| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* IEEE754 floating point arithmetic |
| 3 |
* double precision: common utilities |
| 4 |
*/ |
| 5 |
/* |
| 6 |
* MIPS floating point support |
| 7 |
* Copyright (C) 1994-2000 Algorithmics Ltd. |
| 8 |
*/ |
| 9 |
|
| 10 |
#include <linux/compiler.h> |
| 11 |
|
| 12 |
#include "ieee754dp.h" |
| 13 |
|
| 14 |
int ieee754dp_class(union ieee754dp x) |
| 15 |
{ |
| 16 |
COMPXDP; |
| 17 |
EXPLODEXDP; |
| 18 |
return xc; |
| 19 |
} |
| 20 |
|
| 21 |
static inline int ieee754dp_isnan(union ieee754dp x) |
| 22 |
{ |
| 23 |
return ieee754_class_nan(ieee754dp_class(x)); |
| 24 |
} |
| 25 |
|
| 26 |
static inline int ieee754dp_issnan(union ieee754dp x) |
| 27 |
{ |
| 28 |
int qbit; |
| 29 |
|
| 30 |
assert(ieee754dp_isnan(x)); |
| 31 |
qbit = (DPMANT(x) & DP_MBIT(DP_FBITS - 1)) == DP_MBIT(DP_FBITS - 1); |
| 32 |
return ieee754_csr.nan2008 ^ qbit; |
| 33 |
} |
| 34 |
|
| 35 |
|
| 36 |
/* |
| 37 |
* Raise the Invalid Operation IEEE 754 exception |
| 38 |
* and convert the signaling NaN supplied to a quiet NaN. |
| 39 |
*/ |
| 40 |
union ieee754dp __cold ieee754dp_nanxcpt(union ieee754dp r) |
| 41 |
{ |
| 42 |
assert(ieee754dp_issnan(r)); |
| 43 |
|
| 44 |
ieee754_setcx(IEEE754_INVALID_OPERATION); |
| 45 |
if (ieee754_csr.nan2008) { |
| 46 |
DPMANT(r) |= DP_MBIT(DP_FBITS - 1); |
| 47 |
} else { |
| 48 |
DPMANT(r) &= ~DP_MBIT(DP_FBITS - 1); |
| 49 |
if (!ieee754dp_isnan(r)) |
| 50 |
DPMANT(r) |= DP_MBIT(DP_FBITS - 2); |
| 51 |
} |
| 52 |
|
| 53 |
return r; |
| 54 |
} |
| 55 |
|
| 56 |
static u64 ieee754dp_get_rounding(int sn, u64 xm) |
| 57 |
{ |
| 58 |
/* inexact must round of 3 bits |
| 59 |
*/ |
| 60 |
if (xm & (DP_MBIT(3) - 1)) { |
| 61 |
switch (ieee754_csr.rm) { |
| 62 |
case FPU_CSR_RZ: |
| 63 |
break; |
| 64 |
case FPU_CSR_RN: |
| 65 |
xm += 0x3 + ((xm >> 3) & 1); |
| 66 |
/* xm += (xm&0x8)?0x4:0x3 */ |
| 67 |
break; |
| 68 |
case FPU_CSR_RU: /* toward +Infinity */ |
| 69 |
if (!sn) /* ?? */ |
| 70 |
xm += 0x8; |
| 71 |
break; |
| 72 |
case FPU_CSR_RD: /* toward -Infinity */ |
| 73 |
if (sn) /* ?? */ |
| 74 |
xm += 0x8; |
| 75 |
break; |
| 76 |
} |
| 77 |
} |
| 78 |
return xm; |
| 79 |
} |
| 80 |
|
| 81 |
|
| 82 |
/* generate a normal/denormal number with over,under handling |
| 83 |
* sn is sign |
| 84 |
* xe is an unbiased exponent |
| 85 |
* xm is 3bit extended precision value. |
| 86 |
*/ |
| 87 |
union ieee754dp ieee754dp_format(int sn, int xe, u64 xm) |
| 88 |
{ |
| 89 |
assert(xm); /* we don't gen exact zeros (probably should) */ |
| 90 |
|
| 91 |
assert((xm >> (DP_FBITS + 1 + 3)) == 0); /* no excess */ |
| 92 |
assert(xm & (DP_HIDDEN_BIT << 3)); |
| 93 |
|
| 94 |
if (xe < DP_EMIN) { |
| 95 |
/* strip lower bits */ |
| 96 |
int es = DP_EMIN - xe; |
| 97 |
|
| 98 |
if (ieee754_csr.nod) { |
| 99 |
ieee754_setcx(IEEE754_UNDERFLOW); |
| 100 |
ieee754_setcx(IEEE754_INEXACT); |
| 101 |
|
| 102 |
switch(ieee754_csr.rm) { |
| 103 |
case FPU_CSR_RN: |
| 104 |
case FPU_CSR_RZ: |
| 105 |
return ieee754dp_zero(sn); |
| 106 |
case FPU_CSR_RU: /* toward +Infinity */ |
| 107 |
if (sn == 0) |
| 108 |
return ieee754dp_min(0); |
| 109 |
else |
| 110 |
return ieee754dp_zero(1); |
| 111 |
case FPU_CSR_RD: /* toward -Infinity */ |
| 112 |
if (sn == 0) |
| 113 |
return ieee754dp_zero(0); |
| 114 |
else |
| 115 |
return ieee754dp_min(1); |
| 116 |
} |
| 117 |
} |
| 118 |
|
| 119 |
if (xe == DP_EMIN - 1 && |
| 120 |
ieee754dp_get_rounding(sn, xm) >> (DP_FBITS + 1 + 3)) |
| 121 |
{ |
| 122 |
/* Not tiny after rounding */ |
| 123 |
ieee754_setcx(IEEE754_INEXACT); |
| 124 |
xm = ieee754dp_get_rounding(sn, xm); |
| 125 |
xm >>= 1; |
| 126 |
/* Clear grs bits */ |
| 127 |
xm &= ~(DP_MBIT(3) - 1); |
| 128 |
xe++; |
| 129 |
} |
| 130 |
else { |
| 131 |
/* sticky right shift es bits |
| 132 |
*/ |
| 133 |
xm = XDPSRS(xm, es); |
| 134 |
xe += es; |
| 135 |
assert((xm & (DP_HIDDEN_BIT << 3)) == 0); |
| 136 |
assert(xe == DP_EMIN); |
| 137 |
} |
| 138 |
} |
| 139 |
if (xm & (DP_MBIT(3) - 1)) { |
| 140 |
ieee754_setcx(IEEE754_INEXACT); |
| 141 |
if ((xm & (DP_HIDDEN_BIT << 3)) == 0) { |
| 142 |
ieee754_setcx(IEEE754_UNDERFLOW); |
| 143 |
} |
| 144 |
|
| 145 |
/* inexact must round of 3 bits |
| 146 |
*/ |
| 147 |
xm = ieee754dp_get_rounding(sn, xm); |
| 148 |
/* adjust exponent for rounding add overflowing |
| 149 |
*/ |
| 150 |
if (xm >> (DP_FBITS + 3 + 1)) { |
| 151 |
/* add causes mantissa overflow */ |
| 152 |
xm >>= 1; |
| 153 |
xe++; |
| 154 |
} |
| 155 |
} |
| 156 |
/* strip grs bits */ |
| 157 |
xm >>= 3; |
| 158 |
|
| 159 |
assert((xm >> (DP_FBITS + 1)) == 0); /* no excess */ |
| 160 |
assert(xe >= DP_EMIN); |
| 161 |
|
| 162 |
if (xe > DP_EMAX) { |
| 163 |
ieee754_setcx(IEEE754_OVERFLOW); |
| 164 |
ieee754_setcx(IEEE754_INEXACT); |
| 165 |
/* -O can be table indexed by (rm,sn) */ |
| 166 |
switch (ieee754_csr.rm) { |
| 167 |
case FPU_CSR_RN: |
| 168 |
return ieee754dp_inf(sn); |
| 169 |
case FPU_CSR_RZ: |
| 170 |
return ieee754dp_max(sn); |
| 171 |
case FPU_CSR_RU: /* toward +Infinity */ |
| 172 |
if (sn == 0) |
| 173 |
return ieee754dp_inf(0); |
| 174 |
else |
| 175 |
return ieee754dp_max(1); |
| 176 |
case FPU_CSR_RD: /* toward -Infinity */ |
| 177 |
if (sn == 0) |
| 178 |
return ieee754dp_max(0); |
| 179 |
else |
| 180 |
return ieee754dp_inf(1); |
| 181 |
} |
| 182 |
} |
| 183 |
/* gen norm/denorm/zero */ |
| 184 |
|
| 185 |
if ((xm & DP_HIDDEN_BIT) == 0) { |
| 186 |
/* we underflow (tiny/zero) */ |
| 187 |
assert(xe == DP_EMIN); |
| 188 |
if (ieee754_csr.mx & IEEE754_UNDERFLOW) |
| 189 |
ieee754_setcx(IEEE754_UNDERFLOW); |
| 190 |
return builddp(sn, DP_EMIN - 1 + DP_EBIAS, xm); |
| 191 |
} else { |
| 192 |
assert((xm >> (DP_FBITS + 1)) == 0); /* no excess */ |
| 193 |
assert(xm & DP_HIDDEN_BIT); |
| 194 |
|
| 195 |
return builddp(sn, xe + DP_EBIAS, xm & ~DP_HIDDEN_BIT); |
| 196 |
} |
| 197 |
} |