]>
Commit | Line | Data |
---|---|---|
f2121601 | 1 | /* Vector Optimized 32/64 bit S/390 version of wmemcmp. |
f7a9f785 | 2 | Copyright (C) 2015-2016 Free Software Foundation, Inc. |
f2121601 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 | /* int wmemcmp (const wchar_t *s1, const wchar_t *s2, size_t n) | |
27 | Compare at most n characters of two wchar_t-arrays. | |
28 | ||
29 | Register usage: | |
30 | -r0=tmp | |
31 | -r1=number of blocks | |
32 | -r2=s1 | |
33 | -r3=s2 | |
34 | -r4=n | |
35 | -r5=current_len | |
36 | -v16=part of s1 | |
37 | -v17=part of s2 | |
38 | -v18=index of unequal | |
39 | */ | |
40 | ENTRY(__wmemcmp_vx) | |
41 | .machine "z13" | |
42 | .machinemode "zarch_nohighgprs" | |
43 | ||
44 | # if !defined __s390x__ | |
45 | llgfr %r4,%r4 | |
46 | # endif /* !defined __s390x__ */ | |
47 | clgije %r4,0,.Lend_equal /* Nothing to do if n == 0. */ | |
48 | ||
49 | /* Check range of maxlen and convert to byte-count. */ | |
50 | # ifdef __s390x__ | |
51 | tmhh %r4,49152 /* Test bit 0 or 1 of maxlen. */ | |
52 | lghi %r1,-4 /* Max byte-count is 18446744073709551612. */ | |
53 | # else | |
54 | tmlh %r4,49152 /* Test bit 0 or 1 of maxlen. */ | |
55 | llilf %r1,4294967292 /* Max byte-count is 4294967292. */ | |
56 | # endif /* !__s390x__ */ | |
57 | sllg %r4,%r4,2 /* Convert character-count to byte-count. */ | |
58 | locgrne %r4,%r1 /* Use max byte-count, if bit 0/1 was one. */ | |
59 | ||
60 | lghi %r5,0 /* current_len = 0. */ | |
61 | ||
62 | clgijh %r4,16,.Lgt16 | |
63 | ||
64 | .Lremaining: | |
65 | aghi %r4,-1 /* vstl needs highest index. */ | |
66 | vll %v16,%r4,0(%r2) | |
67 | vll %v17,%r4,0(%r3) | |
68 | vfenef %v18,%v16,%v17 /* Compare not equal. */ | |
69 | vlgvb %r1,%v18,7 /* Load unequal index or 16 if not found. */ | |
70 | clrj %r1,%r4,12,.Lfound2 /* r1 <= r4 -> unequal within loaded | |
71 | bytes. */ | |
72 | ||
73 | .Lend_equal: | |
74 | lghi %r2,0 | |
75 | br %r14 | |
76 | ||
77 | .Lfound: | |
78 | /* vfenezf found an unequal element or zero. | |
79 | This instruction compares unsigned words, but wchar_t is signed. | |
80 | Thus we have to compare the found element again. */ | |
81 | vlgvb %r1,%v18,7 /* Extract not equal byte-index. */ | |
82 | .Lfound2: | |
83 | srl %r1,2 /* And convert it to character-index. */ | |
84 | vlgvf %r0,%v16,0(%r1) /* Load character-values. */ | |
85 | vlgvf %r1,%v17,0(%r1) | |
86 | cr %r0,%r1 | |
87 | je .Lend_equal | |
88 | lghi %r2,1 | |
89 | lghi %r1,-1 | |
90 | locgrl %r2,%r1 | |
91 | br %r14 | |
92 | ||
93 | .Lgt16: | |
94 | clgijh %r4,64,.Lpreloop64 | |
95 | ||
96 | .Lpreloop16: | |
97 | srlg %r1,%r4,4 /* Split into 16byte blocks */ | |
98 | .Lloop16: | |
99 | vl %v16,0(%r5,%r2) | |
100 | vl %v17,0(%r5,%r3) | |
101 | aghi %r5,16 | |
102 | vfenefs %v18,%v16,%v17 /* Compare not equal. */ | |
103 | jno .Lfound | |
104 | brctg %r1,.Lloop16 /* Loop until all blocks are processed. */ | |
105 | ||
106 | llgfr %r4,%r4 | |
107 | nilf %r4,15 /* Get remaining bytes */ | |
108 | locgre %r2,%r4 | |
109 | ber %r14 | |
110 | la %r2,0(%r5,%r2) | |
111 | la %r3,0(%r5,%r3) | |
112 | j .Lremaining | |
113 | ||
114 | .Lpreloop64: | |
115 | srlg %r1,%r4,6 /* Split into 64byte blocks */ | |
116 | .Lloop64: | |
117 | vl %v16,0(%r5,%r2) | |
118 | vl %v17,0(%r5,%r3) | |
119 | vfenefs %v18,%v16,%v17 /* Compare not equal. */ | |
120 | jno .Lfound | |
121 | ||
122 | vl %v16,16(%r5,%r2) | |
123 | vl %v17,16(%r5,%r3) | |
124 | vfenefs %v18,%v16,%v17 | |
125 | jno .Lfound | |
126 | ||
127 | vl %v16,32(%r5,%r2) | |
128 | vl %v17,32(%r5,%r3) | |
129 | vfenefs %v18,%v16,%v17 | |
130 | jno .Lfound | |
131 | ||
132 | vl %v16,48(%r5,%r2) | |
133 | vl %v17,48(%r5,%r3) | |
134 | aghi %r5,64 | |
135 | vfenefs %v18,%v16,%v17 | |
136 | jno .Lfound | |
137 | ||
138 | brctg %r1,.Lloop64 /* Loop until all blocks are processed. */ | |
139 | ||
140 | llgfr %r4,%r4 | |
141 | nilf %r4,63 /* Get remaining bytes */ | |
142 | locgre %r2,%r4 | |
143 | ber %r14 | |
144 | clgijh %r4,16,.Lpreloop16 | |
145 | la %r2,0(%r5,%r2) | |
146 | la %r3,0(%r5,%r3) | |
147 | j .Lremaining | |
148 | END(__wmemcmp_vx) | |
149 | #endif /* HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) */ |