xref: /freebsd/sys/powerpc/fpu/fpu_mul.c (revision 2fb4f839f3fc72ce2bab12f9ba4760f97f73e97f)
1 /*	$NetBSD: fpu_mul.c,v 1.4 2005/12/11 12:18:42 christos Exp $ */
2 
3 /*
4  * SPDX-License-Identifier: BSD-3-Clause
5  *
6  * Copyright (c) 1992, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This software was developed by the Computer Systems Engineering group
10  * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
11  * contributed to Berkeley.
12  *
13  * All advertising materials mentioning features or use of this software
14  * must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Lawrence Berkeley Laboratory.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  * 3. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)fpu_mul.c	8.1 (Berkeley) 6/11/93
43  */
44 
45 /*
46  * Perform an FPU multiply (return x * y).
47  */
48 
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
51 
52 #include <sys/types.h>
53 #include <sys/systm.h>
54 
55 #include <machine/fpu.h>
56 
57 #include <powerpc/fpu/fpu_arith.h>
58 #include <powerpc/fpu/fpu_emu.h>
59 
60 /*
61  * The multiplication algorithm for normal numbers is as follows:
62  *
63  * The fraction of the product is built in the usual stepwise fashion.
64  * Each step consists of shifting the accumulator right one bit
65  * (maintaining any guard bits) and, if the next bit in y is set,
66  * adding the multiplicand (x) to the accumulator.  Then, in any case,
67  * we advance one bit leftward in y.  Algorithmically:
68  *
69  *	A = 0;
70  *	for (bit = 0; bit < FP_NMANT; bit++) {
71  *		sticky |= A & 1, A >>= 1;
72  *		if (Y & (1 << bit))
73  *			A += X;
74  *	}
75  *
76  * (X and Y here represent the mantissas of x and y respectively.)
77  * The resultant accumulator (A) is the product's mantissa.  It may
78  * be as large as 11.11111... in binary and hence may need to be
79  * shifted right, but at most one bit.
80  *
81  * Since we do not have efficient multiword arithmetic, we code the
82  * accumulator as four separate words, just like any other mantissa.
83  * We use local variables in the hope that this is faster than memory.
84  * We keep x->fp_mant in locals for the same reason.
85  *
86  * In the algorithm above, the bits in y are inspected one at a time.
87  * We will pick them up 32 at a time and then deal with those 32, one
88  * at a time.  Note, however, that we know several things about y:
89  *
90  *    - the guard and round bits at the bottom are sure to be zero;
91  *
92  *    - often many low bits are zero (y is often from a single or double
93  *	precision source);
94  *
95  *    - bit FP_NMANT-1 is set, and FP_1*2 fits in a word.
96  *
97  * We can also test for 32-zero-bits swiftly.  In this case, the center
98  * part of the loop---setting sticky, shifting A, and not adding---will
99  * run 32 times without adding X to A.  We can do a 32-bit shift faster
100  * by simply moving words.  Since zeros are common, we optimize this case.
101  * Furthermore, since A is initially zero, we can omit the shift as well
102  * until we reach a nonzero word.
103  */
104 struct fpn *
105 fpu_mul(struct fpemu *fe)
106 {
107 	struct fpn *x = &fe->fe_f1, *y = &fe->fe_f2;
108 	u_int a3, a2, a1, a0, x3, x2, x1, x0, bit, m;
109 	int sticky;
110 	FPU_DECL_CARRY;
111 
112 	/*
113 	 * Put the `heavier' operand on the right (see fpu_emu.h).
114 	 * Then we will have one of the following cases, taken in the
115 	 * following order:
116 	 *
117 	 *  - y = NaN.  Implied: if only one is a signalling NaN, y is.
118 	 *	The result is y.
119 	 *  - y = Inf.  Implied: x != NaN (is 0, number, or Inf: the NaN
120 	 *    case was taken care of earlier).
121 	 *	If x = 0, the result is NaN.  Otherwise the result
122 	 *	is y, with its sign reversed if x is negative.
123 	 *  - x = 0.  Implied: y is 0 or number.
124 	 *	The result is 0 (with XORed sign as usual).
125 	 *  - other.  Implied: both x and y are numbers.
126 	 *	The result is x * y (XOR sign, multiply bits, add exponents).
127 	 */
128 	DPRINTF(FPE_REG, ("fpu_mul:\n"));
129 	DUMPFPN(FPE_REG, x);
130 	DUMPFPN(FPE_REG, y);
131 	DPRINTF(FPE_REG, ("=>\n"));
132 
133 	ORDER(x, y);
134 	if (ISNAN(y)) {
135 		y->fp_sign ^= x->fp_sign;
136 		fe->fe_cx |= FPSCR_VXSNAN;
137 		DUMPFPN(FPE_REG, y);
138 		return (y);
139 	}
140 	if (ISINF(y)) {
141 		if (ISZERO(x)) {
142 			fe->fe_cx |= FPSCR_VXIMZ;
143 			return (fpu_newnan(fe));
144 		}
145 		y->fp_sign ^= x->fp_sign;
146 			DUMPFPN(FPE_REG, y);
147 		return (y);
148 	}
149 	if (ISZERO(x)) {
150 		x->fp_sign ^= y->fp_sign;
151 		DUMPFPN(FPE_REG, x);
152 		return (x);
153 	}
154 
155 	/*
156 	 * Setup.  In the code below, the mask `m' will hold the current
157 	 * mantissa byte from y.  The variable `bit' denotes the bit
158 	 * within m.  We also define some macros to deal with everything.
159 	 */
160 	x3 = x->fp_mant[3];
161 	x2 = x->fp_mant[2];
162 	x1 = x->fp_mant[1];
163 	x0 = x->fp_mant[0];
164 	sticky = a3 = a2 = a1 = a0 = 0;
165 
166 #define	ADD	/* A += X */ \
167 	FPU_ADDS(a3, a3, x3); \
168 	FPU_ADDCS(a2, a2, x2); \
169 	FPU_ADDCS(a1, a1, x1); \
170 	FPU_ADDC(a0, a0, x0)
171 
172 #define	SHR1	/* A >>= 1, with sticky */ \
173 	sticky |= a3 & 1, a3 = (a3 >> 1) | (a2 << 31), \
174 	a2 = (a2 >> 1) | (a1 << 31), a1 = (a1 >> 1) | (a0 << 31), a0 >>= 1
175 
176 #define	SHR32	/* A >>= 32, with sticky */ \
177 	sticky |= a3, a3 = a2, a2 = a1, a1 = a0, a0 = 0
178 
179 #define	STEP	/* each 1-bit step of the multiplication */ \
180 	SHR1; if (bit & m) { ADD; }; bit <<= 1
181 
182 	/*
183 	 * We are ready to begin.  The multiply loop runs once for each
184 	 * of the four 32-bit words.  Some words, however, are special.
185 	 * As noted above, the low order bits of Y are often zero.  Even
186 	 * if not, the first loop can certainly skip the guard bits.
187 	 * The last word of y has its highest 1-bit in position FP_NMANT-1,
188 	 * so we stop the loop when we move past that bit.
189 	 */
190 	if ((m = y->fp_mant[3]) == 0) {
191 		/* SHR32; */			/* unneeded since A==0 */
192 	} else {
193 		bit = 1 << FP_NG;
194 		do {
195 			STEP;
196 		} while (bit != 0);
197 	}
198 	if ((m = y->fp_mant[2]) == 0) {
199 		SHR32;
200 	} else {
201 		bit = 1;
202 		do {
203 			STEP;
204 		} while (bit != 0);
205 	}
206 	if ((m = y->fp_mant[1]) == 0) {
207 		SHR32;
208 	} else {
209 		bit = 1;
210 		do {
211 			STEP;
212 		} while (bit != 0);
213 	}
214 	m = y->fp_mant[0];		/* definitely != 0 */
215 	bit = 1;
216 	do {
217 		STEP;
218 	} while (bit <= m);
219 
220 	/*
221 	 * Done with mantissa calculation.  Get exponent and handle
222 	 * 11.111...1 case, then put result in place.  We reuse x since
223 	 * it already has the right class (FP_NUM).
224 	 */
225 	m = x->fp_exp + y->fp_exp;
226 	if (a0 >= FP_2) {
227 		SHR1;
228 		m++;
229 	}
230 	x->fp_sign ^= y->fp_sign;
231 	x->fp_exp = m;
232 	x->fp_sticky = sticky;
233 	x->fp_mant[3] = a3;
234 	x->fp_mant[2] = a2;
235 	x->fp_mant[1] = a1;
236 	x->fp_mant[0] = a0;
237 
238 	DUMPFPN(FPE_REG, x);
239 	return (x);
240 }
241