xref: /openbsd/sys/arch/sparc64/fpu/fpu_mul.c (revision 3d8817e4)
1 /*	$OpenBSD: fpu_mul.c,v 1.2 2003/06/02 23:27:55 millert Exp $	*/
2 /*	$NetBSD: fpu_mul.c,v 1.2 1994/11/20 20:52:44 deraadt Exp $ */
3 
4 /*
5  * Copyright (c) 1992, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This software was developed by the Computer Systems Engineering group
9  * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
10  * contributed to Berkeley.
11  *
12  * All advertising materials mentioning features or use of this software
13  * must display the following acknowledgement:
14  *	This product includes software developed by the University of
15  *	California, Lawrence Berkeley Laboratory.
16  *
17  * Redistribution and use in source and binary forms, with or without
18  * modification, are permitted provided that the following conditions
19  * are met:
20  * 1. Redistributions of source code must retain the above copyright
21  *    notice, this list of conditions and the following disclaimer.
22  * 2. Redistributions in binary form must reproduce the above copyright
23  *    notice, this list of conditions and the following disclaimer in the
24  *    documentation and/or other materials provided with the distribution.
25  * 3. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)fpu_mul.c	8.1 (Berkeley) 6/11/93
42  */
43 
44 /*
45  * Perform an FPU multiply (return x * y).
46  */
47 
48 #include <sys/types.h>
49 
50 #include <machine/reg.h>
51 
52 #include <sparc64/fpu/fpu_arith.h>
53 #include <sparc64/fpu/fpu_emu.h>
54 
55 /*
56  * The multiplication algorithm for normal numbers is as follows:
57  *
58  * The fraction of the product is built in the usual stepwise fashion.
59  * Each step consists of shifting the accumulator right one bit
60  * (maintaining any guard bits) and, if the next bit in y is set,
61  * adding the multiplicand (x) to the accumulator.  Then, in any case,
62  * we advance one bit leftward in y.  Algorithmically:
63  *
64  *	A = 0;
65  *	for (bit = 0; bit < FP_NMANT; bit++) {
66  *		sticky |= A & 1, A >>= 1;
67  *		if (Y & (1 << bit))
68  *			A += X;
69  *	}
70  *
71  * (X and Y here represent the mantissas of x and y respectively.)
72  * The resultant accumulator (A) is the product's mantissa.  It may
73  * be as large as 11.11111... in binary and hence may need to be
74  * shifted right, but at most one bit.
75  *
76  * Since we do not have efficient multiword arithmetic, we code the
77  * accumulator as four separate words, just like any other mantissa.
78  * We use local `register' variables in the hope that this is faster
79  * than memory.  We keep x->fp_mant in locals for the same reason.
80  *
81  * In the algorithm above, the bits in y are inspected one at a time.
82  * We will pick them up 32 at a time and then deal with those 32, one
83  * at a time.  Note, however, that we know several things about y:
84  *
85  *    - the guard and round bits at the bottom are sure to be zero;
86  *
87  *    - often many low bits are zero (y is often from a single or double
88  *	precision source);
89  *
90  *    - bit FP_NMANT-1 is set, and FP_1*2 fits in a word.
91  *
92  * We can also test for 32-zero-bits swiftly.  In this case, the center
93  * part of the loop---setting sticky, shifting A, and not adding---will
94  * run 32 times without adding X to A.  We can do a 32-bit shift faster
95  * by simply moving words.  Since zeros are common, we optimize this case.
96  * Furthermore, since A is initially zero, we can omit the shift as well
97  * until we reach a nonzero word.
98  */
99 struct fpn *
100 fpu_mul(fe)
101 	register struct fpemu *fe;
102 {
103 	register struct fpn *x = &fe->fe_f1, *y = &fe->fe_f2;
104 	register u_int a3, a2, a1, a0, x3, x2, x1, x0, bit, m;
105 	register int sticky;
106 	FPU_DECL_CARRY
107 
108 	/*
109 	 * Put the `heavier' operand on the right (see fpu_emu.h).
110 	 * Then we will have one of the following cases, taken in the
111 	 * following order:
112 	 *
113 	 *  - y = NaN.  Implied: if only one is a signalling NaN, y is.
114 	 *	The result is y.
115 	 *  - y = Inf.  Implied: x != NaN (is 0, number, or Inf: the NaN
116 	 *    case was taken care of earlier).
117 	 *	If x = 0, the result is NaN.  Otherwise the result
118 	 *	is y, with its sign reversed if x is negative.
119 	 *  - x = 0.  Implied: y is 0 or number.
120 	 *	The result is 0 (with XORed sign as usual).
121 	 *  - other.  Implied: both x and y are numbers.
122 	 *	The result is x * y (XOR sign, multiply bits, add exponents).
123 	 */
124 	ORDER(x, y);
125 	if (ISNAN(y)) {
126 		y->fp_sign ^= x->fp_sign;
127 		return (y);
128 	}
129 	if (ISINF(y)) {
130 		if (ISZERO(x))
131 			return (fpu_newnan(fe));
132 		y->fp_sign ^= x->fp_sign;
133 		return (y);
134 	}
135 	if (ISZERO(x)) {
136 		x->fp_sign ^= y->fp_sign;
137 		return (x);
138 	}
139 
140 	/*
141 	 * Setup.  In the code below, the mask `m' will hold the current
142 	 * mantissa byte from y.  The variable `bit' denotes the bit
143 	 * within m.  We also define some macros to deal with everything.
144 	 */
145 	x3 = x->fp_mant[3];
146 	x2 = x->fp_mant[2];
147 	x1 = x->fp_mant[1];
148 	x0 = x->fp_mant[0];
149 	sticky = a3 = a2 = a1 = a0 = 0;
150 
151 #define	ADD	/* A += X */ \
152 	FPU_ADDS(a3, a3, x3); \
153 	FPU_ADDCS(a2, a2, x2); \
154 	FPU_ADDCS(a1, a1, x1); \
155 	FPU_ADDC(a0, a0, x0)
156 
157 #define	SHR1	/* A >>= 1, with sticky */ \
158 	sticky |= a3 & 1, a3 = (a3 >> 1) | (a2 << 31), \
159 	a2 = (a2 >> 1) | (a1 << 31), a1 = (a1 >> 1) | (a0 << 31), a0 >>= 1
160 
161 #define	SHR32	/* A >>= 32, with sticky */ \
162 	sticky |= a3, a3 = a2, a2 = a1, a1 = a0, a0 = 0
163 
164 #define	STEP	/* each 1-bit step of the multiplication */ \
165 	SHR1; if (bit & m) { ADD; }; bit <<= 1
166 
167 	/*
168 	 * We are ready to begin.  The multiply loop runs once for each
169 	 * of the four 32-bit words.  Some words, however, are special.
170 	 * As noted above, the low order bits of Y are often zero.  Even
171 	 * if not, the first loop can certainly skip the guard bits.
172 	 * The last word of y has its highest 1-bit in position FP_NMANT-1,
173 	 * so we stop the loop when we move past that bit.
174 	 */
175 	if ((m = y->fp_mant[3]) == 0) {
176 		/* SHR32; */			/* unneeded since A==0 */
177 	} else {
178 		bit = 1 << FP_NG;
179 		do {
180 			STEP;
181 		} while (bit != 0);
182 	}
183 	if ((m = y->fp_mant[2]) == 0) {
184 		SHR32;
185 	} else {
186 		bit = 1;
187 		do {
188 			STEP;
189 		} while (bit != 0);
190 	}
191 	if ((m = y->fp_mant[1]) == 0) {
192 		SHR32;
193 	} else {
194 		bit = 1;
195 		do {
196 			STEP;
197 		} while (bit != 0);
198 	}
199 	m = y->fp_mant[0];		/* definitely != 0 */
200 	bit = 1;
201 	do {
202 		STEP;
203 	} while (bit <= m);
204 
205 	/*
206 	 * Done with mantissa calculation.  Get exponent and handle
207 	 * 11.111...1 case, then put result in place.  We reuse x since
208 	 * it already has the right class (FP_NUM).
209 	 */
210 	m = x->fp_exp + y->fp_exp;
211 	if (a0 >= FP_2) {
212 		SHR1;
213 		m++;
214 	}
215 	x->fp_sign ^= y->fp_sign;
216 	x->fp_exp = m;
217 	x->fp_sticky = sticky;
218 	x->fp_mant[3] = a3;
219 	x->fp_mant[2] = a2;
220 	x->fp_mant[1] = a1;
221 	x->fp_mant[0] = a0;
222 	return (x);
223 }
224