]>
Commit | Line | Data |
---|---|---|
fcf40ebe | 1 | /* Vector optimized 32/64 bit S/390 version of strnlen. |
f7a9f785 | 2 | Copyright (C) 2015-2016 Free Software Foundation, Inc. |
fcf40ebe SL |
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 | <http://www.gnu.org/licenses/>. */ | |
18 | ||
19 | #if defined HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) | |
20 | ||
21 | # include "sysdep.h" | |
22 | # include "asm-syntax.h" | |
23 | ||
24 | .text | |
25 | ||
26 | /* size_t strnlen (const char *s, size_t maxlen) | |
27 | Returns the number of characters in s or at most maxlen. | |
28 | ||
29 | Register usage: | |
30 | -r1=tmp | |
31 | -r2=address of string | |
32 | -r3=maxlen (number of characters to be read) | |
33 | -r4=tmp | |
34 | -r5=current_len and return_value | |
35 | -v16=part of s | |
36 | */ | |
37 | ENTRY(__strnlen_vx) | |
38 | .machine "z13" | |
39 | .machinemode "zarch_nohighgprs" | |
40 | ||
41 | # if !defined __s390x__ | |
42 | llgfr %r3,%r3 | |
43 | # endif /* !defined __s390x__ */ | |
44 | ||
45 | clgfi %r3,0 /* if maxlen == 0, return 0. */ | |
46 | locgre %r2,%r3 | |
47 | ber %r14 | |
48 | ||
49 | vlbb %v16,0(%r2),6 /* Load s until next 4k-byte boundary. */ | |
50 | lcbb %r1,0(%r2),6 /* Get bytes to 4k-byte boundary or 16. */ | |
51 | llgfr %r1,%r1 /* Convert 32bit to 64bit. */ | |
52 | ||
53 | vfenezb %v16,%v16,%v16 /* Find element not equal with zero search. */ | |
54 | clgr %r1,%r3 | |
55 | locgrh %r1,%r3 /* loaded_byte_count | |
56 | = min (loaded_byte_count, maxlen) */ | |
57 | ||
58 | vlgvb %r5,%v16,7 /* Load zero index or 16 if not found. */ | |
59 | clr %r5,%r1 /* If found zero within loaded bytes? */ | |
60 | locgrl %r2,%r5 /* Then copy return value. */ | |
61 | blr %r14 /* And return. */ | |
62 | ||
63 | clgr %r1,%r3 /* If loaded_byte_count == maxlen? */ | |
64 | locgre %r2,%r3 /* Then copy return value. */ | |
65 | ber %r14 /* And return. */ | |
66 | ||
67 | /* Align s to 16 byte. */ | |
68 | risbgn %r4,%r2,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */ | |
69 | lghi %r5,16 /* current_len = 16. */ | |
70 | slr %r5,%r4 /* Compute bytes to 16bytes boundary. */ | |
71 | ||
72 | lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */ | |
73 | aghi %r1,64 | |
74 | clgrjl %r1,%r3,.Lloop64 | |
75 | ||
76 | /* Find zero in max 64byte with aligned s. */ | |
77 | .Llt64: | |
78 | vl %v16,0(%r5,%r2) /* Load s. */ | |
79 | vfenezbs %v16,%v16,%v16 /* Find element not equal with zero search. */ | |
80 | je .Lfound /* Jump away if zero was found. */ | |
81 | aghi %r5,16 | |
82 | clgrjhe %r5,%r3,.Lfound /* current_len >= maxlen -> end. */ | |
83 | vl %v16,0(%r5,%r2) | |
84 | vfenezbs %v16,%v16,%v16 | |
85 | je .Lfound | |
86 | aghi %r5,16 | |
87 | clgrjhe %r5,%r3,.Lfound | |
88 | vl %v16,0(%r5,%r2) | |
89 | vfenezbs %v16,%v16,%v16 | |
90 | je .Lfound | |
91 | aghi %r5,16 | |
92 | clgrjhe %r5,%r3,.Lfound | |
93 | vl %v16,0(%r5,%r2) | |
94 | vfenezbs %v16,%v16,%v16 | |
95 | j .Lfound | |
96 | ||
97 | .Lfound48: | |
98 | aghi %r5,16 | |
99 | .Lfound32: | |
100 | aghi %r5,16 | |
101 | .Lfound16: | |
102 | aghi %r5,16 | |
103 | .Lfound: | |
104 | vlgvb %r4,%v16,7 /* Load byte index of zero or 16 if no zero. */ | |
105 | algr %r5,%r4 | |
106 | ||
107 | clgr %r5,%r3 | |
108 | locgrh %r5,%r3 /* Return min (current_len, maxlen). */ | |
109 | lgr %r2,%r5 | |
110 | br %r14 | |
111 | ||
112 | /* Find zero in 16 byte aligned loop. */ | |
113 | .Lloop64: | |
114 | vl %v16,0(%r5,%r2) /* Load s. */ | |
115 | vfenezbs %v16,%v16,%v16 /* Find element not equal with zero search. */ | |
116 | je .Lfound /* Jump away if zero was found. */ | |
117 | vl %v16,16(%r5,%r2) | |
118 | vfenezbs %v16,%v16,%v16 | |
119 | je .Lfound16 | |
120 | vl %v16,32(%r5,%r2) | |
121 | vfenezbs %v16,%v16,%v16 | |
122 | je .Lfound32 | |
123 | vl %v16,48(%r5,%r2) | |
124 | vfenezbs %v16,%v16,%v16 | |
125 | je .Lfound48 | |
126 | ||
127 | aghi %r5,64 | |
128 | lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */ | |
129 | aghi %r1,64 | |
130 | clgrjl %r1,%r3,.Lloop64 | |
131 | ||
132 | j .Llt64 | |
133 | END(__strnlen_vx) | |
134 | #endif /* HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) */ |