Blame sysdeps/s390/wcscpy-vx.S

Packit Service 82fcde
/* Vector optimized 32/64 bit S/390 version of wcscpy.
Packit Service 82fcde
   Copyright (C) 2015-2018 Free Software Foundation, Inc.
Packit Service 82fcde
   This file is part of the GNU C Library.
Packit Service 82fcde
Packit Service 82fcde
   The GNU C Library is free software; you can redistribute it and/or
Packit Service 82fcde
   modify it under the terms of the GNU Lesser General Public
Packit Service 82fcde
   License as published by the Free Software Foundation; either
Packit Service 82fcde
   version 2.1 of the License, or (at your option) any later version.
Packit Service 82fcde
Packit Service 82fcde
   The GNU C Library is distributed in the hope that it will be useful,
Packit Service 82fcde
   but WITHOUT ANY WARRANTY; without even the implied warranty of
Packit Service 82fcde
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
Packit Service 82fcde
   Lesser General Public License for more details.
Packit Service 82fcde
Packit Service 82fcde
   You should have received a copy of the GNU Lesser General Public
Packit Service 82fcde
   License along with the GNU C Library; if not, see
Packit Service 82fcde
   <http://www.gnu.org/licenses/>.  */
Packit Service 82fcde
Packit Service 2f485d
#include <ifunc-wcscpy.h>
Packit Service 2f485d
#if HAVE_WCSCPY_Z13
Packit Service 82fcde
Packit Service 82fcde
# include "sysdep.h"
Packit Service 82fcde
# include "asm-syntax.h"
Packit Service 82fcde
Packit Service 82fcde
	.text
Packit Service 82fcde
Packit Service 82fcde
/* char * wcscpy (const wchar_t *dest, const wchar_t *src)
Packit Service 82fcde
   Copy string src to dest.
Packit Service 82fcde
Packit Service 82fcde
   Register usage:
Packit Service 82fcde
   -r0=border-len for switching to vector-instructions
Packit Service 82fcde
   -r1=tmp
Packit Service 82fcde
   -r2=dest and return value
Packit Service 82fcde
   -r3=src
Packit Service 82fcde
   -r4=tmp
Packit Service 82fcde
   -r5=current_len
Packit Service 82fcde
   -v16=part of src
Packit Service 82fcde
   -v17=index of zero
Packit Service 82fcde
   -v18=part of src
Packit Service 82fcde
*/
Packit Service 2f485d
ENTRY(WCSCPY_Z13)
Packit Service 82fcde
	.machine "z13"
Packit Service 82fcde
	.machinemode "zarch_nohighgprs"
Packit Service 82fcde
Packit Service 82fcde
	vlbb	%v16,0(%r3),6	/* Load s until next 4k-byte boundary.  */
Packit Service 82fcde
	lcbb	%r1,0(%r3),6	/* Get bytes to 4k-byte boundary or 16.  */
Packit Service 82fcde
Packit Service 82fcde
	tmll	%r3,3		/* Test if s is 4-byte aligned?  */
Packit Service 82fcde
	jne	.Lfallback	/* And use common-code variant if not.  */
Packit Service 82fcde
Packit Service 82fcde
	vfenezf	%v17,%v16,%v16	/* Find element not equal with zero search.  */
Packit Service 82fcde
	vlgvb	%r5,%v17,7	/* Load zero index or 16 if not found.  */
Packit Service 82fcde
	clrjl	%r5,%r1,.Lfound_align /* If found zero within loaded bytes,
Packit Service 82fcde
					 copy bytes before and return.  */
Packit Service 82fcde
Packit Service 82fcde
	/* Align s to 16 byte.  */
Packit Service 82fcde
	risbgn	%r4,%r3,60,128+63,0 /* %r3 = bits 60-63 of %r2 'and' 15.  */
Packit Service 82fcde
	lghi	%r5,15		/* current_len = 15.  */
Packit Service 82fcde
	slr	%r5,%r4		/* Compute highest index to 16byte boundary.  */
Packit Service 82fcde
Packit Service 82fcde
	vstl	%v16,%r5,0(%r2)	/* Copy loaded characters - no zero.  */
Packit Service 82fcde
	ahi	%r5,1		/* Start loop at next character.  */
Packit Service 82fcde
Packit Service 82fcde
	/* Find zero in 16byte aligned loop.  */
Packit Service 82fcde
.Lloop:
Packit Service 82fcde
	vl	%v16,0(%r5,%r3)	/* Load s.  */
Packit Service 82fcde
	vfenezfs %v17,%v16,%v16	/* Find element not equal with zero search.  */
Packit Service 82fcde
	je	.Lfound_v16_0	/* Jump away if zero was found.  */
Packit Service 82fcde
	vl	%v18,16(%r5,%r3) /* Load next part of s.  */
Packit Service 82fcde
	vst	%v16,0(%r5,%r2)	/* Store previous part without zero to dst.  */
Packit Service 82fcde
	vfenezfs %v17,%v18,%v18
Packit Service 82fcde
	je	.Lfound_v18_16
Packit Service 82fcde
	vl	%v16,32(%r5,%r3)
Packit Service 82fcde
	vst	%v18,16(%r5,%r2)
Packit Service 82fcde
	vfenezfs %v17,%v16,%v16
Packit Service 82fcde
	je	.Lfound_v16_32
Packit Service 82fcde
	vl	%v18,48(%r5,%r3)
Packit Service 82fcde
	vst	%v16,32(%r5,%r2)
Packit Service 82fcde
	vfenezfs %v17,%v18,%v18
Packit Service 82fcde
	je	.Lfound_v18_48
Packit Service 82fcde
	vst	%v18,48(%r5,%r2)
Packit Service 82fcde
Packit Service 82fcde
	aghi	%r5,64
Packit Service 82fcde
	j	.Lloop		/* No zero found -> loop.  */
Packit Service 82fcde
Packit Service 82fcde
.Lfound_v16_32:
Packit Service 82fcde
	aghi	%r5,32
Packit Service 82fcde
.Lfound_v16_0:
Packit Service 82fcde
	la	%r3,0(%r5,%r2)
Packit Service 82fcde
	vlgvb	%r1,%v17,7	/* Load byte index of zero.  */
Packit Service 82fcde
	aghi	%r1,3		/* Also copy remaining bytes of zero.  */
Packit Service 82fcde
	vstl	%v16,%r1,0(%r3)	/* Copy characters including zero.  */
Packit Service 82fcde
	br	%r14
Packit Service 82fcde
Packit Service 82fcde
.Lfound_v18_48:
Packit Service 82fcde
	aghi	%r5,32
Packit Service 82fcde
.Lfound_v18_16:
Packit Service 82fcde
	la	%r3,16(%r5,%r2)
Packit Service 82fcde
	vlgvb	%r1,%v17,7	/* Load byte index of zero.  */
Packit Service 82fcde
	aghi	%r1,3		/* Also copy remaining bytes of zero.  */
Packit Service 82fcde
	vstl	%v18,%r1,0(%r3)	/* Copy characters including zero.  */
Packit Service 82fcde
	br	%r14
Packit Service 82fcde
Packit Service 82fcde
.Lfound_align:
Packit Service 82fcde
	aghi	%r5,3		/* Also copy remaining bytes of zero.  */
Packit Service 82fcde
	vstl	%v16,%r5,0(%r2)	/* Copy characters including zero.  */
Packit Service 82fcde
	br	%r14
Packit Service 82fcde
Packit Service 82fcde
.Lfallback:
Packit Service 2f485d
	jg	WCSCPY_C
Packit Service 2f485d
END(WCSCPY_Z13)
Packit Service 2f485d
Packit Service 2f485d
# if ! HAVE_WCSCPY_IFUNC
Packit Service 2f485d
strong_alias (WCSCPY_Z13, wcscpy)
Packit Service 2f485d
# endif
Packit Service 2f485d
#endif