blob: fc17a781b9ae868890303f57b08eb6f7b7d72cd3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* IEEE754 floating point arithmetic
2 * double precision: common utilities
3 */
4/*
5 * MIPS floating point support
6 * Copyright (C) 1994-2000 Algorithmics Ltd.
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * 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.,
Ralf Baechle3f7cac42014-04-26 01:49:14 +020019 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 */
21
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "ieee754dp.h"
23
Ralf Baechle2209bcb2014-04-16 01:31:11 +020024union ieee754dp ieee754dp_sub(union ieee754dp x, union ieee754dp y)
Linus Torvalds1da177e2005-04-16 15:20:36 -070025{
Ralf Baechle3f7cac42014-04-26 01:49:14 +020026 int s;
27
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 COMPXDP;
29 COMPYDP;
30
31 EXPLODEXDP;
32 EXPLODEYDP;
33
Ralf Baechle9e8bad12014-04-19 00:36:32 +020034 ieee754_clearcx();
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36 FLUSHXDP;
37 FLUSHYDP;
38
39 switch (CLPAIR(xc, yc)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN):
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN):
42 case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN):
43 case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN):
44 case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN):
Maciej W. Rozyckid5afa7e2015-04-03 23:25:34 +010045 return ieee754dp_nanxcpt(y);
46
47 case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN):
48 case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN):
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO):
50 case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM):
51 case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM):
52 case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF):
Maciej W. Rozyckid5afa7e2015-04-03 23:25:34 +010053 return ieee754dp_nanxcpt(x);
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55 case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN):
56 case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN):
57 case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN):
58 case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN):
59 return y;
60
61 case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN):
62 case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO):
63 case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM):
64 case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM):
65 case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF):
66 return x;
67
68
Ralf Baechle3f7cac42014-04-26 01:49:14 +020069 /*
70 * Infinity handling
71 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
73 if (xs != ys)
74 return x;
Ralf Baechle9e8bad12014-04-19 00:36:32 +020075 ieee754_setcx(IEEE754_INVALID_OPERATION);
Ralf Baechle90efba32014-04-25 03:19:57 +020076 return ieee754dp_indef();
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78 case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
79 case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
80 case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
81 return ieee754dp_inf(ys ^ 1);
82
83 case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
84 case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
85 case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
86 return x;
87
Ralf Baechle3f7cac42014-04-26 01:49:14 +020088 /*
89 * Zero handling
90 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
92 if (xs != ys)
93 return x;
94 else
Ralf Baechle56a64732014-04-30 11:21:55 +020095 return ieee754dp_zero(ieee754_csr.rm == FPU_CSR_RD);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97 case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
98 case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
99 return x;
100
101 case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
102 case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
103 /* quick fix up */
104 DPSIGN(y) ^= 1;
105 return y;
106
107 case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
108 DPDNORMX;
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200109 /* FALL THROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
111 case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
112 /* normalize ym,ye */
113 DPDNORMY;
114 break;
115
116 case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
117 /* normalize xm,xe */
118 DPDNORMX;
119 break;
120
121 case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
122 break;
123 }
124 /* flip sign of y and handle as add */
125 ys ^= 1;
126
127 assert(xm & DP_HIDDEN_BIT);
128 assert(ym & DP_HIDDEN_BIT);
129
130
131 /* provide guard,round and stick bit dpace */
132 xm <<= 3;
133 ym <<= 3;
134
135 if (xe > ye) {
Ralf Baechle3f7cac42014-04-26 01:49:14 +0200136 /*
137 * Have to shift y fraction right to align
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 */
Ralf Baechle3f7cac42014-04-26 01:49:14 +0200139 s = xe - ye;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 ym = XDPSRS(ym, s);
141 ye += s;
142 } else if (ye > xe) {
Ralf Baechle3f7cac42014-04-26 01:49:14 +0200143 /*
144 * Have to shift x fraction right to align
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 */
Ralf Baechle3f7cac42014-04-26 01:49:14 +0200146 s = ye - xe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 xm = XDPSRS(xm, s);
148 xe += s;
149 }
150 assert(xe == ye);
151 assert(xe <= DP_EMAX);
152
153 if (xs == ys) {
154 /* generate 28 bit result of adding two 27 bit numbers
155 */
156 xm = xm + ym;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Ralf Baechlead8fb552014-04-22 15:51:55 +0200158 if (xm >> (DP_FBITS + 1 + 3)) { /* carry out */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 xm = XDPSRS1(xm); /* shift preserving sticky */
160 xe++;
161 }
162 } else {
163 if (xm >= ym) {
164 xm = xm - ym;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 } else {
166 xm = ym - xm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 xs = ys;
168 }
169 if (xm == 0) {
Ralf Baechle56a64732014-04-30 11:21:55 +0200170 if (ieee754_csr.rm == FPU_CSR_RD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 return ieee754dp_zero(1); /* round negative inf. => sign = -1 */
172 else
173 return ieee754dp_zero(0); /* other round modes => sign = 1 */
174 }
175
176 /* normalize to rounding precision
177 */
Ralf Baechlead8fb552014-04-22 15:51:55 +0200178 while ((xm >> (DP_FBITS + 3)) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 xm <<= 1;
180 xe--;
181 }
182 }
Ralf Baechle90efba32014-04-25 03:19:57 +0200183
184 return ieee754dp_format(xs, xe, xm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185}