Merge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
[linux-2.6] / arch / mips / math-emu / dp_mul.c
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  * http://www.algor.co.uk
8  *
9  * ########################################################################
10  *
11  *  This program is free software; you can distribute it and/or modify it
12  *  under the terms of the GNU General Public License (Version 2) as
13  *  published by the Free Software Foundation.
14  *
15  *  This program is distributed in the hope it will be useful, but WITHOUT
16  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
17  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
18  *  for more details.
19  *
20  *  You should have received a copy of the GNU General Public License along
21  *  with this program; if not, write to the Free Software Foundation, Inc.,
22  *  59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
23  *
24  * ########################################################################
25  */
26
27
28 #include "ieee754dp.h"
29
30 ieee754dp ieee754dp_mul(ieee754dp x, ieee754dp y)
31 {
32         COMPXDP;
33         COMPYDP;
34
35         EXPLODEXDP;
36         EXPLODEYDP;
37
38         CLEARCX;
39
40         FLUSHXDP;
41         FLUSHYDP;
42
43         switch (CLPAIR(xc, yc)) {
44         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN):
45         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN):
46         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN):
47         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN):
48         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN):
49         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN):
50         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN):
51         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO):
52         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM):
53         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM):
54         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF):
55                 SETCX(IEEE754_INVALID_OPERATION);
56                 return ieee754dp_nanxcpt(ieee754dp_indef(), "mul", x, y);
57
58         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN):
59         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN):
60         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN):
61         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN):
62                 return y;
63
64         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN):
65         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO):
66         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM):
67         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM):
68         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF):
69                 return x;
70
71
72                 /* Infinity handling */
73
74         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
75         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
76                 SETCX(IEEE754_INVALID_OPERATION);
77                 return ieee754dp_xcpt(ieee754dp_indef(), "mul", x, y);
78
79         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
80         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
81         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
82         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
83         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
84                 return ieee754dp_inf(xs ^ ys);
85
86         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
87         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
88         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
89         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
90         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
91                 return ieee754dp_zero(xs ^ ys);
92
93
94         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
95                 DPDNORMX;
96
97         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
98                 DPDNORMY;
99                 break;
100
101         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
102                 DPDNORMX;
103                 break;
104
105         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
106                 break;
107         }
108         /* rm = xm * ym, re = xe+ye basicly */
109         assert(xm & DP_HIDDEN_BIT);
110         assert(ym & DP_HIDDEN_BIT);
111         {
112                 int re = xe + ye;
113                 int rs = xs ^ ys;
114                 u64 rm;
115
116                 /* shunt to top of word */
117                 xm <<= 64 - (DP_MBITS + 1);
118                 ym <<= 64 - (DP_MBITS + 1);
119
120                 /* multiply 32bits xm,ym to give high 32bits rm with stickness
121                  */
122
123                 /* 32 * 32 => 64 */
124 #define DPXMULT(x, y)   ((u64)(x) * (u64)y)
125
126                 {
127                         unsigned lxm = xm;
128                         unsigned hxm = xm >> 32;
129                         unsigned lym = ym;
130                         unsigned hym = ym >> 32;
131                         u64 lrm;
132                         u64 hrm;
133
134                         lrm = DPXMULT(lxm, lym);
135                         hrm = DPXMULT(hxm, hym);
136
137                         {
138                                 u64 t = DPXMULT(lxm, hym);
139                                 {
140                                         u64 at =
141                                             lrm + (t << 32);
142                                         hrm += at < lrm;
143                                         lrm = at;
144                                 }
145                                 hrm = hrm + (t >> 32);
146                         }
147
148                         {
149                                 u64 t = DPXMULT(hxm, lym);
150                                 {
151                                         u64 at =
152                                             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                 /*
162                  * sticky shift down to normal rounding precision
163                  */
164                 if ((s64) rm < 0) {
165                         rm =
166                             (rm >> (64 - (DP_MBITS + 1 + 3))) |
167                             ((rm << (DP_MBITS + 1 + 3)) != 0);
168                         re++;
169                 } else {
170                         rm =
171                             (rm >> (64 - (DP_MBITS + 1 + 3 + 1))) |
172                             ((rm << (DP_MBITS + 1 + 3 + 1)) != 0);
173                 }
174                 assert(rm & (DP_HIDDEN_BIT << 3));
175                 DPNORMRET2(rs, re, rm, "mul", x, y);
176         }
177 }