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. |
| 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 | int ieee754sp_class(ieee754sp x) |
| 31 | { |
| 32 | COMPXSP; |
| 33 | EXPLODEXSP; |
| 34 | return xc; |
| 35 | } |
| 36 | |
| 37 | int ieee754sp_isnan(ieee754sp x) |
| 38 | { |
| 39 | return ieee754sp_class(x) >= IEEE754_CLASS_SNAN; |
| 40 | } |
| 41 | |
| 42 | int ieee754sp_issnan(ieee754sp x) |
| 43 | { |
| 44 | assert(ieee754sp_isnan(x)); |
| 45 | return (SPMANT(x) & SP_MBIT(SP_MBITS-1)); |
| 46 | } |
| 47 | |
| 48 | |
| 49 | ieee754sp ieee754sp_xcpt(ieee754sp r, const char *op, ...) |
| 50 | { |
| 51 | struct ieee754xctx ax; |
| 52 | |
| 53 | if (!TSTX()) |
| 54 | return r; |
| 55 | |
| 56 | ax.op = op; |
| 57 | ax.rt = IEEE754_RT_SP; |
| 58 | ax.rv.sp = r; |
| 59 | va_start(ax.ap, op); |
| 60 | ieee754_xcpt(&ax); |
Richard Knutsson | 8142294 | 2007-11-24 22:22:19 +0100 | [diff] [blame] | 61 | va_end(ax.ap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | return ax.rv.sp; |
| 63 | } |
| 64 | |
| 65 | ieee754sp ieee754sp_nanxcpt(ieee754sp r, const char *op, ...) |
| 66 | { |
| 67 | struct ieee754xctx ax; |
| 68 | |
| 69 | assert(ieee754sp_isnan(r)); |
| 70 | |
| 71 | if (!ieee754sp_issnan(r)) /* QNAN does not cause invalid op !! */ |
| 72 | return r; |
| 73 | |
| 74 | if (!SETANDTESTCX(IEEE754_INVALID_OPERATION)) { |
| 75 | /* not enabled convert to a quiet NaN */ |
| 76 | SPMANT(r) &= (~SP_MBIT(SP_MBITS-1)); |
| 77 | if (ieee754sp_isnan(r)) |
| 78 | return r; |
| 79 | else |
| 80 | return ieee754sp_indef(); |
| 81 | } |
| 82 | |
| 83 | ax.op = op; |
| 84 | ax.rt = 0; |
| 85 | ax.rv.sp = r; |
| 86 | va_start(ax.ap, op); |
| 87 | ieee754_xcpt(&ax); |
Richard Knutsson | 8142294 | 2007-11-24 22:22:19 +0100 | [diff] [blame] | 88 | va_end(ax.ap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | return ax.rv.sp; |
| 90 | } |
| 91 | |
| 92 | ieee754sp ieee754sp_bestnan(ieee754sp x, ieee754sp y) |
| 93 | { |
| 94 | assert(ieee754sp_isnan(x)); |
| 95 | assert(ieee754sp_isnan(y)); |
| 96 | |
| 97 | if (SPMANT(x) > SPMANT(y)) |
| 98 | return x; |
| 99 | else |
| 100 | return y; |
| 101 | } |
| 102 | |
| 103 | |
| 104 | static unsigned get_rounding(int sn, unsigned xm) |
| 105 | { |
| 106 | /* inexact must round of 3 bits |
| 107 | */ |
| 108 | if (xm & (SP_MBIT(3) - 1)) { |
| 109 | switch (ieee754_csr.rm) { |
| 110 | case IEEE754_RZ: |
| 111 | break; |
| 112 | case IEEE754_RN: |
| 113 | xm += 0x3 + ((xm >> 3) & 1); |
| 114 | /* xm += (xm&0x8)?0x4:0x3 */ |
| 115 | break; |
| 116 | case IEEE754_RU: /* toward +Infinity */ |
| 117 | if (!sn) /* ?? */ |
| 118 | xm += 0x8; |
| 119 | break; |
| 120 | case IEEE754_RD: /* toward -Infinity */ |
| 121 | if (sn) /* ?? */ |
| 122 | xm += 0x8; |
| 123 | break; |
| 124 | } |
| 125 | } |
| 126 | return xm; |
| 127 | } |
| 128 | |
| 129 | |
| 130 | /* generate a normal/denormal number with over,under handling |
| 131 | * sn is sign |
| 132 | * xe is an unbiased exponent |
| 133 | * xm is 3bit extended precision value. |
| 134 | */ |
| 135 | ieee754sp ieee754sp_format(int sn, int xe, unsigned xm) |
| 136 | { |
| 137 | assert(xm); /* we don't gen exact zeros (probably should) */ |
| 138 | |
| 139 | assert((xm >> (SP_MBITS + 1 + 3)) == 0); /* no execess */ |
| 140 | assert(xm & (SP_HIDDEN_BIT << 3)); |
| 141 | |
| 142 | if (xe < SP_EMIN) { |
| 143 | /* strip lower bits */ |
| 144 | int es = SP_EMIN - xe; |
| 145 | |
| 146 | if (ieee754_csr.nod) { |
| 147 | SETCX(IEEE754_UNDERFLOW); |
| 148 | SETCX(IEEE754_INEXACT); |
| 149 | |
| 150 | switch(ieee754_csr.rm) { |
| 151 | case IEEE754_RN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | case IEEE754_RZ: |
| 153 | return ieee754sp_zero(sn); |
| 154 | case IEEE754_RU: /* toward +Infinity */ |
| 155 | if(sn == 0) |
| 156 | return ieee754sp_min(0); |
| 157 | else |
| 158 | return ieee754sp_zero(1); |
| 159 | case IEEE754_RD: /* toward -Infinity */ |
| 160 | if(sn == 0) |
| 161 | return ieee754sp_zero(0); |
| 162 | else |
| 163 | return ieee754sp_min(1); |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | if (xe == SP_EMIN - 1 |
| 168 | && get_rounding(sn, xm) >> (SP_MBITS + 1 + 3)) |
| 169 | { |
| 170 | /* Not tiny after rounding */ |
| 171 | SETCX(IEEE754_INEXACT); |
| 172 | xm = get_rounding(sn, xm); |
| 173 | xm >>= 1; |
| 174 | /* Clear grs bits */ |
| 175 | xm &= ~(SP_MBIT(3) - 1); |
| 176 | xe++; |
| 177 | } |
| 178 | else { |
| 179 | /* sticky right shift es bits |
| 180 | */ |
| 181 | SPXSRSXn(es); |
| 182 | assert((xm & (SP_HIDDEN_BIT << 3)) == 0); |
| 183 | assert(xe == SP_EMIN); |
| 184 | } |
| 185 | } |
| 186 | if (xm & (SP_MBIT(3) - 1)) { |
| 187 | SETCX(IEEE754_INEXACT); |
| 188 | if ((xm & (SP_HIDDEN_BIT << 3)) == 0) { |
| 189 | SETCX(IEEE754_UNDERFLOW); |
| 190 | } |
| 191 | |
| 192 | /* inexact must round of 3 bits |
| 193 | */ |
| 194 | xm = get_rounding(sn, xm); |
| 195 | /* adjust exponent for rounding add overflowing |
| 196 | */ |
| 197 | if (xm >> (SP_MBITS + 1 + 3)) { |
| 198 | /* add causes mantissa overflow */ |
| 199 | xm >>= 1; |
| 200 | xe++; |
| 201 | } |
| 202 | } |
| 203 | /* strip grs bits */ |
| 204 | xm >>= 3; |
| 205 | |
| 206 | assert((xm >> (SP_MBITS + 1)) == 0); /* no execess */ |
| 207 | assert(xe >= SP_EMIN); |
| 208 | |
| 209 | if (xe > SP_EMAX) { |
| 210 | SETCX(IEEE754_OVERFLOW); |
| 211 | SETCX(IEEE754_INEXACT); |
| 212 | /* -O can be table indexed by (rm,sn) */ |
| 213 | switch (ieee754_csr.rm) { |
| 214 | case IEEE754_RN: |
| 215 | return ieee754sp_inf(sn); |
| 216 | case IEEE754_RZ: |
| 217 | return ieee754sp_max(sn); |
| 218 | case IEEE754_RU: /* toward +Infinity */ |
| 219 | if (sn == 0) |
| 220 | return ieee754sp_inf(0); |
| 221 | else |
| 222 | return ieee754sp_max(1); |
| 223 | case IEEE754_RD: /* toward -Infinity */ |
| 224 | if (sn == 0) |
| 225 | return ieee754sp_max(0); |
| 226 | else |
| 227 | return ieee754sp_inf(1); |
| 228 | } |
| 229 | } |
| 230 | /* gen norm/denorm/zero */ |
| 231 | |
| 232 | if ((xm & SP_HIDDEN_BIT) == 0) { |
| 233 | /* we underflow (tiny/zero) */ |
| 234 | assert(xe == SP_EMIN); |
| 235 | if (ieee754_csr.mx & IEEE754_UNDERFLOW) |
| 236 | SETCX(IEEE754_UNDERFLOW); |
| 237 | return buildsp(sn, SP_EMIN - 1 + SP_EBIAS, xm); |
| 238 | } else { |
| 239 | assert((xm >> (SP_MBITS + 1)) == 0); /* no execess */ |
| 240 | assert(xm & SP_HIDDEN_BIT); |
| 241 | |
| 242 | return buildsp(sn, xe + SP_EBIAS, xm & ~SP_HIDDEN_BIT); |
| 243 | } |
| 244 | } |