Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* IEEE754 floating point arithmetic |
| 2 | * single precision |
| 3 | */ |
| 4 | /* |
| 5 | * MIPS floating point support |
| 6 | * Copyright (C) 1994-2000 Algorithmics Ltd. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 "ieee754sp.h" |
| 28 | |
Ralf Baechle | 2209bcb | 2014-04-16 01:31:11 +0200 | [diff] [blame] | 29 | union ieee754sp ieee754sp_sub(union ieee754sp x, union ieee754sp y) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | { |
| 31 | COMPXSP; |
| 32 | COMPYSP; |
| 33 | |
| 34 | EXPLODEXSP; |
| 35 | EXPLODEYSP; |
| 36 | |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 37 | ieee754_clearcx(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | |
| 39 | FLUSHXSP; |
| 40 | FLUSHYSP; |
| 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): |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 54 | ieee754_setcx(IEEE754_INVALID_OPERATION); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | return ieee754sp_nanxcpt(ieee754sp_indef(), "sub", 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 | |
| 74 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF): |
| 75 | if (xs != ys) |
| 76 | return x; |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 77 | ieee754_setcx(IEEE754_INVALID_OPERATION); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | return ieee754sp_xcpt(ieee754sp_indef(), "sub", x, y); |
| 79 | |
| 80 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF): |
| 81 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF): |
| 82 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF): |
| 83 | return ieee754sp_inf(ys ^ 1); |
| 84 | |
| 85 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO): |
| 86 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM): |
| 87 | case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM): |
| 88 | return x; |
| 89 | |
| 90 | /* Zero handling |
| 91 | */ |
| 92 | |
| 93 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO): |
| 94 | if (xs != ys) |
| 95 | return x; |
| 96 | else |
| 97 | return ieee754sp_zero(ieee754_csr.rm == |
| 98 | IEEE754_RD); |
| 99 | |
| 100 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO): |
| 101 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO): |
| 102 | return x; |
| 103 | |
| 104 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM): |
| 105 | case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM): |
| 106 | /* quick fix up */ |
Ralf Baechle | 3e160aa | 2014-04-22 16:02:03 +0200 | [diff] [blame^] | 107 | SPSIGN(y) ^= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | return y; |
| 109 | |
| 110 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
| 111 | SPDNORMX; |
| 112 | |
| 113 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM): |
| 114 | SPDNORMY; |
| 115 | break; |
| 116 | |
| 117 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM): |
| 118 | SPDNORMX; |
| 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 & SP_HIDDEN_BIT); |
| 128 | assert(ym & SP_HIDDEN_BIT); |
| 129 | |
| 130 | |
| 131 | /* provide guard,round and stick bit space */ |
| 132 | xm <<= 3; |
| 133 | ym <<= 3; |
| 134 | |
| 135 | if (xe > ye) { |
| 136 | /* have to shift y fraction right to align |
| 137 | */ |
| 138 | int s = xe - ye; |
| 139 | SPXSRSYn(s); |
| 140 | } else if (ye > xe) { |
| 141 | /* have to shift x fraction right to align |
| 142 | */ |
| 143 | int s = ye - xe; |
| 144 | SPXSRSXn(s); |
| 145 | } |
| 146 | assert(xe == ye); |
| 147 | assert(xe <= SP_EMAX); |
| 148 | |
| 149 | if (xs == ys) { |
| 150 | /* generate 28 bit result of adding two 27 bit numbers |
| 151 | */ |
| 152 | xm = xm + ym; |
| 153 | xe = xe; |
| 154 | xs = xs; |
| 155 | |
Ralf Baechle | ad8fb55 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 156 | if (xm >> (SP_FBITS + 1 + 3)) { /* carry out */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | SPXSRSX1(); /* shift preserving sticky */ |
| 158 | } |
| 159 | } else { |
| 160 | if (xm >= ym) { |
| 161 | xm = xm - ym; |
| 162 | xe = xe; |
| 163 | xs = xs; |
| 164 | } else { |
| 165 | xm = ym - xm; |
| 166 | xe = xe; |
| 167 | xs = ys; |
| 168 | } |
| 169 | if (xm == 0) { |
| 170 | if (ieee754_csr.rm == IEEE754_RD) |
| 171 | return ieee754sp_zero(1); /* round negative inf. => sign = -1 */ |
| 172 | else |
| 173 | return ieee754sp_zero(0); /* other round modes => sign = 1 */ |
| 174 | } |
| 175 | /* normalize to rounding precision |
| 176 | */ |
Ralf Baechle | ad8fb55 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 177 | while ((xm >> (SP_FBITS + 3)) == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | xm <<= 1; |
| 179 | xe--; |
| 180 | } |
| 181 | } |
| 182 | SPNORMRET2(xs, xe, xm, "sub", x, y); |
| 183 | } |