8) | |
72 | cmpb r10,r12,r0 | |
73 | cmpb r9,r12,r4 | |
74 | or r5,r9,r10 | |
75 | cmpdi cr7,r5,0 | |
76 | bne cr7,L(done) | |
77 | b L(loop) /* We branch here (rather than falling through) | |
78 | to skip the nops due to heavy alignment | |
79 | of the loop below. */ | |
80 | ||
81 | .p2align 5 | |
82 | L(loop): | |
83 | /* Load two doublewords, compare and merge in a | |
84 | single register for speed. This is an attempt | |
85 | to speed up the null-checking process for bigger strings. */ | |
86 | ld r12,8(r8) | |
87 | ldu r11,16(r8) | |
88 | cmpb r10,r12,r0 | |
89 | cmpb r9,r12,r4 | |
90 | cmpb r6,r11,r0 | |
91 | cmpb r7,r11,r4 | |
92 | or r5,r9,r10 | |
93 | or r10,r6,r7 | |
94 | or r11,r5,r10 | |
95 | cmpdi cr7,r11,0 | |
96 | beq cr7,L(loop) | |
97 | ||
98 | /* OK, one (or both) of the doublewords contains a c/null byte. Check | |
99 | the first doubleword and decrement the address in case the first | |
100 | doubleword really contains a c/null byte. */ | |
101 | ||
102 | cmpdi cr6,r5,0 | |
103 | addi r8,r8,-8 | |
104 | bne cr6,L(done) | |
105 | ||
106 | /* The c/null byte must be in the second doubleword. Adjust the | |
664318c3 | 107 | address again and move the result of cmpb to r5 so we can calculate |
fe2f79db LM |
108 | the pointer. */ |
109 | mr r5,r10 | |
110 | addi r8,r8,8 | |
111 | ||
112 | /* r5 has the output of the cmpb instruction, that is, it contains | |
113 | 0xff in the same position as the c/null byte in the original | |
114 | doubleword from the string. Use that to calculate the pointer. */ | |
115 | L(done): | |
664318c3 AM |
116 | #ifdef __LITTLE_ENDIAN__ |
117 | addi r0,r5,-1 | |
118 | andc r0,r0,r5 | |
119 | popcntd r0,r0 | |
120 | #else | |
fe2f79db | 121 | cntlzd r0,r5 /* Count leading zeros before the match. */ |
664318c3 | 122 | #endif |
fe2f79db LM |
123 | srdi r0,r0,3 /* Convert leading zeros to bytes. */ |
124 | add r3,r8,r0 /* Return address of matching c/null byte. */ | |
125 | blr | |
2d67d91a | 126 | END (__strchrnul) |
fe2f79db LM |
127 | weak_alias (__strchrnul,strchrnul) |
128 | libc_hidden_builtin_def (__strchrnul) |