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 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | * 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 Baechle | 3f7cac4 | 2014-04-26 01:49:14 +0200 | [diff] [blame] | 19 | * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | */ |
| 21 | |
Ralf Baechle | cae5506 | 2014-04-16 00:47:59 +0200 | [diff] [blame] | 22 | #include <linux/compiler.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | |
| 24 | #include "ieee754dp.h" |
| 25 | |
Ralf Baechle | 2209bcb | 2014-04-16 01:31:11 +0200 | [diff] [blame] | 26 | int ieee754dp_class(union ieee754dp x) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | { |
| 28 | COMPXDP; |
| 29 | EXPLODEXDP; |
| 30 | return xc; |
| 31 | } |
| 32 | |
Ralf Baechle | 2209bcb | 2014-04-16 01:31:11 +0200 | [diff] [blame] | 33 | int ieee754dp_isnan(union ieee754dp x) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | { |
Maciej W. Rozycki | c9a1084 | 2015-04-03 23:25:38 +0100 | [diff] [blame^] | 35 | return ieee754_class_nan(ieee754dp_class(x)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | } |
| 37 | |
Ralf Baechle | f71baa1 | 2014-04-25 00:40:42 +0200 | [diff] [blame] | 38 | static inline int ieee754dp_issnan(union ieee754dp x) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | { |
| 40 | assert(ieee754dp_isnan(x)); |
Ralf Baechle | 635c9907 | 2014-10-21 14:12:49 +0200 | [diff] [blame] | 41 | return (DPMANT(x) & DP_MBIT(DP_FBITS - 1)) == DP_MBIT(DP_FBITS - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | } |
| 43 | |
| 44 | |
Maciej W. Rozycki | d5afa7e | 2015-04-03 23:25:34 +0100 | [diff] [blame] | 45 | /* |
| 46 | * Raise the Invalid Operation IEEE 754 exception |
| 47 | * and convert the signaling NaN supplied to a quiet NaN. |
| 48 | */ |
Ralf Baechle | 90efba3 | 2014-04-25 03:19:57 +0200 | [diff] [blame] | 49 | union ieee754dp __cold ieee754dp_nanxcpt(union ieee754dp r) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | { |
Maciej W. Rozycki | d5afa7e | 2015-04-03 23:25:34 +0100 | [diff] [blame] | 51 | assert(ieee754dp_issnan(r)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | |
Maciej W. Rozycki | d5afa7e | 2015-04-03 23:25:34 +0100 | [diff] [blame] | 53 | ieee754_setcx(IEEE754_INVALID_OPERATION); |
| 54 | return ieee754dp_indef(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | } |
| 56 | |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 57 | static u64 ieee754dp_get_rounding(int sn, u64 xm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | { |
| 59 | /* inexact must round of 3 bits |
| 60 | */ |
| 61 | if (xm & (DP_MBIT(3) - 1)) { |
| 62 | switch (ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 63 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 65 | case FPU_CSR_RN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | xm += 0x3 + ((xm >> 3) & 1); |
| 67 | /* xm += (xm&0x8)?0x4:0x3 */ |
| 68 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 69 | case FPU_CSR_RU: /* toward +Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | if (!sn) /* ?? */ |
| 71 | xm += 0x8; |
| 72 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 73 | case FPU_CSR_RD: /* toward -Infinity */ |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 74 | if (sn) /* ?? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | xm += 0x8; |
| 76 | break; |
| 77 | } |
| 78 | } |
| 79 | return xm; |
| 80 | } |
| 81 | |
| 82 | |
| 83 | /* generate a normal/denormal number with over,under handling |
| 84 | * sn is sign |
| 85 | * xe is an unbiased exponent |
| 86 | * xm is 3bit extended precision value. |
| 87 | */ |
Ralf Baechle | 2209bcb | 2014-04-16 01:31:11 +0200 | [diff] [blame] | 88 | union ieee754dp ieee754dp_format(int sn, int xe, u64 xm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | { |
| 90 | assert(xm); /* we don't gen exact zeros (probably should) */ |
| 91 | |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 92 | assert((xm >> (DP_FBITS + 1 + 3)) == 0); /* no execess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | assert(xm & (DP_HIDDEN_BIT << 3)); |
| 94 | |
| 95 | if (xe < DP_EMIN) { |
| 96 | /* strip lower bits */ |
| 97 | int es = DP_EMIN - xe; |
| 98 | |
| 99 | if (ieee754_csr.nod) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 100 | ieee754_setcx(IEEE754_UNDERFLOW); |
| 101 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | |
| 103 | switch(ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 104 | case FPU_CSR_RN: |
| 105 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | return ieee754dp_zero(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 107 | case FPU_CSR_RU: /* toward +Infinity */ |
Ralf Baechle | 47fa0c0 | 2014-04-16 11:00:12 +0200 | [diff] [blame] | 108 | if (sn == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | return ieee754dp_min(0); |
| 110 | else |
| 111 | return ieee754dp_zero(1); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 112 | case FPU_CSR_RD: /* toward -Infinity */ |
Ralf Baechle | 47fa0c0 | 2014-04-16 11:00:12 +0200 | [diff] [blame] | 113 | if (sn == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | return ieee754dp_zero(0); |
| 115 | else |
| 116 | return ieee754dp_min(1); |
| 117 | } |
| 118 | } |
| 119 | |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 120 | if (xe == DP_EMIN - 1 && |
| 121 | ieee754dp_get_rounding(sn, xm) >> (DP_FBITS + 1 + 3)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | { |
| 123 | /* Not tiny after rounding */ |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 124 | ieee754_setcx(IEEE754_INEXACT); |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 125 | xm = ieee754dp_get_rounding(sn, xm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | xm >>= 1; |
| 127 | /* Clear grs bits */ |
| 128 | xm &= ~(DP_MBIT(3) - 1); |
| 129 | xe++; |
| 130 | } |
| 131 | else { |
| 132 | /* sticky right shift es bits |
| 133 | */ |
| 134 | xm = XDPSRS(xm, es); |
| 135 | xe += es; |
| 136 | assert((xm & (DP_HIDDEN_BIT << 3)) == 0); |
| 137 | assert(xe == DP_EMIN); |
| 138 | } |
| 139 | } |
| 140 | if (xm & (DP_MBIT(3) - 1)) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 141 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | if ((xm & (DP_HIDDEN_BIT << 3)) == 0) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 143 | ieee754_setcx(IEEE754_UNDERFLOW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | } |
| 145 | |
| 146 | /* inexact must round of 3 bits |
| 147 | */ |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 148 | xm = ieee754dp_get_rounding(sn, xm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | /* adjust exponent for rounding add overflowing |
| 150 | */ |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 151 | if (xm >> (DP_FBITS + 3 + 1)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | /* add causes mantissa overflow */ |
| 153 | xm >>= 1; |
| 154 | xe++; |
| 155 | } |
| 156 | } |
| 157 | /* strip grs bits */ |
| 158 | xm >>= 3; |
| 159 | |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 160 | assert((xm >> (DP_FBITS + 1)) == 0); /* no execess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | assert(xe >= DP_EMIN); |
| 162 | |
| 163 | if (xe > DP_EMAX) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 164 | ieee754_setcx(IEEE754_OVERFLOW); |
| 165 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | /* -O can be table indexed by (rm,sn) */ |
| 167 | switch (ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 168 | case FPU_CSR_RN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 169 | return ieee754dp_inf(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 170 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | return ieee754dp_max(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 172 | case FPU_CSR_RU: /* toward +Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | if (sn == 0) |
| 174 | return ieee754dp_inf(0); |
| 175 | else |
| 176 | return ieee754dp_max(1); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 177 | case FPU_CSR_RD: /* toward -Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | if (sn == 0) |
| 179 | return ieee754dp_max(0); |
| 180 | else |
| 181 | return ieee754dp_inf(1); |
| 182 | } |
| 183 | } |
| 184 | /* gen norm/denorm/zero */ |
| 185 | |
| 186 | if ((xm & DP_HIDDEN_BIT) == 0) { |
| 187 | /* we underflow (tiny/zero) */ |
| 188 | assert(xe == DP_EMIN); |
| 189 | if (ieee754_csr.mx & IEEE754_UNDERFLOW) |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 190 | ieee754_setcx(IEEE754_UNDERFLOW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | return builddp(sn, DP_EMIN - 1 + DP_EBIAS, xm); |
| 192 | } else { |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 193 | assert((xm >> (DP_FBITS + 1)) == 0); /* no execess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | assert(xm & DP_HIDDEN_BIT); |
| 195 | |
| 196 | return builddp(sn, xe + DP_EBIAS, xm & ~DP_HIDDEN_BIT); |
| 197 | } |
| 198 | } |