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 | { |
| 35 | return ieee754dp_class(x) >= IEEE754_CLASS_SNAN; |
| 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 | ad8fb553 | 2014-04-22 15:51:55 +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 | |
Ralf Baechle | 90efba3 | 2014-04-25 03:19:57 +0200 | [diff] [blame] | 45 | union ieee754dp __cold ieee754dp_nanxcpt(union ieee754dp r) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | assert(ieee754dp_isnan(r)); |
| 48 | |
| 49 | if (!ieee754dp_issnan(r)) /* QNAN does not cause invalid op !! */ |
| 50 | return r; |
| 51 | |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 52 | if (!ieee754_setandtestcx(IEEE754_INVALID_OPERATION)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 53 | /* not enabled convert to a quiet NaN */ |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 54 | DPMANT(r) &= (~DP_MBIT(DP_FBITS-1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | if (ieee754dp_isnan(r)) |
| 56 | return r; |
| 57 | else |
| 58 | return ieee754dp_indef(); |
| 59 | } |
| 60 | |
Ralf Baechle | 90efba3 | 2014-04-25 03:19:57 +0200 | [diff] [blame] | 61 | return r; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | } |
| 63 | |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 64 | static u64 ieee754dp_get_rounding(int sn, u64 xm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | { |
| 66 | /* inexact must round of 3 bits |
| 67 | */ |
| 68 | if (xm & (DP_MBIT(3) - 1)) { |
| 69 | switch (ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 70 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 71 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 72 | case FPU_CSR_RN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | xm += 0x3 + ((xm >> 3) & 1); |
| 74 | /* xm += (xm&0x8)?0x4:0x3 */ |
| 75 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 76 | case FPU_CSR_RU: /* toward +Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | if (!sn) /* ?? */ |
| 78 | xm += 0x8; |
| 79 | break; |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 80 | case FPU_CSR_RD: /* toward -Infinity */ |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 81 | if (sn) /* ?? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | xm += 0x8; |
| 83 | break; |
| 84 | } |
| 85 | } |
| 86 | return xm; |
| 87 | } |
| 88 | |
| 89 | |
| 90 | /* generate a normal/denormal number with over,under handling |
| 91 | * sn is sign |
| 92 | * xe is an unbiased exponent |
| 93 | * xm is 3bit extended precision value. |
| 94 | */ |
Ralf Baechle | 2209bcb | 2014-04-16 01:31:11 +0200 | [diff] [blame] | 95 | union ieee754dp ieee754dp_format(int sn, int xe, u64 xm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | { |
| 97 | assert(xm); /* we don't gen exact zeros (probably should) */ |
| 98 | |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 99 | assert((xm >> (DP_FBITS + 1 + 3)) == 0); /* no execess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | assert(xm & (DP_HIDDEN_BIT << 3)); |
| 101 | |
| 102 | if (xe < DP_EMIN) { |
| 103 | /* strip lower bits */ |
| 104 | int es = DP_EMIN - xe; |
| 105 | |
| 106 | if (ieee754_csr.nod) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 107 | ieee754_setcx(IEEE754_UNDERFLOW); |
| 108 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | |
| 110 | switch(ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 111 | case FPU_CSR_RN: |
| 112 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | return ieee754dp_zero(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 114 | case FPU_CSR_RU: /* toward +Infinity */ |
Ralf Baechle | 47fa0c0 | 2014-04-16 11:00:12 +0200 | [diff] [blame] | 115 | if (sn == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | return ieee754dp_min(0); |
| 117 | else |
| 118 | return ieee754dp_zero(1); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 119 | case FPU_CSR_RD: /* toward -Infinity */ |
Ralf Baechle | 47fa0c0 | 2014-04-16 11:00:12 +0200 | [diff] [blame] | 120 | if (sn == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | return ieee754dp_zero(0); |
| 122 | else |
| 123 | return ieee754dp_min(1); |
| 124 | } |
| 125 | } |
| 126 | |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 127 | if (xe == DP_EMIN - 1 && |
| 128 | ieee754dp_get_rounding(sn, xm) >> (DP_FBITS + 1 + 3)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 129 | { |
| 130 | /* Not tiny after rounding */ |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 131 | ieee754_setcx(IEEE754_INEXACT); |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 132 | xm = ieee754dp_get_rounding(sn, xm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | xm >>= 1; |
| 134 | /* Clear grs bits */ |
| 135 | xm &= ~(DP_MBIT(3) - 1); |
| 136 | xe++; |
| 137 | } |
| 138 | else { |
| 139 | /* sticky right shift es bits |
| 140 | */ |
| 141 | xm = XDPSRS(xm, es); |
| 142 | xe += es; |
| 143 | assert((xm & (DP_HIDDEN_BIT << 3)) == 0); |
| 144 | assert(xe == DP_EMIN); |
| 145 | } |
| 146 | } |
| 147 | if (xm & (DP_MBIT(3) - 1)) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 148 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | if ((xm & (DP_HIDDEN_BIT << 3)) == 0) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 150 | ieee754_setcx(IEEE754_UNDERFLOW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | } |
| 152 | |
| 153 | /* inexact must round of 3 bits |
| 154 | */ |
Ralf Baechle | de2fc34 | 2014-04-25 10:54:06 +0200 | [diff] [blame] | 155 | xm = ieee754dp_get_rounding(sn, xm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | /* adjust exponent for rounding add overflowing |
| 157 | */ |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 158 | if (xm >> (DP_FBITS + 3 + 1)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | /* add causes mantissa overflow */ |
| 160 | xm >>= 1; |
| 161 | xe++; |
| 162 | } |
| 163 | } |
| 164 | /* strip grs bits */ |
| 165 | xm >>= 3; |
| 166 | |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 167 | assert((xm >> (DP_FBITS + 1)) == 0); /* no execess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | assert(xe >= DP_EMIN); |
| 169 | |
| 170 | if (xe > DP_EMAX) { |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 171 | ieee754_setcx(IEEE754_OVERFLOW); |
| 172 | ieee754_setcx(IEEE754_INEXACT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | /* -O can be table indexed by (rm,sn) */ |
| 174 | switch (ieee754_csr.rm) { |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 175 | case FPU_CSR_RN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | return ieee754dp_inf(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 177 | case FPU_CSR_RZ: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | return ieee754dp_max(sn); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 179 | case FPU_CSR_RU: /* toward +Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | if (sn == 0) |
| 181 | return ieee754dp_inf(0); |
| 182 | else |
| 183 | return ieee754dp_max(1); |
Ralf Baechle | 56a6473 | 2014-04-30 11:21:55 +0200 | [diff] [blame^] | 184 | case FPU_CSR_RD: /* toward -Infinity */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | if (sn == 0) |
| 186 | return ieee754dp_max(0); |
| 187 | else |
| 188 | return ieee754dp_inf(1); |
| 189 | } |
| 190 | } |
| 191 | /* gen norm/denorm/zero */ |
| 192 | |
| 193 | if ((xm & DP_HIDDEN_BIT) == 0) { |
| 194 | /* we underflow (tiny/zero) */ |
| 195 | assert(xe == DP_EMIN); |
| 196 | if (ieee754_csr.mx & IEEE754_UNDERFLOW) |
Ralf Baechle | 9e8bad1 | 2014-04-19 00:36:32 +0200 | [diff] [blame] | 197 | ieee754_setcx(IEEE754_UNDERFLOW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | return builddp(sn, DP_EMIN - 1 + DP_EBIAS, xm); |
| 199 | } else { |
Ralf Baechle | ad8fb553 | 2014-04-22 15:51:55 +0200 | [diff] [blame] | 200 | assert((xm >> (DP_FBITS + 1)) == 0); /* no execess */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | assert(xm & DP_HIDDEN_BIT); |
| 202 | |
| 203 | return builddp(sn, xe + DP_EBIAS, xm & ~DP_HIDDEN_BIT); |
| 204 | } |
| 205 | } |