1 | /* Vector optimized 32/64 bit S/390 version of strncat. |
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-strncat.h> |
20 | #if HAVE_STRNCAT_Z13 |
21 | |
22 | # include "sysdep.h" |
23 | # include "asm-syntax.h" |
24 | |
25 | .text |
26 | |
27 | /* char * strncat (const char *dest, const char *src, size_t n) |
28 | Concatenate two strings - at most n characters of src. |
29 | |
30 | Register usage: |
31 | -r0=saved dest pointer for return |
32 | -r1=tmp |
33 | -r2=dest |
34 | -r3=src |
35 | -r4=n |
36 | -r5=current_len |
37 | -r6=tmp |
38 | -r7=tmp |
39 | -v16=part of src |
40 | -v17=index of zero |
41 | -v18=part of src |
42 | -v31=register save area for r6, r7 |
43 | */ |
44 | ENTRY(STRNCAT_Z13) |
45 | .machine "z13" |
46 | .machinemode "zarch_nohighgprs" |
47 | |
48 | # if !defined __s390x__ |
49 | llgfr %r4,%r4 |
50 | # endif /* !defined __s390x__ */ |
51 | |
52 | clgfi %r4,0 |
53 | ber %r14 /* Nothing to do, if n == 0. */ |
54 | lgr %r0,%r2 /* Save destination pointer for return. */ |
55 | vlvgp %v31,%r6,%r7 /* Save registers. */ |
56 | |
57 | /* STRLEN |
58 | %r1 = loaded bytes (tmp) |
59 | %r6 = zero byte index (tmp) |
60 | %r2 = dst |
61 | */ |
62 | vlbb %v16,0(%r2),6 /* Load s until next 4k-byte boundary. */ |
63 | lcbb %r1,0(%r2),6 /* Get bytes to 4k-byte boundary or 16. */ |
64 | |
65 | vfenezb %v16,%v16,%v16 /* Find element not equal with zero search. */ |
66 | vlgvb %r5,%v16,7 /* Load zero index or 16 if not found. */ |
67 | clrjl %r5,%r1,.Llen_end /* Found zero within loaded bytes, end. */ |
68 | |
69 | /* Align s to 16 byte. */ |
70 | risbgn %r1,%r2,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */ |
71 | lghi %r5,16 /* current_len = 16. */ |
72 | slr %r5,%r1 /* Compute bytes to 16bytes boundary. */ |
73 | |
74 | /* Find zero in 16byte aligned loop. */ |
75 | .Llen_loop: |
76 | vl %v16,0(%r5,%r2) /* Load s. */ |
77 | vfenezbs %v16,%v16,%v16 /* Find element not equal with zero search. */ |
78 | je .Llen_found /* Jump away if zero was found. */ |
79 | vl %v16,16(%r5,%r2) |
80 | vfenezbs %v16,%v16,%v16 |
81 | je .Llen_found16 |
82 | vl %v16,32(%r5,%r2) |
83 | vfenezbs %v16,%v16,%v16 |
84 | je .Llen_found32 |
85 | vl %v16,48(%r5,%r2) |
86 | vfenezbs %v16,%v16,%v16 |
87 | je .Llen_found48 |
88 | |
89 | aghi %r5,64 |
90 | j .Llen_loop /* No zero -> loop. */ |
91 | |
92 | .Llen_found48: |
93 | aghi %r5,16 |
94 | .Llen_found32: |
95 | aghi %r5,16 |
96 | .Llen_found16: |
97 | aghi %r5,16 |
98 | .Llen_found: |
99 | vlgvb %r1,%v16,7 /* Load byte index of zero. */ |
100 | algr %r5,%r1 |
101 | |
102 | .Llen_end: |
103 | /* STRCPY |
104 | %r1 = zero byte index (tmp) |
105 | %r6 = loaded bytes (tmp) |
106 | %r3 = curr src pointer |
107 | %r2 = curr dst pointer |
108 | %r7 = border, tmp |
109 | */ |
110 | la %r2,0(%r5,%r2) /* strcpy at end of dst-string. */ |
111 | |
112 | vlbb %v16,0(%r3),6 /* Load s until next 4k-byte boundary. */ |
113 | lcbb %r6,0(%r3),6 /* Get bytes to 4k-byte boundary or 16. */ |
114 | llgfr %r6,%r6 /* Convert 32bit to 64bit. */ |
115 | |
116 | lghi %r5,0 /* current_len = 0. */ |
117 | |
118 | clgrjle %r4,%r6,.Lcpy_remaining_v16 /* If n <= loaded-bytes |
119 | -> process remaining. */ |
120 | |
121 | /* n > loaded-byte-count. */ |
122 | vfenezb %v17,%v16,%v16 /* Find element not equal with zero search. */ |
123 | vlgvb %r1,%v17,7 /* Load zero index or 16 if not found. */ |
124 | clrjl %r1,%r6,.Lcpy_found_v16_store /* Found zero within loaded |
125 | bytes, copy and return. */ |
126 | |
127 | /* Align s to 16 byte. */ |
128 | risbgn %r7,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */ |
129 | lghi %r5,15 /* current_len = 15. */ |
130 | slr %r5,%r7 /* Compute highest index to 16byte boundary. */ |
131 | |
132 | /* Zero not found and n > loaded-byte-count. */ |
133 | vstl %v16,%r5,0(%r2) /* Copy loaded characters - no zero. */ |
134 | ahi %r5,1 /* Start loop at next character. */ |
135 | |
136 | /* |
137 | Now we are 16byte aligned, so we can load a full vreg |
138 | without page fault. |
139 | */ |
140 | lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */ |
141 | aghi %r1,64 |
142 | clgrjl %r1,%r4,.Lcpy_loop64 |
143 | |
144 | vl %v16,0(%r5,%r3) /* Load s. */ |
145 | clgijl %r4,17,.Lcpy_remaining_v16 /* If n <=16, |
146 | process remaining bytes. */ |
147 | .Lcpy_lt64: |
148 | lgr %r7,%r4 |
149 | slgfi %r7,16 /* border_len = n - 16. */ |
150 | |
151 | /* If current_len >= border then process remaining bytes. */ |
152 | clgrjhe %r5,%r7,.Lcpy_remaining_v16 |
153 | vfenezbs %v17,%v16,%v16 /* Find element not equal with zero search. */ |
154 | je .Lcpy_found_v16 /* Jump away if zero was found. */ |
155 | vl %v18,16(%r5,%r3) /* Load next part of s. */ |
156 | vst %v16,0(%r5,%r2) /* Store previous part without zero to dst. */ |
157 | aghi %r5,16 |
158 | |
159 | clgrjhe %r5,%r7,.Lcpy_remaining_v18 |
160 | vfenezbs %v17,%v18,%v18 |
161 | je .Lcpy_found_v18 |
162 | vl %v16,16(%r5,%r3) |
163 | vst %v18,0(%r5,%r2) |
164 | aghi %r5,16 |
165 | |
166 | clgrjhe %r5,%r7,.Lcpy_remaining_v16 |
167 | vfenezbs %v17,%v16,%v16 |
168 | je .Lcpy_found_v16 |
169 | vl %v18,16(%r5,%r3) |
170 | vst %v16,0(%r5,%r2) |
171 | aghi %r5,16 |
172 | |
173 | .Lcpy_remaining_v18: |
174 | vlr %v16,%v18 |
175 | .Lcpy_remaining_v16: |
176 | /* v16 contains the remaining bytes [1...16]. |
177 | Store remaining bytes and append string-termination. */ |
178 | vfenezb %v17,%v16,%v16 /* Find element not equal with zero search. */ |
179 | slgrk %r7,%r4,%r5 /* Remaining bytes = maxlen - current_len. */ |
180 | aghi %r7,-1 /* vstl needs highest index. */ |
181 | vlgvb %r1,%v17,7 /* Load zero index or 16 if not found. */ |
182 | la %r2,0(%r5,%r2) /* vstl has no index register. */ |
183 | /* Zero-index within remaining-bytes, store up to zero and end. */ |
184 | clgrjle %r1,%r7,.Lcpy_found_v16_store |
185 | vstl %v16,%r7,0(%r2) /* Store remaining bytes. */ |
186 | lghi %r1,0 |
187 | stc %r1,1(%r7,%r2) /* Store string-null-termination beyond n. */ |
188 | .Lcpy_end: |
189 | /* Restore saved registers. */ |
190 | vlgvg %r6,%v31,0 |
191 | vlgvg %r7,%v31,1 |
192 | lgr %r2,%r0 /* Load saved dest-ptr. */ |
193 | br %r14 |
194 | |
195 | .Lcpy_found_v16_32: |
196 | aghi %r5,32 |
197 | j .Lcpy_found_v16 |
198 | .Lcpy_found_v18_48: |
199 | aghi %r5,32 |
200 | .Lcpy_found_v18_16: |
201 | aghi %r5,16 |
202 | .Lcpy_found_v18: |
203 | vlr %v16,%v18 |
204 | .Lcpy_found_v16: |
205 | /* v16 contains a zero. Store remaining bytes to zero. current_len |
206 | has not reached border, thus checking for n is not needed! */ |
207 | vlgvb %r1,%v17,7 /* Load byte index of zero. */ |
208 | la %r2,0(%r5,%r2) |
209 | .Lcpy_found_v16_store: |
210 | vstl %v16,%r1,0(%r2) /* Copy characters including zero. */ |
211 | j .Lcpy_end |
212 | |
213 | /* Find zero in 16byte aligned loop. */ |
214 | .Lcpy_loop64: |
215 | vl %v16,0(%r5,%r3) /* Load s. */ |
216 | vfenezbs %v17,%v16,%v16 /* Find element not equal with zero search. */ |
217 | je .Lcpy_found_v16 /* Jump away if zero was found. */ |
218 | vl %v18,16(%r5,%r3) /* Load next part of s. */ |
219 | vst %v16,0(%r5,%r2) /* Store previous part without zero to dst. */ |
220 | vfenezbs %v17,%v18,%v18 |
221 | je .Lcpy_found_v18_16 |
222 | vl %v16,32(%r5,%r3) |
223 | vst %v18,16(%r5,%r2) |
224 | vfenezbs %v17,%v16,%v16 |
225 | je .Lcpy_found_v16_32 |
226 | vl %v18,48(%r5,%r3) |
227 | vst %v16,32(%r5,%r2) |
228 | vfenezbs %v17,%v18,%v18 |
229 | je .Lcpy_found_v18_48 |
230 | vst %v18,48(%r5,%r2) |
231 | |
232 | aghi %r5,64 |
233 | lgr %r1,%r5 /* If %r5 + 64 < maxlen? -> loop64. */ |
234 | aghi %r1,64 |
235 | clgrjl %r1,%r4,.Lcpy_loop64 |
236 | |
237 | vl %v16,0(%r5,%r3) /* Load s. */ |
238 | j .Lcpy_lt64 |
239 | END(STRNCAT_Z13) |
240 | |
241 | # if ! HAVE_STRNCAT_IFUNC |
242 | strong_alias (STRNCAT_Z13, strncat) |
243 | # endif |
244 | |
245 | # if ! HAVE_STRNCAT_C |
246 | /* See string/strncat.c and define STRNCAT_PRIMARY. */ |
247 | strong_alias (STRNCAT_Z13, __strncat) |
248 | # if defined SHARED && IS_IN (libc) |
249 | strong_alias (__strncat, __GI___strncat) |
250 | # endif |
251 | # endif |
252 | #endif |
253 | |