1 /* Vector optimized 32/64 bit S/390 version of wcscat.
2 Copyright (C) 2015 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
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.
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.
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/>. */
19 #if defined HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc)
22 # include "asm-syntax.h"
26 /* wchar_t * wcscat (wchar_t *dest, const wchar_t *src)
27 Concatenate two strings.
30 -r0=saved dest pointer for return
42 .machinemode "zarch_nohighgprs"
44 vlbb %v16,0(%r2),6 /* Load s until next 4k-byte boundary. */
45 lcbb %r1,0(%r2),6 /* Get bytes to 4k-byte boundary or 16. */
47 /* __wcslen_c can handle non 4byte aligned pointers,
48 but __wcscpy_c not. Thus if either src or dest is
49 not 4byte aligned, use __wcscat_c. */
50 tmll %r2,3 /* Test if s is 4-byte aligned? */
51 jne .Lfallback /* And use common-code variant if not. */
52 tmll %r3,3 /* Test if src is 4-byte aligned? */
53 jne .Lfallback /* And use common-code variant if not. */
55 lgr %r0,%r2 /* Save destination pointer for return. */
58 r1 = loaded bytes (tmp)
59 r4 = zero byte index (tmp)
63 vfenezf %v16,%v16,%v16 /* Find element not equal with zero search. */
64 vlgvb %r5,%v16,7 /* Load zero index or 16 if not found. */
65 clrjl %r5,%r1,.Llen_end /* Found zero within loaded bytes, end. */
67 /* Align s to 16 byte. */
68 risbgn %r1,%r2,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */
69 lghi %r5,16 /* current_len = 16. */
70 slr %r5,%r1 /* Compute bytes to 16bytes boundary. */
72 /* Find zero in 16byte aligned loop. */
74 vl %v16,0(%r5,%r2) /* Load s. */
75 vfenezfs %v16,%v16,%v16 /* Find element not equal with zero search. */
76 je .Llen_found /* Jump away if zero was found. */
78 vfenezfs %v16,%v16,%v16
81 vfenezfs %v16,%v16,%v16
84 vfenezfs %v16,%v16,%v16
88 j .Llen_loop /* No zero -> loop. */
97 vlgvb %r4,%v16,7 /* Load byte index of zero. */
102 %r1 = loaded bytes (tmp)
103 %r4 = zero byte index (tmp)
104 %r3 = curr src pointer
105 %r2 = curr dst pointer
107 la %r2,0(%r5,%r2) /* strcpy at end of dst-string. */
109 vlbb %v16,0(%r3),6 /* Load s until next 4k-byte boundary. */
110 lcbb %r1,0(%r3),6 /* Get bytes to 4k-byte boundary or 16. */
112 vfenezf %v17,%v16,%v16 /* Find element not equal with zero search. */
113 vlgvb %r5,%v17,7 /* Load zero index or 16 if not found. */
114 clrjl %r5,%r1,.Lcpy_found_align /* If found zero within loaded bytes,
115 copy bytes before and return. */
117 /* Align s to 16 byte. */
118 risbgn %r4,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15. */
119 lghi %r5,15 /* current_len = 15. */
120 slr %r5,%r4 /* Compute highest index to 16byte boundary. */
122 vstl %v16,%r5,0(%r2) /* Copy loaded characters - no zero. */
123 ahi %r5,1 /* Start loop at next character. */
125 /* Find zero in 16byte aligned loop. */
127 vl %v16,0(%r5,%r3) /* Load s. */
128 vfenezfs %v17,%v16,%v16 /* Find element not equal with zero search. */
129 je .Lcpy_found_v16_0 /* Jump away if zero was found. */
130 vl %v18,16(%r5,%r3) /* Load next part of s. */
131 vst %v16,0(%r5,%r2) /* Save previous part without zero to dst. */
132 vfenezfs %v17,%v18,%v18
133 je .Lcpy_found_v18_16
136 vfenezfs %v17,%v16,%v16
137 je .Lcpy_found_v16_32
140 vfenezfs %v17,%v18,%v18
141 je .Lcpy_found_v18_48
145 j .Lcpy_loop /* No zero -> loop. */
151 vlgvb %r1,%v17,7 /* Load byte index of zero. */
152 aghi %r1,3 /* Also copy remaining bytes of zero. */
153 vstl %v16,%r1,0(%r4) /* Copy characters including zero. */
154 lgr %r2,%r0 /* Load saved dest-ptr. */
161 vlgvb %r1,%v17,7 /* Load byte index of zero. */
162 aghi %r1,3 /* Also copy remaining bytes of zero. */
163 vstl %v18,%r1,0(%r4) /* Copy characters including zero. */
164 lgr %r2,%r0 /* Load saved dest-ptr. */
168 aghi %r5,3 /* Also copy remaining bytes of found zero. */
169 vstl %v16,%r5,0(%r2) /* Copy characters including zero. */
170 lgr %r2,%r0 /* Load saved dest-ptr. */
175 #endif /* HAVE_S390_VX_ASM_SUPPORT && IS_IN (libc) */