glibc/sysdeps/alpha/lshift.s
Roland McGrath 8f5ca04bc7 Sat Oct 14 02:52:36 1995 Ulrich Drepper <drepper@ipd.info.uni-karlsruhe.de>
* malloc/malloc.c (_malloc_internal): Performance fix.  Move
	if statement out of loop.

	* stdio/_itoa.c, stdio/_itoa.h: Complete rewrite.  Much faster
	implementation using GMP functions.  Contributed by
	Torbjorn Granlund and Ulrich Drepper.

	* stdio/test_rdwr.c: Include <errno.h>.

	* sysdeps/i386/i586/Implies: New file.

	New highly optimized string functions for i[345]86.
	* sysdeps/i386/memchr.S, sysdeps/i386/memcmp.S: New files.
        * sysdeps/i386/stpcpy.S, sysdeps/i386/stpncpy.S: New files.
        * sysdeps/i386/strchr.S, sysdeps/i386/strcspn.S: New files.
        * sysdeps/i386/strpbrk.S, sysdeps/i386/strrchr.S: New files.
        * sysdeps/i386/strspn.S, sysdeps/i386/i486/strcat.S: New files.
        * sysdeps/i386/i486/strlen.S, sysdeps/i386/i586/strchr.S: New files.
        * sysdeps/i386/i586/strlen.S: New file.
	* sysdeps/i386/memchr.c: Removed.  There is now an assembler version.

	* sysdeps/i386/i586/memcopy.h (WORD_COPY_BWD): Parameters did
	not correspond to used values.

	* sysdeps/unix/sysv/linux/nfs/nfs.h: New file.  Simply a wrapper
        around a kernel header file.
	* sysdeps/unix/sysv/linux/Dist: Add it.
	* sysdeps/unix/sysv/linux/Makefile [$(subdir)=sunrpc] (headers):
	Likewise.

	* sysdeps/unix/sysv/linux/local_lim.h: Rewrite.  Instead of
        defining ourself we use a kernel header file.

	* sysdeps/unix/sysv/linux/i386/sysdep.h (DO_CALL): Optimize system
        call handler for i586.

	* sysdeps/unix/sysv/linux/sys/param.h: Add copyright and clean up.
Sat Oct 14 02:52:36 1995  Ulrich Drepper  <drepper@ipd.info.uni-karlsruhe.de>

	* malloc/malloc.c (_malloc_internal): Performance fix.  Move
	if statement out of loop.

	* stdio/_itoa.c, stdio/_itoa.h: Complete rewrite.  Much faster
	implementation using GMP functions.  Contributed by
	Torbjorn Granlund and Ulrich Drepper.

	* stdio/test_rdwr.c: Include <errno.h>.

	* sysdeps/i386/i586/Implies: New file.

	New highly optimized string functions for i[345]86.
	* sysdeps/i386/memchr.S, sysdeps/i386/memcmp.S: New files.
        * sysdeps/i386/stpcpy.S, sysdeps/i386/stpncpy.S: New files.
        * sysdeps/i386/strchr.S, sysdeps/i386/strcspn.S: New files.
        * sysdeps/i386/strpbrk.S, sysdeps/i386/strrchr.S: New files.
        * sysdeps/i386/strspn.S, sysdeps/i386/i486/strcat.S: New files.
        * sysdeps/i386/i486/strlen.S, sysdeps/i386/i586/strchr.S: New files.
        * sysdeps/i386/i586/strlen.S: New file.
	* sysdeps/i386/memchr.c: Removed.  There is now an assembler version.

	* sysdeps/i386/i586/memcopy.h (WORD_COPY_BWD): Parameters did
	not correspond to used values.

	* sysdeps/unix/sysv/linux/nfs/nfs.h: New file.  Simply a wrapper
        around a kernel header file.
	* sysdeps/unix/sysv/linux/Dist: Add it.
	* sysdeps/unix/sysv/linux/Makefile [$(subdir)=sunrpc] (headers):
	Likewise.

	* sysdeps/unix/sysv/linux/local_lim.h: Rewrite.  Instead of
        defining ourself we use a kernel header file.

	* sysdeps/unix/sysv/linux/i386/sysdep.h (DO_CALL): Optimize system
        call handler for i586.

	* sysdeps/unix/sysv/linux/sys/param.h: Add copyright and clean up.
1995-10-16 01:37:51 +00:00

109 lines
2.5 KiB
ArmAsm

# Alpha 21064 __mpn_lshift --
# Copyright (C) 1994, 1995 Free Software Foundation, Inc.
# This file is part of the GNU MP Library.
# The GNU MP Library is free software; you can redistribute it and/or modify
# it under the terms of the GNU Library General Public License as published by
# the Free Software Foundation; either version 2 of the License, or (at your
# option) any later version.
# The GNU MP Library is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
# License for more details.
# You should have received a copy of the GNU Library General Public License
# along with the GNU MP Library; see the file COPYING.LIB. If not, write to
# the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
# INPUT PARAMETERS
# res_ptr r16
# s1_ptr r17
# size r18
# cnt r19
# This code runs at 4.8 cycles/limb on the 21064. With infinite unrolling,
# it would take 4 cycles/limb. It should be possible to get down to 3
# cycles/limb since both ldq and stq can be paired with the other used
# instructions. But there are many restrictions in the 21064 pipeline that
# makes it hard, if not impossible, to get down to 3 cycles/limb:
# 1. ldq has a 3 cycle delay, srl and sll have a 2 cycle delay.
# 2. Only aligned instruction pairs can be paired.
# 3. The store buffer or silo might not be able to deal with the bandwidth.
.set noreorder
.set noat
.text
.align 3
.globl __mpn_lshift
.ent __mpn_lshift
__mpn_lshift:
.frame $30,0,$26,0
s8addq $18,$17,$17 # make r17 point at end of s1
ldq $4,-8($17) # load first limb
subq $17,8,$17
subq $31,$19,$7
s8addq $18,$16,$16 # make r16 point at end of RES
subq $18,1,$18
and $18,4-1,$20 # number of limbs in first loop
srl $4,$7,$0 # compute function result
beq $20,L0
subq $18,$20,$18
.align 3
Loop0:
ldq $3,-8($17)
subq $16,8,$16
subq $17,8,$17
subq $20,1,$20
sll $4,$19,$5
srl $3,$7,$6
bis $3,$3,$4
bis $5,$6,$8
stq $8,0($16)
bne $20,Loop0
L0: beq $18,Lend
.align 3
Loop: ldq $3,-8($17)
subq $16,32,$16
subq $18,4,$18
sll $4,$19,$5
srl $3,$7,$6
ldq $4,-16($17)
sll $3,$19,$1
bis $5,$6,$8
stq $8,24($16)
srl $4,$7,$2
ldq $3,-24($17)
sll $4,$19,$5
bis $1,$2,$8
stq $8,16($16)
srl $3,$7,$6
ldq $4,-32($17)
sll $3,$19,$1
bis $5,$6,$8
stq $8,8($16)
srl $4,$7,$2
subq $17,32,$17
bis $1,$2,$8
stq $8,0($16)
bgt $18,Loop
Lend: sll $4,$19,$8
stq $8,-8($16)
ret $31,($26),1
.end __mpn_lshift