]>
Commit | Line | Data |
---|---|---|
fcf40ebe | 1 | /* Vector optimized 32/64 bit S/390 version of wcsnlen. |
04277e02 | 2 | Copyright (C) 2015-2019 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 | |
5a82c748 | 17 | <https://www.gnu.org/licenses/>. */ |
fcf40ebe | 18 | |
c7e7cd26 SL |
19 | #include <ifunc-wcsnlen.h> |
20 | #if HAVE_WCSNLEN_Z13 | |
fcf40ebe SL |
21 | |
22 | # include "sysdep.h" | |
23 | # include "asm-syntax.h" | |
24 | ||
25 | .text | |
26 | ||
27 | /* size_t wcsnlen (const wchar_t *s, size_t maxlen) | |
28 | Returns the number of characters in s or at most maxlen. | |
29 | ||
30 | Register usage: | |
31 | -r1=tmp | |
32 | -r2=address of string | |
33 | -r3=maxlen (number of characters to be read) | |
34 | -r4=tmp | |
35 | -r5=current_len and return_value | |
36 | -v16=part of s | |
37 | */ | |
c7e7cd26 | 38 | ENTRY(WCSNLEN_Z13) |
fcf40ebe SL |
39 | |
40 | .machine "z13" | |
41 | .machinemode "zarch_nohighgprs" | |
42 | ||
43 | # if !defined __s390x__ | |
44 | llgfr %r3,%r3 | |
45 | # endif /* !defined __s390x__ */ | |
46 | ||
47 | clgfi %r3,0 /* if maxlen == 0, return 0. */ | |
48 | locgre %r2,%r3 | |
49 | ber %r14 | |
50 | ||
51 | vlbb %v16,0(%r2),6 /* Load s until next 4k-byte boundary. */ | |
52 | lcbb %r1,0(%r2),6 /* Get bytes to 4k-byte boundary or 16. */ | |
53 | llgfr %r1,%r1 /* Convert 32bit to 64bit. */ | |
54 | ||
55 | tmll %r2,3 /* Test if s is 4-byte aligned? */ | |
56 | jne .Lfallback /* And use common-code variant if not. */ | |
57 | ||
58 | /* Check range of maxlen and convert to byte-count. */ | |
59 | # ifdef __s390x__ | |
60 | tmhh %r3,49152 /* Test bit 0 or 1 of maxlen. */ | |
61 | lghi %r4,-4 /* Max byte-count is 18446744073709551612. */ | |
62 | # else | |
63 | tmlh %r3,49152 /* Test bit 0 or 1 of maxlen. */ | |
64 | llilf %r4,4294967292 /* Max byte-count is 4294967292. */ | |
65 | # endif /* !__s390x__ */ | |
66 | sllg %r3,%r3,2 /* Convert character-count to byte-count. */ | |
67 | locgrne %r3,%r4 /* Use max byte-count, if bit 0/1 was one. */ | |
68 | ||
69 | vfenezf %v16,%v16,%v16 /* Find element not equal with zero search. */ | |
70 | clgr %r1,%r3 | |
71 | locgrh %r1,%r3 /* loaded_byte_count | |
72 | = min (loaded_byte_count, maxlen) */ | |
73 | ||
74 | vlgvb %r5,%v16,7 /* Load zero index or 16 if not found. */ | |
75 | clrjl %r5,%r1,.Lend /* Found zero within loaded bytes -> return. */ | |
76 | ||
77 | clgr %r1,%r3 /* If loaded_byte_count == maxlen -> end. */ | |
78 | locgre %r5,%r3 | |
79 | je .Lend | |
80 | ||
81 | /* Align s to 16 byte. */ | |
82 | risbgn %r4,%r2,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */ | |
83 | lghi %r5,16 /* current_len = 16. */ | |
84 | slr %r5,%r4 /* Compute bytes to 16bytes boundary. */ | |
85 | ||
86 | lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */ | |
87 | aghi %r1,64 | |
88 | clgrjl %r1,%r3,.Lloop64 | |
89 | ||
90 | /* Find zero in max 64byte with aligned s. */ | |
91 | .Llt64: | |
92 | vl %v16,0(%r5,%r2) /* Load s. */ | |
93 | vfenezfs %v16,%v16,%v16 /* Find element not equal with zero search. */ | |
94 | je .Lfound /* Jump away if zero was found. */ | |
95 | aghi %r5,16 | |
96 | clgrjhe %r5,%r3,.Lfound /* If current_len >= maxlen -> end. */ | |
97 | vl %v16,0(%r5,%r2) | |
98 | vfenezfs %v16,%v16,%v16 | |
99 | je .Lfound | |
100 | aghi %r5,16 | |
101 | clgrjhe %r5,%r3,.Lfound | |
102 | vl %v16,0(%r5,%r2) | |
103 | vfenezfs %v16,%v16,%v16 | |
104 | je .Lfound | |
105 | aghi %r5,16 | |
106 | clgrjhe %r5,%r3,.Lfound | |
107 | vl %v16,0(%r5,%r2) | |
108 | vfenezfs %v16,%v16,%v16 | |
109 | j .Lfound | |
110 | ||
111 | .Lfound48: | |
112 | aghi %r5,16 | |
113 | .Lfound32: | |
114 | aghi %r5,16 | |
115 | .Lfound16: | |
116 | aghi %r5,16 | |
117 | .Lfound: | |
118 | vlgvb %r4,%v16,7 /* Load byte index of zero or 16 if no zero. */ | |
119 | algr %r5,%r4 | |
120 | ||
121 | clgr %r5,%r3 | |
122 | locgrh %r5,%r3 /* Return min (current_len, maxlen). */ | |
123 | .Lend: | |
124 | srlg %r2,%r5,2 /* Convert byte-count to character-count. */ | |
125 | br %r14 | |
126 | ||
127 | /* Find zero in 16byte aligned loop. */ | |
128 | .Lloop64: | |
129 | vl %v16,0(%r5,%r2) /* Load s. */ | |
130 | vfenezfs %v16,%v16,%v16 /* Find element not equal with zero search. */ | |
131 | je .Lfound /* Jump away if zero was found. */ | |
132 | vl %v16,16(%r5,%r2) | |
133 | vfenezfs %v16,%v16,%v16 | |
134 | je .Lfound16 | |
135 | vl %v16,32(%r5,%r2) | |
136 | vfenezfs %v16,%v16,%v16 | |
137 | je .Lfound32 | |
138 | vl %v16,48(%r5,%r2) | |
139 | vfenezfs %v16,%v16,%v16 | |
140 | je .Lfound48 | |
141 | ||
142 | aghi %r5,64 | |
143 | lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */ | |
144 | aghi %r1,64 | |
145 | clgrjl %r1,%r3,.Lloop64 | |
146 | ||
147 | j .Llt64 | |
148 | ||
149 | .Lfallback: | |
c7e7cd26 SL |
150 | jg WCSNLEN_C |
151 | END(WCSNLEN_Z13) | |
152 | ||
153 | # if ! HAVE_WCSNLEN_IFUNC | |
154 | strong_alias (WCSNLEN_Z13, __wcsnlen) | |
155 | weak_alias (__wcsnlen, wcsnlen) | |
156 | # endif | |
157 | #endif |