Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* IEEE754 floating point arithmetic |
| 2 | * double precision: common utilities |
| 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 "ieee754dp.h" |
| 28 | |
| 29 | ieee754dp ieee754dp_sub(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(), "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; |
| 77 | SETCX(IEEE754_INVALID_OPERATION); |
| 78 | return ieee754dp_xcpt(ieee754dp_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 ieee754dp_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 ieee754dp_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 */ |
| 107 | DPSIGN(y) ^= 1; |
| 108 | return y; |
| 109 | |
| 110 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM): |
| 111 | DPDNORMX; |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 112 | /* FALL THROUGH */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | |
| 114 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM): |
| 115 | /* normalize ym,ye */ |
| 116 | DPDNORMY; |
| 117 | break; |
| 118 | |
| 119 | case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM): |
| 120 | /* normalize xm,xe */ |
| 121 | DPDNORMX; |
| 122 | break; |
| 123 | |
| 124 | case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM): |
| 125 | break; |
| 126 | } |
| 127 | /* flip sign of y and handle as add */ |
| 128 | ys ^= 1; |
| 129 | |
| 130 | assert(xm & DP_HIDDEN_BIT); |
| 131 | assert(ym & DP_HIDDEN_BIT); |
| 132 | |
| 133 | |
| 134 | /* provide guard,round and stick bit dpace */ |
| 135 | xm <<= 3; |
| 136 | ym <<= 3; |
| 137 | |
| 138 | if (xe > ye) { |
| 139 | /* have to shift y fraction right to align |
| 140 | */ |
| 141 | int s = xe - ye; |
| 142 | ym = XDPSRS(ym, s); |
| 143 | ye += s; |
| 144 | } else if (ye > xe) { |
| 145 | /* have to shift x fraction right to align |
| 146 | */ |
| 147 | int s = ye - xe; |
| 148 | xm = XDPSRS(xm, s); |
| 149 | xe += s; |
| 150 | } |
| 151 | assert(xe == ye); |
| 152 | assert(xe <= DP_EMAX); |
| 153 | |
| 154 | if (xs == ys) { |
| 155 | /* generate 28 bit result of adding two 27 bit numbers |
| 156 | */ |
| 157 | xm = xm + ym; |
| 158 | xe = xe; |
| 159 | xs = xs; |
| 160 | |
| 161 | if (xm >> (DP_MBITS + 1 + 3)) { /* carry out */ |
| 162 | xm = XDPSRS1(xm); /* shift preserving sticky */ |
| 163 | xe++; |
| 164 | } |
| 165 | } else { |
| 166 | if (xm >= ym) { |
| 167 | xm = xm - ym; |
| 168 | xe = xe; |
| 169 | xs = xs; |
| 170 | } else { |
| 171 | xm = ym - xm; |
| 172 | xe = xe; |
| 173 | xs = ys; |
| 174 | } |
| 175 | if (xm == 0) { |
| 176 | if (ieee754_csr.rm == IEEE754_RD) |
| 177 | return ieee754dp_zero(1); /* round negative inf. => sign = -1 */ |
| 178 | else |
| 179 | return ieee754dp_zero(0); /* other round modes => sign = 1 */ |
| 180 | } |
| 181 | |
| 182 | /* normalize to rounding precision |
| 183 | */ |
| 184 | while ((xm >> (DP_MBITS + 3)) == 0) { |
| 185 | xm <<= 1; |
| 186 | xe--; |
| 187 | } |
| 188 | } |
| 189 | DPNORMRET2(xs, xe, xm, "sub", x, y); |
| 190 | } |