Merge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
[linux-2.6] / arch / mips / math-emu / sp_mul.c
1 /* IEEE754 floating point arithmetic
2  * single precision
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 "ieee754sp.h"
29
30 ieee754sp ieee754sp_mul(ieee754sp x, ieee754sp y)
31 {
32         COMPXSP;
33         COMPYSP;
34
35         EXPLODEXSP;
36         EXPLODEYSP;
37
38         CLEARCX;
39
40         FLUSHXSP;
41         FLUSHYSP;
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 ieee754sp_nanxcpt(ieee754sp_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 ieee754sp_xcpt(ieee754sp_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 ieee754sp_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 ieee754sp_zero(xs ^ ys);
92
93
94         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
95                 SPDNORMX;
96
97         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
98                 SPDNORMY;
99                 break;
100
101         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
102                 SPDNORMX;
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 & SP_HIDDEN_BIT);
110         assert(ym & SP_HIDDEN_BIT);
111
112         {
113                 int re = xe + ye;
114                 int rs = xs ^ ys;
115                 unsigned rm;
116
117                 /* shunt to top of word */
118                 xm <<= 32 - (SP_MBITS + 1);
119                 ym <<= 32 - (SP_MBITS + 1);
120
121                 /* multiply 32bits xm,ym to give high 32bits rm with stickness
122                  */
123                 {
124                         unsigned short lxm = xm & 0xffff;
125                         unsigned short hxm = xm >> 16;
126                         unsigned short lym = ym & 0xffff;
127                         unsigned short hym = ym >> 16;
128                         unsigned lrm;
129                         unsigned hrm;
130
131                         lrm = lxm * lym;        /* 16 * 16 => 32 */
132                         hrm = hxm * hym;        /* 16 * 16 => 32 */
133
134                         {
135                                 unsigned t = lxm * hym; /* 16 * 16 => 32 */
136                                 {
137                                         unsigned at = lrm + (t << 16);
138                                         hrm += at < lrm;
139                                         lrm = at;
140                                 }
141                                 hrm = hrm + (t >> 16);
142                         }
143
144                         {
145                                 unsigned t = hxm * lym; /* 16 * 16 => 32 */
146                                 {
147                                         unsigned at = lrm + (t << 16);
148                                         hrm += at < lrm;
149                                         lrm = at;
150                                 }
151                                 hrm = hrm + (t >> 16);
152                         }
153                         rm = hrm | (lrm != 0);
154                 }
155
156                 /*
157                  * sticky shift down to normal rounding precision
158                  */
159                 if ((int) rm < 0) {
160                         rm = (rm >> (32 - (SP_MBITS + 1 + 3))) |
161                             ((rm << (SP_MBITS + 1 + 3)) != 0);
162                         re++;
163                 } else {
164                         rm = (rm >> (32 - (SP_MBITS + 1 + 3 + 1))) |
165                             ((rm << (SP_MBITS + 1 + 3 + 1)) != 0);
166                 }
167                 assert(rm & (SP_HIDDEN_BIT << 3));
168
169                 SPNORMRET2(rs, re, rm, "mul", x, y);
170         }
171 }