xref: /openbmc/linux/arch/mips/math-emu/dp_mul.c (revision 4f6cce39)
1 /* IEEE754 floating point arithmetic
2  * double precision: common utilities
3  */
4 /*
5  * MIPS floating point support
6  * Copyright (C) 1994-2000 Algorithmics Ltd.
7  *
8  *  This program is free software; you can distribute it and/or modify it
9  *  under the terms of the GNU General Public License (Version 2) as
10  *  published by the Free Software Foundation.
11  *
12  *  This program is distributed in the hope it will be useful, but WITHOUT
13  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
15  *  for more details.
16  *
17  *  You should have received a copy of the GNU General Public License along
18  *  with this program; if not, write to the Free Software Foundation, Inc.,
19  *  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA.
20  */
21 
22 #include "ieee754dp.h"
23 
24 union ieee754dp ieee754dp_mul(union ieee754dp x, union ieee754dp y)
25 {
26 	int re;
27 	int rs;
28 	u64 rm;
29 	unsigned lxm;
30 	unsigned hxm;
31 	unsigned lym;
32 	unsigned hym;
33 	u64 lrm;
34 	u64 hrm;
35 	u64 t;
36 	u64 at;
37 
38 	COMPXDP;
39 	COMPYDP;
40 
41 	EXPLODEXDP;
42 	EXPLODEYDP;
43 
44 	ieee754_clearcx();
45 
46 	FLUSHXDP;
47 	FLUSHYDP;
48 
49 	switch (CLPAIR(xc, yc)) {
50 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN):
51 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN):
52 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN):
53 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN):
54 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN):
55 		return ieee754dp_nanxcpt(y);
56 
57 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN):
58 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN):
59 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO):
60 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM):
61 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM):
62 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF):
63 		return ieee754dp_nanxcpt(x);
64 
65 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN):
66 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN):
67 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN):
68 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN):
69 		return y;
70 
71 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN):
72 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO):
73 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM):
74 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM):
75 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF):
76 		return x;
77 
78 
79 	/*
80 	 * Infinity handling
81 	 */
82 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
83 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
84 		ieee754_setcx(IEEE754_INVALID_OPERATION);
85 		return ieee754dp_indef();
86 
87 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
88 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
89 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
90 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
91 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
92 		return ieee754dp_inf(xs ^ ys);
93 
94 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
95 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
96 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
97 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
98 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
99 		return ieee754dp_zero(xs ^ ys);
100 
101 
102 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
103 		DPDNORMX;
104 
105 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
106 		DPDNORMY;
107 		break;
108 
109 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
110 		DPDNORMX;
111 		break;
112 
113 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
114 		break;
115 	}
116 	/* rm = xm * ym, re = xe+ye basically */
117 	assert(xm & DP_HIDDEN_BIT);
118 	assert(ym & DP_HIDDEN_BIT);
119 
120 	re = xe + ye;
121 	rs = xs ^ ys;
122 
123 	/* shunt to top of word */
124 	xm <<= 64 - (DP_FBITS + 1);
125 	ym <<= 64 - (DP_FBITS + 1);
126 
127 	/*
128 	 * Multiply 64 bits xm, ym to give high 64 bits rm with stickness.
129 	 */
130 
131 	/* 32 * 32 => 64 */
132 #define DPXMULT(x, y)	((u64)(x) * (u64)y)
133 
134 	lxm = xm;
135 	hxm = xm >> 32;
136 	lym = ym;
137 	hym = ym >> 32;
138 
139 	lrm = DPXMULT(lxm, lym);
140 	hrm = DPXMULT(hxm, hym);
141 
142 	t = DPXMULT(lxm, hym);
143 
144 	at = lrm + (t << 32);
145 	hrm += at < lrm;
146 	lrm = at;
147 
148 	hrm = hrm + (t >> 32);
149 
150 	t = DPXMULT(hxm, lym);
151 
152 	at = lrm + (t << 32);
153 	hrm += at < lrm;
154 	lrm = at;
155 
156 	hrm = hrm + (t >> 32);
157 
158 	rm = hrm | (lrm != 0);
159 
160 	/*
161 	 * Sticky shift down to normal rounding precision.
162 	 */
163 	if ((s64) rm < 0) {
164 		rm = (rm >> (64 - (DP_FBITS + 1 + 3))) |
165 		     ((rm << (DP_FBITS + 1 + 3)) != 0);
166 		re++;
167 	} else {
168 		rm = (rm >> (64 - (DP_FBITS + 1 + 3 + 1))) |
169 		     ((rm << (DP_FBITS + 1 + 3 + 1)) != 0);
170 	}
171 	assert(rm & (DP_HIDDEN_BIT << 3));
172 
173 	return ieee754dp_format(rs, re, rm);
174 }
175