1/* Vector optimized 32/64 bit S/390 version of wcscpy.
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-wcscpy.h>
20#if HAVE_WCSCPY_Z13
21
22# include "sysdep.h"
23# include "asm-syntax.h"
24
25	.text
26
27/* char * wcscpy (const wchar_t *dest, const wchar_t *src)
28   Copy string src to dest.
29
30   Register usage:
31   -r0=border-len for switching to vector-instructions
32   -r1=tmp
33   -r2=dest and return value
34   -r3=src
35   -r4=tmp
36   -r5=current_len
37   -v16=part of src
38   -v17=index of zero
39   -v18=part of src
40*/
41ENTRY(WCSCPY_Z13)
42	.machine "z13"
43	.machinemode "zarch_nohighgprs"
44
45	vlbb	%v16,0(%r3),6	/* Load s until next 4k-byte boundary.  */
46	lcbb	%r1,0(%r3),6	/* Get bytes to 4k-byte boundary or 16.  */
47
48	tmll	%r3,3		/* Test if s is 4-byte aligned?  */
49	jne	.Lfallback	/* And use common-code variant if not.  */
50
51	vfenezf	%v17,%v16,%v16	/* Find element not equal with zero search.  */
52	vlgvb	%r5,%v17,7	/* Load zero index or 16 if not found.  */
53	clrjl	%r5,%r1,.Lfound_align /* If found zero within loaded bytes,
54					 copy bytes before and return.  */
55
56	/* Align s to 16 byte.  */
57	risbgn	%r4,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15.  */
58	lghi	%r5,15		/* current_len = 15.  */
59	slr	%r5,%r4		/* Compute highest index to 16byte boundary.  */
60
61	vstl	%v16,%r5,0(%r2)	/* Copy loaded characters - no zero.  */
62	ahi	%r5,1		/* Start loop at next character.  */
63
64	/* Find zero in 16byte aligned loop.  */
65.Lloop:
66	vl	%v16,0(%r5,%r3)	/* Load s.  */
67	vfenezfs %v17,%v16,%v16	/* Find element not equal with zero search.  */
68	je	.Lfound_v16_0	/* Jump away if zero was found.  */
69	vl	%v18,16(%r5,%r3) /* Load next part of s.  */
70	vst	%v16,0(%r5,%r2)	/* Store previous part without zero to dst.  */
71	vfenezfs %v17,%v18,%v18
72	je	.Lfound_v18_16
73	vl	%v16,32(%r5,%r3)
74	vst	%v18,16(%r5,%r2)
75	vfenezfs %v17,%v16,%v16
76	je	.Lfound_v16_32
77	vl	%v18,48(%r5,%r3)
78	vst	%v16,32(%r5,%r2)
79	vfenezfs %v17,%v18,%v18
80	je	.Lfound_v18_48
81	vst	%v18,48(%r5,%r2)
82
83	aghi	%r5,64
84	j	.Lloop		/* No zero found -> loop.  */
85
86.Lfound_v16_32:
87	aghi	%r5,32
88.Lfound_v16_0:
89	la	%r3,0(%r5,%r2)
90	vlgvb	%r1,%v17,7	/* Load byte index of zero.  */
91	aghi	%r1,3		/* Also copy remaining bytes of zero.  */
92	vstl	%v16,%r1,0(%r3)	/* Copy characters including zero.  */
93	br	%r14
94
95.Lfound_v18_48:
96	aghi	%r5,32
97.Lfound_v18_16:
98	la	%r3,16(%r5,%r2)
99	vlgvb	%r1,%v17,7	/* Load byte index of zero.  */
100	aghi	%r1,3		/* Also copy remaining bytes of zero.  */
101	vstl	%v18,%r1,0(%r3)	/* Copy characters including zero.  */
102	br	%r14
103
104.Lfound_align:
105	aghi	%r5,3		/* Also copy remaining bytes of zero.  */
106	vstl	%v16,%r5,0(%r2)	/* Copy characters including zero.  */
107	br	%r14
108
109.Lfallback:
110	jg	WCSCPY_C
111END(WCSCPY_Z13)
112
113# if ! HAVE_WCSCPY_IFUNC
114strong_alias (WCSCPY_Z13, __wcscpy)
115weak_alias (__wcscpy, wcscpy)
116#  if defined SHARED && IS_IN (libc)
117strong_alias (__wcscpy, __GI___wcscpy)
118#  endif
119# endif
120#endif
121