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.

1 files changed, 100 insertions, 0 deletions

diff --git a/sysdeps/alpha/submul_1.s b/sysdeps/alpha/submul_1.s
new file mode 100644
index 0000000000..acaa11c545
--- /dev/null
+++ b/sysdeps/alpha/submul_1.s
@@ -0,0 +1,100 @@
+ # Alpha 21064 __mpn_submul_1 -- Multiply a limb vector with a limb and
+ # subtract the result from a second limb vector.
+
+ # Copyright (C) 1992, 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
+ # s2_limb	r19
+
+ # This code runs at 42 cycles/limb on the 21064.
+
+ # To improve performance for long multiplications, we would use
+ # 'fetch' for S1 and 'fetch_m' for RES.  It's not obvious how to use
+ # these instructions without slowing down the general code: 1. We can
+ # only have two prefetches in operation at any time in the Alpha
+ # architecture.  2. There will seldom be any special alignment
+ # between RES_PTR and S1_PTR.  Maybe we can simply divide the current
+ # loop into an inner and outer loop, having the inner loop handle
+ # exactly one prefetch block?
+
+	.set	noreorder
+	.set	noat
+.text
+	.align	3
+	.globl	__mpn_submul_1
+	.ent	__mpn_submul_1 2
+__mpn_submul_1:
+	.frame	$30,0,$26
+
+	ldq	$2,0($17)	# $2 = s1_limb
+	addq	$17,8,$17	# s1_ptr++
+	subq	$18,1,$18	# size--
+	mulq	$2,$19,$3	# $3 = prod_low
+	ldq	$5,0($16)	# $5 = *res_ptr
+	umulh	$2,$19,$0	# $0 = prod_high
+	beq	$18,Lend1	# jump if size was == 1
+	ldq	$2,0($17)	# $2 = s1_limb
+	addq	$17,8,$17	# s1_ptr++
+	subq	$18,1,$18	# size--
+	subq	$5,$3,$3
+	cmpult	$5,$3,$4
+	stq	$3,0($16)
+	addq	$16,8,$16	# res_ptr++
+	beq	$18,Lend2	# jump if size was == 2
+
+	.align	3
+Loop:	mulq	$2,$19,$3	# $3 = prod_low
+	ldq	$5,0($16)	# $5 = *res_ptr
+	addq	$4,$0,$0	# cy_limb = cy_limb + 'cy'
+	subq	$18,1,$18	# size--
+	umulh	$2,$19,$4	# $4 = cy_limb
+	ldq	$2,0($17)	# $2 = s1_limb
+	addq	$17,8,$17	# s1_ptr++
+	addq	$3,$0,$3	# $3 = cy_limb + prod_low
+	cmpult	$3,$0,$0	# $0 = carry from (cy_limb + prod_low)
+	subq	$5,$3,$3
+	cmpult	$5,$3,$5
+	stq	$3,0($16)
+	addq	$16,8,$16	# res_ptr++
+	addq	$5,$0,$0	# combine carries
+	bne	$18,Loop
+
+Lend2:	mulq	$2,$19,$3	# $3 = prod_low
+	ldq	$5,0($16)	# $5 = *res_ptr
+	addq	$4,$0,$0	# cy_limb = cy_limb + 'cy'
+	umulh	$2,$19,$4	# $4 = cy_limb
+	addq	$3,$0,$3	# $3 = cy_limb + prod_low
+	cmpult	$3,$0,$0	# $0 = carry from (cy_limb + prod_low)
+	subq	$5,$3,$3
+	cmpult	$5,$3,$5
+	stq	$3,0($16)
+	addq	$5,$0,$0	# combine carries
+	addq	$4,$0,$0	# cy_limb = prod_high + cy
+	ret	$31,($26),1
+Lend1:	subq	$5,$3,$3
+	cmpult	$5,$3,$5
+	stq	$3,0($16)
+	addq	$0,$5,$0
+	ret	$31,($26),1
+
+	.end	__mpn_submul_1