1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230
// Copyright © 2018–2024 Trevor Spiteri
// This library is free software: you can redistribute it and/or
// modify it under the terms of either
//
// * the Apache License, Version 2.0 or
// * the MIT License
//
// at your option.
//
// You should have recieved copies of the Apache License and the MIT
// License along with the library. If not, see
// <https://www.apache.org/licenses/LICENSE-2.0> and
// <https://opensource.org/licenses/MIT>.
#![allow(deprecated)]
macro_rules! make_helper {
($Float:ident($Bits:ty, $IBits:ident, $prec:expr) $(; use $path:path)?) => {
#[allow(non_snake_case)]
pub mod $Float {
use crate::{
helpers::{FloatKind, ToFixedHelper, ToFloatHelper, Widest},
int_helper,
};
use core::cmp::Ordering;
$(use $path;)?
const PREC: u32 = $prec;
const EXP_BIAS: i32 = (1 << (<$Bits>::BITS - PREC - 1)) - 1;
const EXP_MIN: i32 = 1 - EXP_BIAS;
const EXP_MAX: i32 = EXP_BIAS;
pub const SIGN_MASK: $Bits = 1 << (<$Bits>::BITS - 1);
pub const EXP_MASK: $Bits = !(SIGN_MASK | MANT_MASK);
const MANT_MASK: $Bits = (1 << (PREC - 1)) - 1;
// zero is NOT negative, that is zero is represented as
// Kind::Finite { neg: false, abs: 0, frac_bits: 0 },
pub enum Kind {
NaN,
Infinite { neg: bool },
Finite { neg: bool, abs: $Bits, frac_bits: i32 },
}
#[inline]
pub fn kind(val: $Float) -> Kind {
let (neg, mut exp, mut mantissa) = parts(val);
if exp > EXP_MAX {
return if mantissa == 0 {
Kind::Infinite { neg }
} else {
Kind::NaN
};
}
// if not subnormal, add implicit bit
if exp >= EXP_MIN {
mantissa |= 1 << (PREC - 1);
} else {
exp = EXP_MIN;
}
if mantissa == 0 {
return Kind::Finite {
neg: false,
abs: 0,
frac_bits: 0,
};
}
Kind::Finite {
neg,
abs: mantissa,
frac_bits: PREC as i32 - 1 - exp,
}
}
#[inline]
fn parts(val: $Float) -> (bool, i32, $Bits) {
let bits = val.to_bits();
let neg = bits & SIGN_MASK != 0;
let biased_exp = (bits & EXP_MASK) >> (PREC - 1);
let exp = biased_exp as i32 - EXP_BIAS;
let mant = bits & MANT_MASK;
(neg, exp, mant)
}
#[inline]
pub fn from_to_float_helper(
val: ToFloatHelper,
frac_bits: u32,
int_bits: u32,
) -> $Float {
let fix_bits = frac_bits + int_bits;
let bits_sign = if val.neg { SIGN_MASK } else { 0 };
let extra_zeros = 128 - fix_bits;
let leading_zeros = val.abs.leading_zeros() - extra_zeros;
let signif_bits = fix_bits - leading_zeros;
if signif_bits == 0 {
return $Float::from_bits(bits_sign);
}
// remove leading zeros and implicit one
let mut mantissa = val.abs << leading_zeros << 1;
let exponent = int_bits as i32 - 1 - leading_zeros as i32;
let biased_exponent = if exponent > EXP_MAX {
return $Float::from_bits(EXP_MASK | bits_sign);
} else if exponent < EXP_MIN {
let lost_prec = EXP_MIN - exponent;
if lost_prec as u32 >= (int_bits + frac_bits) {
mantissa = 0;
} else {
// reinsert implicit one
mantissa = (mantissa >> 1) | !(!0 >> 1);
mantissa >>= lost_prec - 1;
}
0
} else {
(exponent + EXP_MAX) as $Bits
};
// check for rounding
let round_up = (fix_bits >= PREC) && {
let shift = PREC - 1;
let mid_bit = !(!0 >> 1) >> (shift + extra_zeros);
let lower_bits = mid_bit - 1;
if mantissa & mid_bit == 0 {
false
} else if mantissa & lower_bits != 0 {
true
} else {
// round to even
mantissa & (mid_bit << 1) != 0
}
};
let bits_exp = biased_exponent << (PREC - 1);
let bits_mantissa = (if fix_bits >= PREC - 1 {
(mantissa >> (fix_bits - (PREC - 1))) as $Bits
} else {
(mantissa as $Bits) << (PREC - 1 - fix_bits)
}) & !(!0 << (PREC - 1));
let mut bits_exp_mantissa = bits_exp | bits_mantissa;
if round_up {
bits_exp_mantissa += 1;
}
$Float::from_bits(bits_sign | bits_exp_mantissa)
}
#[inline]
pub fn to_float_kind(val: $Float, dst_frac_bits: u32, dst_int_bits: u32) -> FloatKind {
let prec = PREC as i32;
let (neg, mut exp, mut mantissa) = parts(val);
if exp > EXP_MAX {
return if mantissa == 0 {
FloatKind::Infinite { neg }
} else {
FloatKind::NaN
};
}
// if not subnormal add implicit bit, otherwise update exp
if exp >= EXP_MIN {
mantissa |= 1 << (prec - 1);
} else {
exp = EXP_MIN;
}
if mantissa == 0 {
let conv = ToFixedHelper {
bits: Widest::Unsigned(0),
dir: Ordering::Equal,
overflow: false,
};
return FloatKind::Finite { neg, conv };
}
let mut src_frac_bits = prec - 1 - exp;
let need_to_shr = src_frac_bits - dst_frac_bits as i32;
if need_to_shr > prec {
let dir = if neg {
Ordering::Greater
} else {
Ordering::Less
};
let conv = ToFixedHelper {
bits: Widest::Unsigned(0),
dir,
overflow: false,
};
return FloatKind::Finite { neg, conv };
}
let mut dir = Ordering::Equal;
if need_to_shr > 0 {
let removed_bits = mantissa & !(!0 << need_to_shr);
let will_be_lsb = 1 << need_to_shr;
let tie = will_be_lsb >> 1;
if removed_bits == 0 {
// removed nothing
} else if removed_bits < tie {
dir = Ordering::Less;
} else if removed_bits > tie || mantissa & will_be_lsb != 0 {
mantissa += will_be_lsb;
dir = Ordering::Greater;
} else {
dir = Ordering::Less;
};
mantissa >>= need_to_shr;
src_frac_bits -= need_to_shr;
}
let mut mantissa = mantissa as $IBits;
if neg {
mantissa = -mantissa;
dir = dir.reverse();
}
let mut conv = int_helper::$IBits::to_fixed_helper(
mantissa,
src_frac_bits,
dst_frac_bits,
dst_int_bits,
);
conv.dir = dir;
FloatKind::Finite { neg, conv }
}
}
};
}
make_helper! { f16(u16, i16, 11); use half::f16 }
make_helper! { bf16(u16, i16, 8); use half::bf16 }
make_helper! { f32(u32, i32, 24) }
make_helper! { f64(u64, i64, 53) }
make_helper! { F128(u128, i128, 113); use crate::F128 }
make_helper! { F128Bits(u128, i128, 113); use crate::F128Bits }