]>
Commit | Line | Data |
---|---|---|
798f5b4b | 1 | /* Vector optimized 32/64 bit S/390 version of memrchr. |
f7a9f785 | 2 | Copyright (C) 2015-2016 Free Software Foundation, Inc. |
798f5b4b 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 | /* void *memrchr (const void *s, int c, size_t n) | |
27 | Scans memory for character c backwards | |
28 | and returns pointer to first c. | |
29 | ||
30 | Register usage: | |
31 | -r0=tmp | |
32 | -r1=tmp | |
33 | -r2=s | |
34 | -r3=c | |
35 | -r4=n | |
36 | -r5=s in loop | |
37 | ||
38 | -v16=part of s | |
39 | -v17=index of found c | |
40 | -v18=c replicated | |
41 | -v20=permute pattern | |
42 | */ | |
43 | ENTRY(__memrchr_vx) | |
44 | .machine "z13" | |
45 | .machinemode "zarch_nohighgprs" | |
46 | ||
47 | # if !defined __s390x__ | |
48 | llgfr %r4,%r4 | |
49 | # endif /* !defined __s390x__ */ | |
50 | clgije %r4,0,.Lnot_found | |
51 | ||
52 | vlvgb %v18,%r3,0 /* Generate vector which elements are all c. | |
53 | If c > 255, c will be truncated. */ | |
54 | vrepb %v18,%v18,0 | |
55 | ||
56 | llcr %r3,%r3 /* char c_char = (char) c. */ | |
57 | ||
58 | /* check byte n - 1. */ | |
59 | llc %r0,-1(%r4,%r2) | |
60 | slgfi %r4,1 | |
61 | clrje %r0,%r3,.Lfound_end | |
62 | jh .Lnot_found /* Return NULL if n is now 0. */ | |
63 | ||
64 | larl %r1,.Lpermute_mask /* Load permute mask. */ | |
65 | vl %v20,0(%r1) | |
66 | ||
67 | /* check byte n - 2. */ | |
68 | llc %r0,-1(%r4,%r2) | |
69 | slgfi %r4,1 | |
70 | clrje %r0,%r3,.Lfound_end | |
71 | jh .Lnot_found /* Return NULL if n is now 0. */ | |
72 | ||
73 | clgijhe %r4,64,.Lloop64 /* If n >= 64 -> loop64. */ | |
74 | ||
75 | .Llt64: | |
76 | /* Process n < 64 bytes. */ | |
77 | clgijl %r4,16,.Llt16 /* Jump away if n < 16. */ | |
78 | aghi %r4,-16 | |
79 | vl %v16,0(%r4,%r2) | |
80 | vfeebs %v17,%v16,%v18 | |
81 | jno .Lfound0 | |
82 | clgijl %r4,16,.Llt16 | |
83 | aghi %r4,-16 | |
84 | vl %v16,0(%r4,%r2) | |
85 | vfeebs %v17,%v16,%v18 | |
86 | jno .Lfound0 | |
87 | clgijl %r4,16,.Llt16 | |
88 | aghi %r4,-16 | |
89 | vl %v16,0(%r4,%r2) | |
90 | vfeebs %v17,%v16,%v18 | |
91 | jno .Lfound0 | |
92 | .Llt16: | |
93 | clgfi %r4,0 /* if remaining bytes == 0, return NULL. */ | |
94 | locghie %r2,0 | |
95 | ber %r14 | |
96 | ||
97 | aghi %r4,-1 /* vll needs highest index. */ | |
98 | vll %v16,%r4,0(%r2) /* Load remaining bytes. */ | |
99 | ||
100 | /* Right-shift of v16 to mask bytes after highest index. */ | |
101 | lhi %r0,15 | |
102 | slr %r0,%r4 /* Compute byte count for vector shift right. */ | |
103 | sll %r0,3 /* Convert to bit count. */ | |
104 | vlvgb %v17,%r0,7 | |
105 | vsrlb %v16,%v16,%v17 /* Vector shift right by byte by number of bytes | |
106 | specified in bits 1-4 of byte 7 in v17. */ | |
107 | j .Lfound_permute | |
108 | ||
109 | .Lfound48: | |
110 | aghi %r4,16 | |
111 | .Lfound32: | |
112 | aghi %r4,16 | |
113 | .Lfound16: | |
114 | aghi %r4,16 | |
115 | .Lfound0: | |
116 | la %r2,0(%r4,%r2) /* Set pointer to start of v16. */ | |
117 | lghi %r4,15 /* Set highest index in v16 to last index. */ | |
118 | .Lfound_permute: | |
119 | /* Search for a c in v16 in reversed byte order. v16 contains %r4 + 1 | |
120 | bytes. If v16 was not fully loaded, the bytes are already | |
121 | right shifted, so that the bytes in v16 can simply be reversed. */ | |
122 | vperm %v16,%v16,%v16,%v20 /* Permute v16 to reversed order. */ | |
123 | vfeeb %v16,%v16,%v18 /* Find c in reversed v16. */ | |
124 | vlgvb %r1,%v16,7 /* Index of c or 16 if not found. */ | |
125 | ||
126 | /* Return NULL if there is no c in loaded bytes. */ | |
127 | clrjh %r1,%r4,.Lnot_found | |
128 | ||
129 | slgr %r4,%r1 | |
130 | .Lfound_end: | |
131 | la %r2,0(%r4,%r2) /* Return pointer to c. */ | |
132 | br %r14 | |
133 | ||
134 | .Lnot_found: | |
135 | lghi %r2,0 | |
136 | br %r14 | |
137 | ||
138 | .Lpermute_mask: | |
139 | .byte 0x0F,0x0E,0x0D,0x0C,0x0B,0x0A,0x09,0x08 | |
140 | .byte 0x07,0x06,0x05,0x04,0x03,0x02,0x01,0x00 | |
141 | ||
142 | .Lloop64: | |
143 | aghi %r4,-64 | |
144 | vl %v16,48(%r4,%r2) /* Load 16bytes of memory area. */ | |
145 | vfeebs %v17,%v16,%v18 /* Find c. */ | |
146 | jno .Lfound48 /* Jump away if c was found. */ | |
147 | vl %v16,32(%r4,%r2) | |
148 | vfeebs %v17,%v16,%v18 | |
149 | jno .Lfound32 | |
150 | vl %v16,16(%r4,%r2) | |
151 | vfeebs %v17,%v16,%v18 | |
152 | jno .Lfound16 | |
153 | vl %v16,0(%r4,%r2) | |
154 | vfeebs %v17,%v16,%v18 | |
155 | jno .Lfound0 | |
156 | ||
157 | clgijhe %r4,64,.Lloop64 /* If n >= 64 -> loop64. */ | |
158 | j .Llt64 | |
159 | END(__memrchr_vx) | |
160 | #endif /* HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) */ |