xref: /linux/arch/mips/math-emu/dp_mul.c (revision 50a23e6eec6f20d55a3a920e47adb455bff6046e)
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  * ########################################################################
9  *
10  *  This program is free software; you can distribute it and/or modify it
11  *  under the terms of the GNU General Public License (Version 2) as
12  *  published by the Free Software Foundation.
13  *
14  *  This program is distributed in the hope it will be useful, but WITHOUT
15  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
17  *  for more details.
18  *
19  *  You should have received a copy of the GNU General Public License along
20  *  with this program; if not, write to the Free Software Foundation, Inc.,
21  *  59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * ########################################################################
24  */
25 
26 
27 #include "ieee754dp.h"
28 
29 ieee754dp ieee754dp_mul(ieee754dp x, ieee754dp y)
30 {
31 	COMPXDP;
32 	COMPYDP;
33 
34 	EXPLODEXDP;
35 	EXPLODEYDP;
36 
37 	CLEARCX;
38 
39 	FLUSHXDP;
40 	FLUSHYDP;
41 
42 	switch (CLPAIR(xc, yc)) {
43 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN):
44 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN):
45 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN):
46 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN):
47 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN):
48 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN):
49 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN):
50 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO):
51 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM):
52 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM):
53 	case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF):
54 		SETCX(IEEE754_INVALID_OPERATION);
55 		return ieee754dp_nanxcpt(ieee754dp_indef(), "mul", x, y);
56 
57 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN):
58 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN):
59 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN):
60 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN):
61 		return y;
62 
63 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN):
64 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO):
65 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM):
66 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM):
67 	case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF):
68 		return x;
69 
70 
71 		/* Infinity handling */
72 
73 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
74 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
75 		SETCX(IEEE754_INVALID_OPERATION);
76 		return ieee754dp_xcpt(ieee754dp_indef(), "mul", x, y);
77 
78 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
79 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
80 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
81 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
82 	case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
83 		return ieee754dp_inf(xs ^ ys);
84 
85 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
86 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
87 	case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
88 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
89 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
90 		return ieee754dp_zero(xs ^ ys);
91 
92 
93 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
94 		DPDNORMX;
95 
96 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
97 		DPDNORMY;
98 		break;
99 
100 	case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
101 		DPDNORMX;
102 		break;
103 
104 	case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
105 		break;
106 	}
107 	/* rm = xm * ym, re = xe+ye basicly */
108 	assert(xm & DP_HIDDEN_BIT);
109 	assert(ym & DP_HIDDEN_BIT);
110 	{
111 		int re = xe + ye;
112 		int rs = xs ^ ys;
113 		u64 rm;
114 
115 		/* shunt to top of word */
116 		xm <<= 64 - (DP_MBITS + 1);
117 		ym <<= 64 - (DP_MBITS + 1);
118 
119 		/* multiply 32bits xm,ym to give high 32bits rm with stickness
120 		 */
121 
122 		/* 32 * 32 => 64 */
123 #define DPXMULT(x, y)	((u64)(x) * (u64)y)
124 
125 		{
126 			unsigned lxm = xm;
127 			unsigned hxm = xm >> 32;
128 			unsigned lym = ym;
129 			unsigned hym = ym >> 32;
130 			u64 lrm;
131 			u64 hrm;
132 
133 			lrm = DPXMULT(lxm, lym);
134 			hrm = DPXMULT(hxm, hym);
135 
136 			{
137 				u64 t = DPXMULT(lxm, hym);
138 				{
139 					u64 at =
140 					    lrm + (t << 32);
141 					hrm += at < lrm;
142 					lrm = at;
143 				}
144 				hrm = hrm + (t >> 32);
145 			}
146 
147 			{
148 				u64 t = DPXMULT(hxm, lym);
149 				{
150 					u64 at =
151 					    lrm + (t << 32);
152 					hrm += at < lrm;
153 					lrm = at;
154 				}
155 				hrm = hrm + (t >> 32);
156 			}
157 			rm = hrm | (lrm != 0);
158 		}
159 
160 		/*
161 		 * sticky shift down to normal rounding precision
162 		 */
163 		if ((s64) rm < 0) {
164 			rm =
165 			    (rm >> (64 - (DP_MBITS + 1 + 3))) |
166 			    ((rm << (DP_MBITS + 1 + 3)) != 0);
167 			re++;
168 		} else {
169 			rm =
170 			    (rm >> (64 - (DP_MBITS + 1 + 3 + 1))) |
171 			    ((rm << (DP_MBITS + 1 + 3 + 1)) != 0);
172 		}
173 		assert(rm & (DP_HIDDEN_BIT << 3));
174 		DPNORMRET2(rs, re, rm, "mul", x, y);
175 	}
176 }
177