move .mul and .umul into the kernel. if v8 sparc is detected, replace with

the equivelant 2-instruction sequence.

5 files changed, 286 insertions, 358 deletions

diff --git a/sys/arch/sparc/sparc/autoconf.c b/sys/arch/sparc/sparc/autoconf.c
index 68afc63f0d8..814e1bc83d7 100644
--- a/sys/arch/sparc/sparc/autoconf.c
+++ b/sys/arch/sparc/sparc/autoconf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: autoconf.c,v 1.31 1999/09/03 18:01:57 art Exp $	*/
+/*	$OpenBSD: autoconf.c,v 1.32 1999/12/07 00:49:07 deraadt Exp $	*/
 /*	$NetBSD: autoconf.c,v 1.73 1997/07/29 09:41:53 fair Exp $ */
 
 /*
@@ -286,6 +286,15 @@ bootstrap()
 		extern void setpte4m __P((u_int, u_int));
 		extern struct timer_4m *timerreg_4m;
 		extern struct counter_4m *counterreg_4m;
+		extern void *_umulreplace, *_umul, *_umulreplace_end;
+		extern void *_mulreplace, *_mul, *_mulreplace_end;
+
+		/*
+		 * Whack the slow sun4/sun4c umul/mul functions with
+		 * fast V8 ones
+		 */
+		bcopy(_umulreplace, _umul, _umulreplace_end-_umulreplace);
+		bcopy(_mulreplace, _mul, _mulreplace_end-_mulreplace);
 
 		if ((node = opennode("/obio/interrupt")) == 0)
 		    if ((node=search_prom(findroot(),"interrupt"))==0)
diff --git a/sys/arch/sparc/sparc/locore.s b/sys/arch/sparc/sparc/locore.s
index c97c659d2ff..2b98b97deb6 100644
--- a/sys/arch/sparc/sparc/locore.s
+++ b/sys/arch/sparc/sparc/locore.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.s,v 1.27 1999/11/13 00:11:52 deraadt Exp $	*/
+/*	$OpenBSD: locore.s,v 1.28 1999/12/07 00:49:07 deraadt Exp $	*/
 /*	$NetBSD: locore.s,v 1.73 1997/09/13 20:36:48 pk Exp $	*/
 
 /*
@@ -5895,6 +5895,278 @@ ENTRY(ffs)
 	add	%o0, 24, %o0
 
 /*
+ * V8 sparc mul/umul replacements.
+ */
+.globl __mulreplace, __mulreplace_end
+__mulreplace:
+	retl
+	 smul	%o0, %o1, %o0
+__mulreplace_end:
+
+.globl __umulreplace, __umulreplace_end
+__umulreplace:
+	retl
+	 umul	%o0, %o1, %o0
+__umulreplace_end:
+
+/*
+ * Signed multiply, from Appendix E of the Sparc Version 8
+ * Architecture Manual.
+ *
+ * Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the upper 32 bits of
+ * the 64-bit product).
+ *
+ * This code optimizes short (less than 13-bit) multiplies.
+ */
+.globl .mul, __mul
+.mul:
+__mul:
+	mov	%o0, %y		! multiplier -> Y
+	andncc	%o0, 0xfff, %g0	! test bits 12..31
+	be	Lmul_shortway	! if zero, can do it the short way
+	 andcc	%g0, %g0, %o4	! zero the partial product and clear N and V
+
+	/*
+	 * Long multiply.  32 steps, followed by a final shift step.
+	 */
+	mulscc	%o4, %o1, %o4	! 1
+	mulscc	%o4, %o1, %o4	! 2
+	mulscc	%o4, %o1, %o4	! 3
+	mulscc	%o4, %o1, %o4	! 4
+	mulscc	%o4, %o1, %o4	! 5
+	mulscc	%o4, %o1, %o4	! 6
+	mulscc	%o4, %o1, %o4	! 7
+	mulscc	%o4, %o1, %o4	! 8
+	mulscc	%o4, %o1, %o4	! 9
+	mulscc	%o4, %o1, %o4	! 10
+	mulscc	%o4, %o1, %o4	! 11
+	mulscc	%o4, %o1, %o4	! 12
+	mulscc	%o4, %o1, %o4	! 13
+	mulscc	%o4, %o1, %o4	! 14
+	mulscc	%o4, %o1, %o4	! 15
+	mulscc	%o4, %o1, %o4	! 16
+	mulscc	%o4, %o1, %o4	! 17
+	mulscc	%o4, %o1, %o4	! 18
+	mulscc	%o4, %o1, %o4	! 19
+	mulscc	%o4, %o1, %o4	! 20
+	mulscc	%o4, %o1, %o4	! 21
+	mulscc	%o4, %o1, %o4	! 22
+	mulscc	%o4, %o1, %o4	! 23
+	mulscc	%o4, %o1, %o4	! 24
+	mulscc	%o4, %o1, %o4	! 25
+	mulscc	%o4, %o1, %o4	! 26
+	mulscc	%o4, %o1, %o4	! 27
+	mulscc	%o4, %o1, %o4	! 28
+	mulscc	%o4, %o1, %o4	! 29
+	mulscc	%o4, %o1, %o4	! 30
+	mulscc	%o4, %o1, %o4	! 31
+	mulscc	%o4, %o1, %o4	! 32
+	mulscc	%o4, %g0, %o4	! final shift
+
+	! If %o0 was negative, the result is
+	!	(%o0 * %o1) + (%o1 << 32))
+	! We fix that here.
+
+	tst	%o0
+	bge	1f
+	 rd	%y, %o0
+
+	! %o0 was indeed negative; fix upper 32 bits of result by subtracting 
+	! %o1 (i.e., return %o4 - %o1 in %o1).
+	retl
+	 sub	%o4, %o1, %o1
+
+1:
+	retl
+	 mov	%o4, %o1
+
+Lmul_shortway:
+	/*
+	 * Short multiply.  12 steps, followed by a final shift step.
+	 * The resulting bits are off by 12 and (32-12) = 20 bit positions,
+	 * but there is no problem with %o0 being negative (unlike above).
+	 */
+	mulscc	%o4, %o1, %o4	! 1
+	mulscc	%o4, %o1, %o4	! 2
+	mulscc	%o4, %o1, %o4	! 3
+	mulscc	%o4, %o1, %o4	! 4
+	mulscc	%o4, %o1, %o4	! 5
+	mulscc	%o4, %o1, %o4	! 6
+	mulscc	%o4, %o1, %o4	! 7
+	mulscc	%o4, %o1, %o4	! 8
+	mulscc	%o4, %o1, %o4	! 9
+	mulscc	%o4, %o1, %o4	! 10
+	mulscc	%o4, %o1, %o4	! 11
+	mulscc	%o4, %o1, %o4	! 12
+	mulscc	%o4, %g0, %o4	! final shift
+
+	/*
+	 *  %o4 has 20 of the bits that should be in the low part of the
+	 * result; %y has the bottom 12 (as %y's top 12).  That is:
+	 *
+	 *	  %o4		    %y
+	 * +----------------+----------------+
+	 * | -12- |   -20-  | -12- |   -20-  |
+	 * +------(---------+------)---------+
+	 *  --hi-- ----low-part----
+	 *
+	 * The upper 12 bits of %o4 should be sign-extended to form the
+	 * high part of the product (i.e., highpart = %o4 >> 20).
+	 */
+
+	rd	%y, %o5
+	sll	%o4, 12, %o0	! shift middle bits left 12
+	srl	%o5, 20, %o5	! shift low bits right 20, zero fill at left
+	or	%o5, %o0, %o0	! construct low part of result
+	retl
+	 sra	%o4, 20, %o1	! ... and extract high part of result
+
+/*
+ * Unsigned multiply.  Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the
+ * upper 32 bits of the 64-bit product).
+ *
+ * This code optimizes short (less than 13-bit) multiplies.  Short
+ * multiplies require 25 instruction cycles, and long ones require
+ * 45 instruction cycles.
+ *
+ * On return, overflow has occurred (%o1 is not zero) if and only if
+ * the Z condition code is clear, allowing, e.g., the following:
+ *
+ *	call	.umul
+ *	nop
+ *	bnz	overflow	(or tnz)
+ */
+.globl	.umul, __umul
+.umul:
+__umul:
+	or	%o0, %o1, %o4
+	mov	%o0, %y		! multiplier -> Y
+	andncc	%o4, 0xfff, %g0	! test bits 12..31 of *both* args
+	be	Lumul_shortway	! if zero, can do it the short way
+	 andcc	%g0, %g0, %o4	! zero the partial product and clear N and V
+
+	/*
+	 * Long multiply.  32 steps, followed by a final shift step.
+	 */
+	mulscc	%o4, %o1, %o4	! 1
+	mulscc	%o4, %o1, %o4	! 2
+	mulscc	%o4, %o1, %o4	! 3
+	mulscc	%o4, %o1, %o4	! 4
+	mulscc	%o4, %o1, %o4	! 5
+	mulscc	%o4, %o1, %o4	! 6
+	mulscc	%o4, %o1, %o4	! 7
+	mulscc	%o4, %o1, %o4	! 8
+	mulscc	%o4, %o1, %o4	! 9
+	mulscc	%o4, %o1, %o4	! 10
+	mulscc	%o4, %o1, %o4	! 11
+	mulscc	%o4, %o1, %o4	! 12
+	mulscc	%o4, %o1, %o4	! 13
+	mulscc	%o4, %o1, %o4	! 14
+	mulscc	%o4, %o1, %o4	! 15
+	mulscc	%o4, %o1, %o4	! 16
+	mulscc	%o4, %o1, %o4	! 17
+	mulscc	%o4, %o1, %o4	! 18
+	mulscc	%o4, %o1, %o4	! 19
+	mulscc	%o4, %o1, %o4	! 20
+	mulscc	%o4, %o1, %o4	! 21
+	mulscc	%o4, %o1, %o4	! 22
+	mulscc	%o4, %o1, %o4	! 23
+	mulscc	%o4, %o1, %o4	! 24
+	mulscc	%o4, %o1, %o4	! 25
+	mulscc	%o4, %o1, %o4	! 26
+	mulscc	%o4, %o1, %o4	! 27
+	mulscc	%o4, %o1, %o4	! 28
+	mulscc	%o4, %o1, %o4	! 29
+	mulscc	%o4, %o1, %o4	! 30
+	mulscc	%o4, %o1, %o4	! 31
+	mulscc	%o4, %o1, %o4	! 32
+	mulscc	%o4, %g0, %o4	! final shift
+
+
+	/*
+	 * Normally, with the shift-and-add approach, if both numbers are
+	 * positive you get the correct result.  WIth 32-bit two's-complement
+	 * numbers, -x is represented as
+	 *
+	 *		  x		    32
+	 *	( 2  -  ------ ) mod 2  *  2
+	 *		   32
+	 *		  2
+	 *
+	 * (the `mod 2' subtracts 1 from 1.bbbb).  To avoid lots of 2^32s,
+	 * we can treat this as if the radix point were just to the left
+	 * of the sign bit (multiply by 2^32), and get
+	 *
+	 *	-x  =  (2 - x) mod 2
+	 *
+	 * Then, ignoring the `mod 2's for convenience:
+	 *
+	 *   x *  y	= xy
+	 *  -x *  y	= 2y - xy
+	 *   x * -y	= 2x - xy
+	 *  -x * -y	= 4 - 2x - 2y + xy
+	 *
+	 * For signed multiplies, we subtract (x << 32) from the partial
+	 * product to fix this problem for negative multipliers (see mul.s).
+	 * Because of the way the shift into the partial product is calculated
+	 * (N xor V), this term is automatically removed for the multiplicand,
+	 * so we don't have to adjust.
+	 *
+	 * But for unsigned multiplies, the high order bit wasn't a sign bit,
+	 * and the correction is wrong.  So for unsigned multiplies where the
+	 * high order bit is one, we end up with xy - (y << 32).  To fix it
+	 * we add y << 32.
+	 */
+	tst	%o1
+	bl,a	1f		! if %o1 < 0 (high order bit = 1),
+	add	%o4, %o0, %o4	! %o4 += %o0 (add y to upper half)
+1:	rd	%y, %o0		! get lower half of product
+	retl
+	 addcc	%o4, %g0, %o1	! put upper half in place and set Z for %o1==0
+
+Lumul_shortway:
+	/*
+	 * Short multiply.  12 steps, followed by a final shift step.
+	 * The resulting bits are off by 12 and (32-12) = 20 bit positions,
+	 * but there is no problem with %o0 being negative (unlike above),
+	 * and overflow is impossible (the answer is at most 24 bits long).
+	 */
+	mulscc	%o4, %o1, %o4	! 1
+	mulscc	%o4, %o1, %o4	! 2
+	mulscc	%o4, %o1, %o4	! 3
+	mulscc	%o4, %o1, %o4	! 4
+	mulscc	%o4, %o1, %o4	! 5
+	mulscc	%o4, %o1, %o4	! 6
+	mulscc	%o4, %o1, %o4	! 7
+	mulscc	%o4, %o1, %o4	! 8
+	mulscc	%o4, %o1, %o4	! 9
+	mulscc	%o4, %o1, %o4	! 10
+	mulscc	%o4, %o1, %o4	! 11
+	mulscc	%o4, %o1, %o4	! 12
+	mulscc	%o4, %g0, %o4	! final shift
+
+	/*
+	 * %o4 has 20 of the bits that should be in the result; %y has
+	 * the bottom 12 (as %y's top 12).  That is:
+	 *
+	 *	  %o4		    %y
+	 * +----------------+----------------+
+	 * | -12- |   -20-  | -12- |   -20-  |
+	 * +------(---------+------)---------+
+	 *	   -----result-----
+	 *
+	 * The 12 bits of %o4 left of the `result' area are all zero;
+	 * in fact, all top 20 bits of %o4 are zero.
+	 */
+
+	rd	%y, %o5
+	sll	%o4, 12, %o0	! shift middle bits left 12
+	srl	%o5, 20, %o5	! shift low bits right 20
+	or	%o5, %o0, %o0
+	retl
+	 addcc	%g0, %g0, %o1	! %o1 = zero, and set Z
+
+/*
  * Here is a very good random number generator.  This implementation is
  * based on ``Two Fast Implementations of the "Minimal Standard" Random
  * Number Generator", David G. Carta, Communications of the ACM, Jan 1990,
diff --git a/sys/lib/libkern/arch/sparc/Makefile.inc b/sys/lib/libkern/arch/sparc/Makefile.inc
index 49da2f42b7c..3d14262e7a8 100644
--- a/sys/lib/libkern/arch/sparc/Makefile.inc
+++ b/sys/lib/libkern/arch/sparc/Makefile.inc
@@ -1,4 +1,4 @@
-#	$OpenBSD: Makefile.inc,v 1.5 1998/06/02 20:38:26 deraadt Exp $
+#	$OpenBSD: Makefile.inc,v 1.6 1999/12/07 00:49:10 deraadt Exp $
 #	$NetBSD: Makefile.inc,v 1.12 1996/04/23 23:05:22 christos Exp $
 
 SRCS+=  __main.c imax.c imin.c lmax.c lmin.c max.c min.c ulmax.c ulmin.c \
@@ -7,8 +7,8 @@ SRCS+=  __main.c imax.c imin.c lmax.c lmin.c max.c min.c ulmax.c ulmin.c \
 	strncpy.c htonl.S htons.S ntohl.S ntohs.S scanc.c skpc.c \
 	strncasecmp.c
 
-SRCS+=	umul.S mul.S rem.S sdiv.S udiv.S umul.S urem.S
-SRCS+=	mul.S saveregs.S umul.S
+SRCS+=	rem.S sdiv.S udiv.S urem.S
+SRCS+=	saveregs.S
 
 # `source' files built from m4 source
 # the name `div.o' is taken for the ANSI C `div' function, hence sdiv here
diff --git a/sys/lib/libkern/arch/sparc/mul.S b/sys/lib/libkern/arch/sparc/mul.S
deleted file mode 100644
index 028237f2c47..00000000000
--- a/sys/lib/libkern/arch/sparc/mul.S
+++ /dev/null
@@ -1,160 +0,0 @@
-/*	$OpenBSD: mul.S,v 1.2 1997/11/07 15:57:37 niklas Exp $	*/
-/*	$NetBSD: mul.S,v 1.2 1994/10/26 06:40:01 cgd Exp $	*/
-
-/*
- * Copyright (c) 1992, 1993
- *	The Regents of the University of California.  All rights reserved.
- *
- * This software was developed by the Computer Systems Engineering group
- * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
- * contributed to Berkeley.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. All advertising materials mentioning features or use of this software
- *    must display the following acknowledgement:
- *	This product includes software developed by the University of
- *	California, Berkeley and its contributors.
- * 4. Neither the name of the University nor the names of its contributors
- *    may be used to endorse or promote products derived from this software
- *    without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
- * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- *
- * Header: mul.s,v 1.5 92/06/25 13:24:03 torek Exp
- */
-
-#if defined(LIBC_SCCS) && !defined(lint)
-#ifdef notdef
-	.asciz "@(#)mul.s	8.1 (Berkeley) 6/4/93"
-#endif
-	.asciz "$OpenBSD: mul.S,v 1.2 1997/11/07 15:57:37 niklas Exp $"
-#endif /* LIBC_SCCS and not lint */
-
-/*
- * Signed multiply, from Appendix E of the Sparc Version 8
- * Architecture Manual.
- *
- * Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the upper 32 bits of
- * the 64-bit product).
- *
- * This code optimizes short (less than 13-bit) multiplies.
- */
-
-#include "DEFS.h"
-FUNC(.mul)
-	mov	%o0, %y		! multiplier -> Y
-	andncc	%o0, 0xfff, %g0	! test bits 12..31
-	be	Lmul_shortway	! if zero, can do it the short way
-	andcc	%g0, %g0, %o4	! zero the partial product and clear N and V
-
-	/*
-	 * Long multiply.  32 steps, followed by a final shift step.
-	 */
-	mulscc	%o4, %o1, %o4	! 1
-	mulscc	%o4, %o1, %o4	! 2
-	mulscc	%o4, %o1, %o4	! 3
-	mulscc	%o4, %o1, %o4	! 4
-	mulscc	%o4, %o1, %o4	! 5
-	mulscc	%o4, %o1, %o4	! 6
-	mulscc	%o4, %o1, %o4	! 7
-	mulscc	%o4, %o1, %o4	! 8
-	mulscc	%o4, %o1, %o4	! 9
-	mulscc	%o4, %o1, %o4	! 10
-	mulscc	%o4, %o1, %o4	! 11
-	mulscc	%o4, %o1, %o4	! 12
-	mulscc	%o4, %o1, %o4	! 13
-	mulscc	%o4, %o1, %o4	! 14
-	mulscc	%o4, %o1, %o4	! 15
-	mulscc	%o4, %o1, %o4	! 16
-	mulscc	%o4, %o1, %o4	! 17
-	mulscc	%o4, %o1, %o4	! 18
-	mulscc	%o4, %o1, %o4	! 19
-	mulscc	%o4, %o1, %o4	! 20
-	mulscc	%o4, %o1, %o4	! 21
-	mulscc	%o4, %o1, %o4	! 22
-	mulscc	%o4, %o1, %o4	! 23
-	mulscc	%o4, %o1, %o4	! 24
-	mulscc	%o4, %o1, %o4	! 25
-	mulscc	%o4, %o1, %o4	! 26
-	mulscc	%o4, %o1, %o4	! 27
-	mulscc	%o4, %o1, %o4	! 28
-	mulscc	%o4, %o1, %o4	! 29
-	mulscc	%o4, %o1, %o4	! 30
-	mulscc	%o4, %o1, %o4	! 31
-	mulscc	%o4, %o1, %o4	! 32
-	mulscc	%o4, %g0, %o4	! final shift
-
-	! If %o0 was negative, the result is
-	!	(%o0 * %o1) + (%o1 << 32))
-	! We fix that here.
-
-	tst	%o0
-	bge	1f
-	rd	%y, %o0
-
-	! %o0 was indeed negative; fix upper 32 bits of result by subtracting 
-	! %o1 (i.e., return %o4 - %o1 in %o1).
-	retl
-	sub	%o4, %o1, %o1
-
-1:
-	retl
-	mov	%o4, %o1
-
-Lmul_shortway:
-	/*
-	 * Short multiply.  12 steps, followed by a final shift step.
-	 * The resulting bits are off by 12 and (32-12) = 20 bit positions,
-	 * but there is no problem with %o0 being negative (unlike above).
-	 */
-	mulscc	%o4, %o1, %o4	! 1
-	mulscc	%o4, %o1, %o4	! 2
-	mulscc	%o4, %o1, %o4	! 3
-	mulscc	%o4, %o1, %o4	! 4
-	mulscc	%o4, %o1, %o4	! 5
-	mulscc	%o4, %o1, %o4	! 6
-	mulscc	%o4, %o1, %o4	! 7
-	mulscc	%o4, %o1, %o4	! 8
-	mulscc	%o4, %o1, %o4	! 9
-	mulscc	%o4, %o1, %o4	! 10
-	mulscc	%o4, %o1, %o4	! 11
-	mulscc	%o4, %o1, %o4	! 12
-	mulscc	%o4, %g0, %o4	! final shift
-
-	/*
-	 *  %o4 has 20 of the bits that should be in the low part of the
-	 * result; %y has the bottom 12 (as %y's top 12).  That is:
-	 *
-	 *	  %o4		    %y
-	 * +----------------+----------------+
-	 * | -12- |   -20-  | -12- |   -20-  |
-	 * +------(---------+------)---------+
-	 *  --hi-- ----low-part----
-	 *
-	 * The upper 12 bits of %o4 should be sign-extended to form the
-	 * high part of the product (i.e., highpart = %o4 >> 20).
-	 */
-
-	rd	%y, %o5
-	sll	%o4, 12, %o0	! shift middle bits left 12
-	srl	%o5, 20, %o5	! shift low bits right 20, zero fill at left
-	or	%o5, %o0, %o0	! construct low part of result
-	retl
-	sra	%o4, 20, %o1	! ... and extract high part of result
diff --git a/sys/lib/libkern/arch/sparc/umul.S b/sys/lib/libkern/arch/sparc/umul.S
deleted file mode 100644
index 6a5b7005e22..00000000000
--- a/sys/lib/libkern/arch/sparc/umul.S
+++ /dev/null
@@ -1,193 +0,0 @@
-/*	$OpenBSD: umul.S,v 1.2 1997/11/07 15:57:42 niklas Exp $	*/
-/*	$NetBSD: umul.S,v 1.2 1994/10/26 06:40:10 cgd Exp $	*/
-
-/*
- * Copyright (c) 1992, 1993
- *	The Regents of the University of California.  All rights reserved.
- *
- * This software was developed by the Computer Systems Engineering group
- * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
- * contributed to Berkeley.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. All advertising materials mentioning features or use of this software
- *    must display the following acknowledgement:
- *	This product includes software developed by the University of
- *	California, Berkeley and its contributors.
- * 4. Neither the name of the University nor the names of its contributors
- *    may be used to endorse or promote products derived from this software
- *    without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
- * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- *
- * Header: umul.s,v 1.4 92/06/25 13:24:05 torek Exp
- */
-
-#if defined(LIBC_SCCS) && !defined(lint)
-#ifdef notdef
-	.asciz "@(#)umul.s	8.1 (Berkeley) 6/4/93"
-#endif
-	.asciz "$OpenBSD: umul.S,v 1.2 1997/11/07 15:57:42 niklas Exp $"
-#endif /* LIBC_SCCS and not lint */
-
-/*
- * Unsigned multiply.  Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the
- * upper 32 bits of the 64-bit product).
- *
- * This code optimizes short (less than 13-bit) multiplies.  Short
- * multiplies require 25 instruction cycles, and long ones require
- * 45 instruction cycles.
- *
- * On return, overflow has occurred (%o1 is not zero) if and only if
- * the Z condition code is clear, allowing, e.g., the following:
- *
- *	call	.umul
- *	nop
- *	bnz	overflow	(or tnz)
- */
-
-#include "DEFS.h"
-FUNC(.umul)
-	or	%o0, %o1, %o4
-	mov	%o0, %y		! multiplier -> Y
-	andncc	%o4, 0xfff, %g0	! test bits 12..31 of *both* args
-	be	Lmul_shortway	! if zero, can do it the short way
-	andcc	%g0, %g0, %o4	! zero the partial product and clear N and V
-
-	/*
-	 * Long multiply.  32 steps, followed by a final shift step.
-	 */
-	mulscc	%o4, %o1, %o4	! 1
-	mulscc	%o4, %o1, %o4	! 2
-	mulscc	%o4, %o1, %o4	! 3
-	mulscc	%o4, %o1, %o4	! 4
-	mulscc	%o4, %o1, %o4	! 5
-	mulscc	%o4, %o1, %o4	! 6
-	mulscc	%o4, %o1, %o4	! 7
-	mulscc	%o4, %o1, %o4	! 8
-	mulscc	%o4, %o1, %o4	! 9
-	mulscc	%o4, %o1, %o4	! 10
-	mulscc	%o4, %o1, %o4	! 11
-	mulscc	%o4, %o1, %o4	! 12
-	mulscc	%o4, %o1, %o4	! 13
-	mulscc	%o4, %o1, %o4	! 14
-	mulscc	%o4, %o1, %o4	! 15
-	mulscc	%o4, %o1, %o4	! 16
-	mulscc	%o4, %o1, %o4	! 17
-	mulscc	%o4, %o1, %o4	! 18
-	mulscc	%o4, %o1, %o4	! 19
-	mulscc	%o4, %o1, %o4	! 20
-	mulscc	%o4, %o1, %o4	! 21
-	mulscc	%o4, %o1, %o4	! 22
-	mulscc	%o4, %o1, %o4	! 23
-	mulscc	%o4, %o1, %o4	! 24
-	mulscc	%o4, %o1, %o4	! 25
-	mulscc	%o4, %o1, %o4	! 26
-	mulscc	%o4, %o1, %o4	! 27
-	mulscc	%o4, %o1, %o4	! 28
-	mulscc	%o4, %o1, %o4	! 29
-	mulscc	%o4, %o1, %o4	! 30
-	mulscc	%o4, %o1, %o4	! 31
-	mulscc	%o4, %o1, %o4	! 32
-	mulscc	%o4, %g0, %o4	! final shift
-
-
-	/*
-	 * Normally, with the shift-and-add approach, if both numbers are
-	 * positive you get the correct result.  WIth 32-bit two's-complement
-	 * numbers, -x is represented as
-	 *
-	 *		  x		    32
-	 *	( 2  -  ------ ) mod 2  *  2
-	 *		   32
-	 *		  2
-	 *
-	 * (the `mod 2' subtracts 1 from 1.bbbb).  To avoid lots of 2^32s,
-	 * we can treat this as if the radix point were just to the left
-	 * of the sign bit (multiply by 2^32), and get
-	 *
-	 *	-x  =  (2 - x) mod 2
-	 *
-	 * Then, ignoring the `mod 2's for convenience:
-	 *
-	 *   x *  y	= xy
-	 *  -x *  y	= 2y - xy
-	 *   x * -y	= 2x - xy
-	 *  -x * -y	= 4 - 2x - 2y + xy
-	 *
-	 * For signed multiplies, we subtract (x << 32) from the partial
-	 * product to fix this problem for negative multipliers (see mul.s).
-	 * Because of the way the shift into the partial product is calculated
-	 * (N xor V), this term is automatically removed for the multiplicand,
-	 * so we don't have to adjust.
-	 *
-	 * But for unsigned multiplies, the high order bit wasn't a sign bit,
-	 * and the correction is wrong.  So for unsigned multiplies where the
-	 * high order bit is one, we end up with xy - (y << 32).  To fix it
-	 * we add y << 32.
-	 */
-	tst	%o1
-	bl,a	1f		! if %o1 < 0 (high order bit = 1),
-	add	%o4, %o0, %o4	! %o4 += %o0 (add y to upper half)
-1:	rd	%y, %o0		! get lower half of product
-	retl
-	addcc	%o4, %g0, %o1	! put upper half in place and set Z for %o1==0
-
-Lmul_shortway:
-	/*
-	 * Short multiply.  12 steps, followed by a final shift step.
-	 * The resulting bits are off by 12 and (32-12) = 20 bit positions,
-	 * but there is no problem with %o0 being negative (unlike above),
-	 * and overflow is impossible (the answer is at most 24 bits long).
-	 */
-	mulscc	%o4, %o1, %o4	! 1
-	mulscc	%o4, %o1, %o4	! 2
-	mulscc	%o4, %o1, %o4	! 3
-	mulscc	%o4, %o1, %o4	! 4
-	mulscc	%o4, %o1, %o4	! 5
-	mulscc	%o4, %o1, %o4	! 6
-	mulscc	%o4, %o1, %o4	! 7
-	mulscc	%o4, %o1, %o4	! 8
-	mulscc	%o4, %o1, %o4	! 9
-	mulscc	%o4, %o1, %o4	! 10
-	mulscc	%o4, %o1, %o4	! 11
-	mulscc	%o4, %o1, %o4	! 12
-	mulscc	%o4, %g0, %o4	! final shift
-
-	/*
-	 * %o4 has 20 of the bits that should be in the result; %y has
-	 * the bottom 12 (as %y's top 12).  That is:
-	 *
-	 *	  %o4		    %y
-	 * +----------------+----------------+
-	 * | -12- |   -20-  | -12- |   -20-  |
-	 * +------(---------+------)---------+
-	 *	   -----result-----
-	 *
-	 * The 12 bits of %o4 left of the `result' area are all zero;
-	 * in fact, all top 20 bits of %o4 are zero.
-	 */
-
-	rd	%y, %o5
-	sll	%o4, 12, %o0	! shift middle bits left 12
-	srl	%o5, 20, %o5	! shift low bits right 20
-	or	%o5, %o0, %o0
-	retl
-	addcc	%g0, %g0, %o1	! %o1 = zero, and set Z