1 /* $NetBSD: umul.S,v 1.3 1997/07/16 14:37:44 christos Exp $ */
4 * Copyright (c) 1992, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This software was developed by the Computer Systems Engineering group
8 * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
9 * contributed to Berkeley.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: Header: umul.s,v 1.4 92/06/25 13:24:05 torek Exp
42 #include <machine/asm.h>
43 #if defined(LIBC_SCCS) && !defined(lint)
45 .asciz "@(#)umul.s 8.1 (Berkeley) 6/4/93"
47 RCSID("$NetBSD: umul.S,v 1.3 1997/07/16 14:37:44 christos Exp $")
49 #endif /* LIBC_SCCS and not lint */
52 * Unsigned multiply. Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the
53 * upper 32 bits of the 64-bit product).
55 * This code optimizes short (less than 13-bit) multiplies. Short
56 * multiplies require 25 instruction cycles, and long ones require
57 * 45 instruction cycles.
59 * On return, overflow has occurred (%o1 is not zero) if and only if
60 * the Z condition code is clear, allowing, e.g., the following:
64 * bnz overflow (or tnz)
69 mov %o0, %y ! multiplier -> Y
70 andncc %o4, 0xfff, %g0 ! test bits 12..31 of *both* args
71 be Lmul_shortway ! if zero, can do it the short way
72 andcc %g0, %g0, %o4 ! zero the partial product and clear N and V
75 * Long multiply. 32 steps, followed by a final shift step.
77 mulscc %o4, %o1, %o4 ! 1
78 mulscc %o4, %o1, %o4 ! 2
79 mulscc %o4, %o1, %o4 ! 3
80 mulscc %o4, %o1, %o4 ! 4
81 mulscc %o4, %o1, %o4 ! 5
82 mulscc %o4, %o1, %o4 ! 6
83 mulscc %o4, %o1, %o4 ! 7
84 mulscc %o4, %o1, %o4 ! 8
85 mulscc %o4, %o1, %o4 ! 9
86 mulscc %o4, %o1, %o4 ! 10
87 mulscc %o4, %o1, %o4 ! 11
88 mulscc %o4, %o1, %o4 ! 12
89 mulscc %o4, %o1, %o4 ! 13
90 mulscc %o4, %o1, %o4 ! 14
91 mulscc %o4, %o1, %o4 ! 15
92 mulscc %o4, %o1, %o4 ! 16
93 mulscc %o4, %o1, %o4 ! 17
94 mulscc %o4, %o1, %o4 ! 18
95 mulscc %o4, %o1, %o4 ! 19
96 mulscc %o4, %o1, %o4 ! 20
97 mulscc %o4, %o1, %o4 ! 21
98 mulscc %o4, %o1, %o4 ! 22
99 mulscc %o4, %o1, %o4 ! 23
100 mulscc %o4, %o1, %o4 ! 24
101 mulscc %o4, %o1, %o4 ! 25
102 mulscc %o4, %o1, %o4 ! 26
103 mulscc %o4, %o1, %o4 ! 27
104 mulscc %o4, %o1, %o4 ! 28
105 mulscc %o4, %o1, %o4 ! 29
106 mulscc %o4, %o1, %o4 ! 30
107 mulscc %o4, %o1, %o4 ! 31
108 mulscc %o4, %o1, %o4 ! 32
109 mulscc %o4, %g0, %o4 ! final shift
113 * Normally, with the shift-and-add approach, if both numbers are
114 * positive you get the correct result. WIth 32-bit two's-complement
115 * numbers, -x is represented as
118 * ( 2 - ------ ) mod 2 * 2
122 * (the `mod 2' subtracts 1 from 1.bbbb). To avoid lots of 2^32s,
123 * we can treat this as if the radix point were just to the left
124 * of the sign bit (multiply by 2^32), and get
128 * Then, ignoring the `mod 2's for convenience:
133 * -x * -y = 4 - 2x - 2y + xy
135 * For signed multiplies, we subtract (x << 32) from the partial
136 * product to fix this problem for negative multipliers (see mul.s).
137 * Because of the way the shift into the partial product is calculated
138 * (N xor V), this term is automatically removed for the multiplicand,
139 * so we don't have to adjust.
141 * But for unsigned multiplies, the high order bit wasn't a sign bit,
142 * and the correction is wrong. So for unsigned multiplies where the
143 * high order bit is one, we end up with xy - (y << 32). To fix it
147 bl,a 1f ! if %o1 < 0 (high order bit = 1),
148 add %o4, %o0, %o4 ! %o4 += %o0 (add y to upper half)
149 1: rd %y, %o0 ! get lower half of product
151 addcc %o4, %g0, %o1 ! put upper half in place and set Z for %o1==0
155 * Short multiply. 12 steps, followed by a final shift step.
156 * The resulting bits are off by 12 and (32-12) = 20 bit positions,
157 * but there is no problem with %o0 being negative (unlike above),
158 * and overflow is impossible (the answer is at most 24 bits long).
160 mulscc %o4, %o1, %o4 ! 1
161 mulscc %o4, %o1, %o4 ! 2
162 mulscc %o4, %o1, %o4 ! 3
163 mulscc %o4, %o1, %o4 ! 4
164 mulscc %o4, %o1, %o4 ! 5
165 mulscc %o4, %o1, %o4 ! 6
166 mulscc %o4, %o1, %o4 ! 7
167 mulscc %o4, %o1, %o4 ! 8
168 mulscc %o4, %o1, %o4 ! 9
169 mulscc %o4, %o1, %o4 ! 10
170 mulscc %o4, %o1, %o4 ! 11
171 mulscc %o4, %o1, %o4 ! 12
172 mulscc %o4, %g0, %o4 ! final shift
175 * %o4 has 20 of the bits that should be in the result; %y has
176 * the bottom 12 (as %y's top 12). That is:
179 * +----------------+----------------+
180 * | -12- | -20- | -12- | -20- |
181 * +------(---------+------)---------+
184 * The 12 bits of %o4 left of the `result' area are all zero;
185 * in fact, all top 20 bits of %o4 are zero.
189 sll %o4, 12, %o0 ! shift middle bits left 12
190 srl %o5, 20, %o5 ! shift low bits right 20
193 addcc %g0, %g0, %o1 ! %o1 = zero, and set Z