Blame sysdeps/alpha/strlen.S

Packit Service 82fcde
/* Copyright (C) 1996-2018 Free Software Foundation, Inc.
Packit Service 82fcde
   Contributed by David Mosberger (davidm@cs.arizona.edu).
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 82fcde
/* Finds length of a 0-terminated string.  Optimized for the Alpha
Packit Service 82fcde
   architecture:
Packit Service 82fcde
Packit Service 82fcde
      - memory accessed as aligned quadwords only
Packit Service 82fcde
      - uses cmpbge to compare 8 bytes in parallel
Packit Service 82fcde
      - does binary search to find 0 byte in last quadword (HAKMEM
Packit Service 82fcde
	needed 12 instructions to do this instead of the 8 instructions
Packit Service 82fcde
	that the binary search needs).
Packit Service 82fcde
*/
Packit Service 82fcde
Packit Service 82fcde
#include <sysdep.h>
Packit Service 82fcde
Packit Service 82fcde
	.set noreorder
Packit Service 82fcde
	.set noat
Packit Service 82fcde
Packit Service 82fcde
ENTRY(strlen)
Packit Service 82fcde
#ifdef PROF
Packit Service 82fcde
	ldgp	gp, 0(pv)
Packit Service 82fcde
	lda	AT, _mcount
Packit Service 82fcde
	jsr	AT, (AT), _mcount
Packit Service 82fcde
	.prologue 1
Packit Service 82fcde
#else
Packit Service 82fcde
	.prologue 0
Packit Service 82fcde
#endif
Packit Service 82fcde
Packit Service 82fcde
	ldq_u   t0, 0(a0)	# load first quadword (a0 may be misaligned)
Packit Service 82fcde
	lda     t1, -1(zero)
Packit Service 82fcde
	insqh   t1, a0, t1
Packit Service 82fcde
	andnot  a0, 7, v0
Packit Service 82fcde
	or      t1, t0, t0
Packit Service 82fcde
	nop			# dual issue the next two on ev5
Packit Service 82fcde
	cmpbge  zero, t0, t1	# t1 <- bitmask: bit i == 1 <==> i-th byte == 0
Packit Service 82fcde
	bne     t1, $found
Packit Service 82fcde
Packit Service 82fcde
$loop:	ldq     t0, 8(v0)
Packit Service 82fcde
	addq    v0, 8, v0	# addr += 8
Packit Service 82fcde
	cmpbge  zero, t0, t1
Packit Service 82fcde
	beq     t1, $loop
Packit Service 82fcde
Packit Service 82fcde
$found:	negq    t1, t2		# clear all but least set bit
Packit Service 82fcde
	and     t1, t2, t1
Packit Service 82fcde
Packit Service 82fcde
	and     t1, 0xf0, t2	# binary search for that set bit
Packit Service 82fcde
	and	t1, 0xcc, t3
Packit Service 82fcde
	and	t1, 0xaa, t4
Packit Service 82fcde
	cmovne	t2, 4, t2
Packit Service 82fcde
	cmovne	t3, 2, t3
Packit Service 82fcde
	cmovne	t4, 1, t4
Packit Service 82fcde
	addq	t2, t3, t2
Packit Service 82fcde
	addq	v0, t4, v0
Packit Service 82fcde
	addq	v0, t2, v0
Packit Service 82fcde
	nop			# dual issue next two on ev4 and ev5
Packit Service 82fcde
Packit Service 82fcde
	subq    v0, a0, v0
Packit Service 82fcde
	ret
Packit Service 82fcde
Packit Service 82fcde
	END(strlen)
Packit Service 82fcde
libc_hidden_builtin_def (strlen)