Blame sysdeps/sparc/sparc64/rawmemchr.S

Packit 6c4009
/* rawmemchr (str, ch) -- Return pointer to first occurrence of CH in STR.
Packit 6c4009
   For SPARC v9.
Packit 6c4009
   Copyright (C) 1999-2018 Free Software Foundation, Inc.
Packit 6c4009
   This file is part of the GNU C Library.
Packit 6c4009
   Contributed by Jakub Jelinek <jj@ultra.linux.cz>.
Packit 6c4009
   This version is developed using the same algorithm as the fast C
Packit 6c4009
   version which carries the following introduction:
Packit 6c4009
   Based on strlen implementation by Torbjorn Granlund (tege@sics.se),
Packit 6c4009
   with help from Dan Sahlin (dan@sics.se) and
Packit 6c4009
   commentary by Jim Blandy (jimb@ai.mit.edu);
Packit 6c4009
   adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu),
Packit 6c4009
   and implemented by Roland McGrath (roland@ai.mit.edu).
Packit 6c4009
Packit 6c4009
   The GNU C Library is free software; you can redistribute it and/or
Packit 6c4009
   modify it under the terms of the GNU Lesser General Public
Packit 6c4009
   License as published by the Free Software Foundation; either
Packit 6c4009
   version 2.1 of the License, or (at your option) any later version.
Packit 6c4009
Packit 6c4009
   The GNU C Library is distributed in the hope that it will be useful,
Packit 6c4009
   but WITHOUT ANY WARRANTY; without even the implied warranty of
Packit 6c4009
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
Packit 6c4009
   Lesser General Public License for more details.
Packit 6c4009
Packit 6c4009
   You should have received a copy of the GNU Lesser General Public
Packit 6c4009
   License along with the GNU C Library; if not, see
Packit 6c4009
   <http://www.gnu.org/licenses/>.  */
Packit 6c4009
Packit 6c4009
#include <sysdep.h>
Packit 6c4009
#include <asm/asi.h>
Packit 6c4009
#ifndef XCC
Packit 6c4009
#define XCC xcc
Packit 6c4009
#define USE_BPR
Packit 6c4009
	.register	%g2, #scratch
Packit 6c4009
	.register	%g3, #scratch
Packit 6c4009
#endif
Packit 6c4009
Packit 6c4009
	/* Normally, this uses
Packit 6c4009
	   ((xword - 0x0101010101010101) & 0x8080808080808080) test
Packit 6c4009
	   to find out if any byte in xword could be zero. This is fast, but
Packit 6c4009
	   also gives false alarm for any byte in range 0x81-0xff. It does
Packit 6c4009
	   not matter for correctness, as if this test tells us there could
Packit 6c4009
	   be some zero byte, we check it byte by byte, but if bytes with
Packit 6c4009
	   high bits set are common in the strings, then this will give poor
Packit 6c4009
	   performance. You can #define EIGHTBIT_NOT_RARE and the algorithm
Packit 6c4009
	   will use one tick slower, but more precise test
Packit 6c4009
	   ((xword - 0x0101010101010101) & (~xword) & 0x8080808080808080),
Packit 6c4009
	   which does not give any false alarms (but if some bits are set,
Packit 6c4009
	   one cannot assume from it which bytes are zero and which are not).
Packit 6c4009
	   It is yet to be measured, what is the correct default for glibc
Packit 6c4009
	   in these days for an average user.
Packit 6c4009
	 */
Packit 6c4009
Packit 6c4009
	.text
Packit 6c4009
	.align		32
Packit 6c4009
ENTRY(__rawmemchr)
Packit 6c4009
	and		%o1, 0xff, %o1			/* IEU0		Group		*/
Packit 6c4009
	sethi		%hi(0x01010101), %g1		/* IEU1				*/
Packit 6c4009
	ldub		[%o0], %o3			/* Load				*/
Packit 6c4009
	sll		%o1, 8, %o4			/* IEU0		Group		*/
Packit 6c4009
Packit 6c4009
	or		%g1, %lo(0x01010101), %g1	/* IEU1				*/
Packit 6c4009
	sllx		%g1, 32, %g2			/* IEU0		Group		*/
Packit 6c4009
	or		%o4, %o1, %o4			/* IEU1				*/
Packit 6c4009
	andcc		%o0, 7, %g0			/* IEU1		Group		*/
Packit 6c4009
Packit 6c4009
	sll		%o4, 16, %g5			/* IEU0				*/
Packit 6c4009
	or		%o4, %g5, %o4			/* IEU0		Group		*/
Packit 6c4009
	or		%g1, %g2, %g1			/* IEU1				*/
Packit 6c4009
	bne,pn		%icc, 32f			/* CTI				*/
Packit 6c4009
Packit 6c4009
	 sllx		%o4, 32, %g5			/* IEU0		Group		*/
Packit 6c4009
	cmp		%o3, %o1			/* IEU1				*/
Packit 6c4009
	be,pn		%icc, 30f			/* CTI				*/
Packit 6c4009
	 sllx		%g1, 7, %g2			/* IEU0		Group		*/
Packit 6c4009
Packit 6c4009
18:	ldx		[%o0], %o3			/* Load				*/
Packit 6c4009
	or		%o4, %g5, %o4			/* IEU1				*/
Packit 6c4009
	add		%o0, 8, %o0			/* IEU0		Group		*/
Packit 6c4009
19:	xor		%o3, %o4, %o3			/* IEU0		Group		*/
Packit 6c4009
Packit 6c4009
	sub		%o3, %g1, %o2			/* IEU0		Group		*/
Packit 6c4009
#ifdef EIGHTBIT_NOT_RARE
Packit 6c4009
	andn		%o2, %o3, %o5			/* IEU0		Group		*/
Packit 6c4009
	ldxa		[%o0] ASI_PNF, %o3		/* Load				*/
Packit 6c4009
	andcc		%o5, %g2, %g0			/* IEU1		Group		*/
Packit 6c4009
#else
Packit 6c4009
	ldxa		[%o0] ASI_PNF, %o3		/* Load				*/
Packit 6c4009
	andcc		%o2, %g2, %g0			/* IEU1		Group		*/
Packit 6c4009
#endif
Packit 6c4009
	be,pt		%xcc, 19b			/* CTI				*/
Packit 6c4009
Packit 6c4009
	 add		%o0, 8, %o0			/* IEU0				*/
Packit 6c4009
 	addcc		%o2, %g1, %g3			/* IEU1		Group		*/
Packit 6c4009
	srlx		%o2, 32, %o2			/* IEU0				*/
Packit 6c4009
20:	andcc		%o2, %g2, %g0			/* IEU1		Group		*/
Packit 6c4009
Packit 6c4009
	be,pn		%xcc, 21f			/* CTI				*/
Packit 6c4009
	 srlx		%g3, 56, %o2			/* IEU0				*/
Packit 6c4009
	andcc		%o2, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
	be,pn		%icc, 29f			/* CTI				*/
Packit 6c4009
Packit 6c4009
	 srlx		%g3, 48, %o2			/* IEU0				*/
Packit 6c4009
	andcc		%o2, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
	be,pn		%icc, 28f			/* CTI				*/
Packit 6c4009
	 srlx		%g3, 40, %o2			/* IEU0				*/
Packit 6c4009
Packit 6c4009
	andcc		%o2, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
	be,pn		%icc, 27f			/* CTI				*/
Packit 6c4009
	 srlx		%g3, 32, %o2			/* IEU0				*/
Packit 6c4009
	andcc		%o2, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
Packit 6c4009
	be,pn		%icc, 26f			/* CTI				*/
Packit 6c4009
21:	 srlx		%g3, 24, %o2			/* IEU0				*/
Packit 6c4009
	andcc		%o2, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
	be,pn		%icc, 25f			/* CTI				*/
Packit 6c4009
Packit 6c4009
	 srlx		%g3, 16, %o2			/* IEU0				*/
Packit 6c4009
	andcc		%o2, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
	be,pn		%icc, 24f			/* CTI				*/
Packit 6c4009
	 srlx		%g3, 8, %o2			/* IEU0				*/
Packit 6c4009
Packit 6c4009
	andcc		%o2, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
	be,pn		%icc, 23f			/* CTI				*/
Packit 6c4009
	 xor		%o3, %o4, %o3			/* IEU0				*/
Packit 6c4009
	andcc		%g3, 0xff, %g0			/* IEU1		Group		*/
Packit 6c4009
Packit 6c4009
	be,pn		%icc, 22f			/* CTI				*/
Packit 6c4009
	 sub		%o3, %g1, %o2			/* IEU0				*/
Packit 6c4009
	ldxa		[%o0] ASI_PNF, %o3		/* Load				*/
Packit 6c4009
	andcc		%o2, %g2, %g0			/* IEU1		Group		*/
Packit 6c4009
Packit 6c4009
	be,pt		%xcc, 19b			/* CTI				*/
Packit 6c4009
	 add		%o0, 8, %o0			/* IEU0				*/
Packit 6c4009
	addcc		%o2, %g1, %g3			/* IEU1		Group		*/
Packit 6c4009
	ba,pt		%xcc, 20b			/* CTI				*/
Packit 6c4009
Packit 6c4009
	 srlx		%o2, 32, %o2			/* IEU0				*/
Packit 6c4009
Packit 6c4009
	.align		16
Packit 6c4009
22:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -9, %o0			/* IEU0				*/
Packit 6c4009
23:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -10, %o0			/* IEU0				*/
Packit 6c4009
Packit 6c4009
24:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -11, %o0			/* IEU0				*/
Packit 6c4009
25:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -12, %o0			/* IEU0				*/
Packit 6c4009
Packit 6c4009
26:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -13, %o0			/* IEU0				*/
Packit 6c4009
27:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -14, %o0			/* IEU0				*/
Packit 6c4009
Packit 6c4009
28:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -15, %o0			/* IEU0				*/
Packit 6c4009
29:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 add		%o0, -16, %o0			/* IEU0				*/
Packit 6c4009
Packit 6c4009
30:	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
	 nop						/* IEU0				*/
Packit 6c4009
Packit 6c4009
	.align		16
Packit 6c4009
32:	andcc		%o0, 7, %g0			/* IEU1		Group		*/
Packit 6c4009
	be,a,pn		%icc, 18b			/* CTI				*/
Packit 6c4009
	 sllx		%g1, 7, %g2			/* IEU0				*/
Packit 6c4009
	add		%o0, 1, %o0			/* IEU0		Group		*/
Packit 6c4009
Packit 6c4009
	cmp		%o3, %o1			/* IEU1				*/
Packit 6c4009
	bne,a,pt	%icc, 32b			/* CTI				*/
Packit 6c4009
	 lduba		[%o0] ASI_PNF, %o3		/* Load				*/
Packit 6c4009
	retl						/* CTI+IEU1	Group		*/
Packit 6c4009
Packit 6c4009
	 add		%o0, -1, %o0			/* IEU0				*/
Packit 6c4009
END(__rawmemchr)
Packit 6c4009
Packit 6c4009
libc_hidden_def (__rawmemchr)
Packit 6c4009
weak_alias (__rawmemchr, rawmemchr)