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 | |
Maciej W. Rozycki | e06b530 | 2015-04-03 23:25:57 +0100 | [diff] [blame] | 33 | static inline 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 | { |
Maciej W. Rozycki | 90d53a9 | 2015-11-13 00:47:28 +0000 | [diff] [blame] | 40 | int qbit; |
| 41 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | assert(ieee754dp_isnan(x)); |
Maciej W. Rozycki | 90d53a9 | 2015-11-13 00:47:28 +0000 | [diff] [blame] | 43 | qbit = (DPMANT(x) & DP_MBIT(DP_FBITS - 1)) == DP_MBIT(DP_FBITS - 1); |
| 44 | return ieee754_csr.nan2008 ^ qbit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | } |
| 46 | |
| 47 | |
Maciej W. Rozycki | d5afa7e | 2015-04-03 23:25:34 +0100 | [diff] [blame] | 48 | /* |
| 49 | * Raise the Invalid Operation IEEE 754 exception |
| 50 | * and convert the signaling NaN supplied to a quiet NaN. |
| 51 | */ |
Ralf Baechle | 90efba3 | 2014-04-25 03:19:57 +0200 | [diff] [blame] | 52 | union ieee754dp __cold ieee754dp_nanxcpt(union ieee754dp r) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 53 | { |
Maciej W. Rozycki | d5afa7e | 2015-04-03 23:25:34 +0100 | [diff] [blame] | 54 | assert(ieee754dp_issnan(r)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | |
Maciej W. Rozycki | d5afa7e | 2015-04-03 23:25:34 +0100 | [diff] [blame] | 56 | ieee754_setcx(IEEE754_INVALID_OPERATION); |
Maciej W. Rozycki | acd9e20 | 2015-11-13 00:48:48 +0000 | [diff] [blame] | 57 | if (ieee754_csr.nan2008) { |
Maciej W. Rozycki | 90d53a9 | 2015-11-13 00:47:28 +0000 | [diff] [blame] | 58 | DPMANT(r) |= DP_MBIT(DP_FBITS - 1); |
Maciej W. Rozycki | acd9e20 | 2015-11-13 00:48:48 +0000 | [diff] [blame] | 59 | } else { |
| 60 | DPMANT(r) &= ~DP_MBIT(DP_FBITS - 1); |
| 61 | if (!ieee754dp_isnan(r)) |
| 62 | DPMANT(r) |= DP_MBIT(DP_FBITS - 2); |
| 63 | } |
Maciej W. Rozycki | 90d53a9 | 2015-11-13 00:47:28 +0000 | [diff] [blame] | 64 | |
| 65 | return r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | } |
| 67 | |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 68 | static u64 ieee754dp_get_rounding(int sn, u64 xm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | { |
| 70 | /* inexact must round of 3 bits |
| 71 | */ |
| 72 | if (xm & (DP_MBIT(3) - 1)) { |
| 73 | switch (ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 74 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 76 | case FPU_CSR_RN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | xm += 0x3 + ((xm >> 3) & 1); |
| 78 | /* xm += (xm&0x8)?0x4:0x3 */ |
| 79 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 80 | case FPU_CSR_RU: /* toward +Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | if (!sn) /* ?? */ |
| 82 | xm += 0x8; |
| 83 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 84 | case FPU_CSR_RD: /* toward -Infinity */ |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 85 | if (sn) /* ?? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | xm += 0x8; |
| 87 | break; |
| 88 | } |
| 89 | } |
| 90 | return xm; |
| 91 | } |
| 92 | |
| 93 | |
| 94 | /* generate a normal/denormal number with over,under handling |
| 95 | * sn is sign |
| 96 | * xe is an unbiased exponent |
| 97 | * xm is 3bit extended precision value. |
| 98 | */ |
Ralf Baechle | 2209bcb | 2014-04-16 01:31:11 +0200 | [diff] [blame] | 99 | union ieee754dp ieee754dp_format(int sn, int xe, u64 xm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | { |
| 101 | assert(xm); /* we don't gen exact zeros (probably should) */ |
| 102 | |
Adam Buchbinder | 92a76f6 | 2016-02-25 00:44:58 -0800 | [diff] [blame] | 103 | assert((xm >> (DP_FBITS + 1 + 3)) == 0); /* no excess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | assert(xm & (DP_HIDDEN_BIT << 3)); |
| 105 | |
| 106 | if (xe < DP_EMIN) { |
| 107 | /* strip lower bits */ |
| 108 | int es = DP_EMIN - xe; |
| 109 | |
| 110 | if (ieee754_csr.nod) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 111 | ieee754_setcx(IEEE754_UNDERFLOW); |
| 112 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | |
| 114 | switch(ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 115 | case FPU_CSR_RN: |
| 116 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | return ieee754dp_zero(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 118 | case FPU_CSR_RU: /* toward +Infinity */ |
Ralf Baechle | 47fa0c0 | 2014-04-16 11:00:12 +0200 | [diff] [blame] | 119 | if (sn == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | return ieee754dp_min(0); |
| 121 | else |
| 122 | return ieee754dp_zero(1); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 123 | case FPU_CSR_RD: /* toward -Infinity */ |
Ralf Baechle | 47fa0c0 | 2014-04-16 11:00:12 +0200 | [diff] [blame] | 124 | if (sn == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | return ieee754dp_zero(0); |
| 126 | else |
| 127 | return ieee754dp_min(1); |
| 128 | } |
| 129 | } |
| 130 | |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 131 | if (xe == DP_EMIN - 1 && |
| 132 | ieee754dp_get_rounding(sn, xm) >> (DP_FBITS + 1 + 3)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | { |
| 134 | /* Not tiny after rounding */ |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 135 | ieee754_setcx(IEEE754_INEXACT); |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 136 | xm = ieee754dp_get_rounding(sn, xm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | xm >>= 1; |
| 138 | /* Clear grs bits */ |
| 139 | xm &= ~(DP_MBIT(3) - 1); |
| 140 | xe++; |
| 141 | } |
| 142 | else { |
| 143 | /* sticky right shift es bits |
| 144 | */ |
| 145 | xm = XDPSRS(xm, es); |
| 146 | xe += es; |
| 147 | assert((xm & (DP_HIDDEN_BIT << 3)) == 0); |
| 148 | assert(xe == DP_EMIN); |
| 149 | } |
| 150 | } |
| 151 | if (xm & (DP_MBIT(3) - 1)) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 152 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | if ((xm & (DP_HIDDEN_BIT << 3)) == 0) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 154 | ieee754_setcx(IEEE754_UNDERFLOW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | } |
| 156 | |
| 157 | /* inexact must round of 3 bits |
| 158 | */ |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 159 | xm = ieee754dp_get_rounding(sn, xm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | /* adjust exponent for rounding add overflowing |
| 161 | */ |
Ralf Baechle | ad8fb55 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 162 | if (xm >> (DP_FBITS + 3 + 1)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | /* add causes mantissa overflow */ |
| 164 | xm >>= 1; |
| 165 | xe++; |
| 166 | } |
| 167 | } |
| 168 | /* strip grs bits */ |
| 169 | xm >>= 3; |
| 170 | |
Adam Buchbinder | 92a76f6 | 2016-02-25 00:44:58 -0800 | [diff] [blame] | 171 | assert((xm >> (DP_FBITS + 1)) == 0); /* no excess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | assert(xe >= DP_EMIN); |
| 173 | |
| 174 | if (xe > DP_EMAX) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 175 | ieee754_setcx(IEEE754_OVERFLOW); |
| 176 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | /* -O can be table indexed by (rm,sn) */ |
| 178 | switch (ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 179 | case FPU_CSR_RN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | return ieee754dp_inf(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 181 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | return ieee754dp_max(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 183 | case FPU_CSR_RU: /* toward +Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | if (sn == 0) |
| 185 | return ieee754dp_inf(0); |
| 186 | else |
| 187 | return ieee754dp_max(1); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame] | 188 | case FPU_CSR_RD: /* toward -Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | if (sn == 0) |
| 190 | return ieee754dp_max(0); |
| 191 | else |
| 192 | return ieee754dp_inf(1); |
| 193 | } |
| 194 | } |
| 195 | /* gen norm/denorm/zero */ |
| 196 | |
| 197 | if ((xm & DP_HIDDEN_BIT) == 0) { |
| 198 | /* we underflow (tiny/zero) */ |
| 199 | assert(xe == DP_EMIN); |
| 200 | if (ieee754_csr.mx & IEEE754_UNDERFLOW) |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 201 | ieee754_setcx(IEEE754_UNDERFLOW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | return builddp(sn, DP_EMIN - 1 + DP_EBIAS, xm); |
| 203 | } else { |
Adam Buchbinder | 92a76f6 | 2016-02-25 00:44:58 -0800 | [diff] [blame] | 204 | assert((xm >> (DP_FBITS + 1)) == 0); /* no excess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | assert(xm & DP_HIDDEN_BIT); |
| 206 | |
| 207 | return builddp(sn, xe + DP_EBIAS, xm & ~DP_HIDDEN_BIT); |
| 208 | } |
| 209 | } |