1 | /* Vector optimized 32/64 bit S/390 version of memccpy. |
2 | Copyright (C) 2015-2022 Free Software Foundation, Inc. |
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 | <https://www.gnu.org/licenses/>. */ |
18 | |
19 | #include <ifunc-memccpy.h> |
20 | |
21 | #if HAVE_MEMCCPY_Z13 |
22 | |
23 | # include "sysdep.h" |
24 | # include "asm-syntax.h" |
25 | |
26 | .text |
27 | |
28 | /* void *memccpy (void * dest, const void *src, int c, size_t n) |
29 | Copies no more than n bytes from src to dest, |
30 | stopping when the character c is found |
31 | and returns pointer next to c in dest or null if c not found. |
32 | |
33 | Register usage: |
34 | -r0=tmp |
35 | -r1=tmp |
36 | -r2=dest |
37 | -r3=src |
38 | -r4=c |
39 | -r5=n |
40 | -r6=current_len |
41 | -v16=part of s |
42 | -v17=index of found c |
43 | -v18=c replicated |
44 | -v19=part #2 of s |
45 | -v31=save area for r6 |
46 | */ |
47 | ENTRY(MEMCCPY_Z13) |
48 | .machine "z13" |
49 | .machinemode "zarch_nohighgprs" |
50 | |
51 | # if !defined __s390x__ |
52 | llgfr %r5,%r5 |
53 | # endif /* !defined __s390x__ */ |
54 | |
55 | vlvgp %v31,%r6,%r7 /* Save registers. */ |
56 | clgije %r5,0,.Lnf_end /* If len == 0 then exit. */ |
57 | |
58 | vlbb %v16,0(%r3),6 /* Load s until next 4k-byte boundary. */ |
59 | lcbb %r0,0(%r3),6 /* Get bytes to 4k-byte boundary or 16. */ |
60 | llgfr %r0,%r0 /* Convert 32bit to 64bit. */ |
61 | |
62 | vlvgb %v18,%r4,0 /* Generate vector which elements are all c. |
63 | if c > 255, c will be truncated. */ |
64 | vrepb %v18,%v18,0 |
65 | lghi %r6,0 /* current_len = 0. */ |
66 | |
67 | clgrjle %r5,%r0,.Lremaining_v16 /* If maxlen <= loaded-bytes |
68 | -> Process remaining. */ |
69 | |
70 | vfeebs %v17,%v16,%v18 /* Find c. */ |
71 | vlgvb %r1,%v17,7 /* Load byte index of c. */ |
72 | clgrjl %r1,%r0,.Lfound_v16 /* Found c is within loaded bytes. */ |
73 | |
74 | /* Align s to 16 byte. */ |
75 | risbgn %r1,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */ |
76 | lghi %r6,15 /* current_len = 15. */ |
77 | slr %r6,%r1 /* Compute highest index to 16byte boundary. */ |
78 | |
79 | vstl %v16,%r6,0(%r2) /* Store prcessed bytes */ |
80 | ahi %r6,1 |
81 | |
82 | .Lpreloop1: |
83 | /* Now we are 16byte aligned, so we can load |
84 | a full vreg without page fault. */ |
85 | vl %v16,0(%r6,%r3) /* Load s. */ |
86 | clgijl %r5,17,.Lremaining_v16 /* If n <= 16, |
87 | process remaining bytes. */ |
88 | lgr %r7,%r5 |
89 | slgfi %r7,16 /* border_len = n - 16. */ |
90 | j .Lloop1 |
91 | |
92 | .Lloop2: |
93 | vl %v16,16(%r6,%r3) |
94 | vst %v19,0(%r6,%r2) |
95 | aghi %r6,16 |
96 | |
97 | .Lloop1: |
98 | clgrjhe %r6,%r7,.Lremaining_v16 /* If current_len >= border |
99 | then process remaining bytes. */ |
100 | vfeebs %v17,%v16,%v18 /* Find c. */ |
101 | jl .Lfound_v16 /* Jump away if c was found. */ |
102 | vl %v19,16(%r6,%r3) /* Load next s part. */ |
103 | vst %v16,0(%r6,%r2) /* Store previous part without c. */ |
104 | aghi %r6,16 |
105 | |
106 | clgrjhe %r6,%r7,.Lremaining_v19 |
107 | vfeebs %v17,%v19,%v18 |
108 | jl .Lfound_v19 |
109 | vl %v16,16(%r6,%r3) |
110 | vst %v19,0(%r6,%r2) |
111 | aghi %r6,16 |
112 | |
113 | clgrjhe %r6,%r7,.Lremaining_v16 |
114 | vfeebs %v17,%v16,%v18 |
115 | jl .Lfound_v16 |
116 | vl %v19,16(%r6,%r3) |
117 | vst %v16,0(%r6,%r2) |
118 | aghi %r6,16 |
119 | |
120 | clgrjhe %r6,%r7,.Lremaining_v19 |
121 | vfeebs %v17,%v19,%v18 |
122 | jo .Lloop2 |
123 | |
124 | .Lfound_v19: |
125 | vlr %v16,%v19 |
126 | .Lfound_v16: |
127 | /* v16 contains c. Store remaining bytes to c. currlen hasnĀ“t |
128 | reached border, thus checking for maxlen is not needed! */ |
129 | vlgvb %r1,%v17,7 /* Load byte index of c. */ |
130 | la %r2,0(%r6,%r2) /* vstl has no support for index-register. */ |
131 | .Lfound_v16_store: |
132 | vstl %v16,%r1,0(%r2) /* Copy bytes including c. */ |
133 | la %r2,1(%r1,%r2) /* Return pointer next to c in dest. */ |
134 | vlgvg %r6,%v31,0 |
135 | vlgvg %r7,%v31,1 |
136 | br %r14 |
137 | |
138 | .Lremaining_v19: |
139 | vlr %v16,%v19 |
140 | .Lremaining_v16: |
141 | /* v16 contains the remaining bytes [1...16]. |
142 | Check and store remaining bytes. */ |
143 | vfeebs %v17,%v16,%v18 |
144 | slgrk %r7,%r5,%r6 /* Remaining bytes = maxlen - current_len. */ |
145 | aghi %r7,-1 /* vstl needs highest index. */ |
146 | la %r2,0(%r6,%r2) /* vstl has no index register. */ |
147 | vlgvb %r1,%v17,7 /* Load index of c or 16 if not found. */ |
148 | /* c in remaining bytes? -> Jump away (c-index <= max-index) */ |
149 | clrjle %r1,%r7,.Lfound_v16_store |
150 | vstl %v16,%r7,0(%r2) /* Store remaining bytes. */ |
151 | |
152 | .Lnf_end: |
153 | vlgvg %r6,%v31,0 |
154 | vlgvg %r7,%v31,1 |
155 | lghi %r2,0 /* Return null. */ |
156 | br %r14 |
157 | END(MEMCCPY_Z13) |
158 | |
159 | # if ! HAVE_MEMCCPY_IFUNC |
160 | strong_alias (MEMCCPY_Z13, __memccpy) |
161 | weak_alias (__memccpy, memccpy) |
162 | # endif |
163 | |
164 | #endif /* HAVE_MEMCCPY_Z13 */ |
165 | |