Crate num_bigfloat

Expand description

Multiple precision floating point numbers implemented purely in Rust. Number has fixed-size mantissa and exponent, but increased precision compared to f32 or f64 values.

Number characteristics:

Name	Value
Decimal positions in mantissa	40
Exponent minimum value	-128
Exponent maximum value	127

Examples

use num_bigfloat::BigFloat;
use num_bigfloat::ONE;
use num_bigfloat::PI;
 
// compute pi: pi = 6*arctan(1/sqrt(3))
let six: BigFloat = 6.0.into(); // note: conversion from f64,f32 are not loss-less.
let three: BigFloat = BigFloat::parse("3.0").unwrap();
let pi = six * (ONE / three.sqrt()).atan();
let epsilon = 1.0e-38.into();
 
assert!((pi - PI).abs() < epsilon);
 
println!("{}", pi);
// output: 3.141592653589793238462643383279502884196e-39

no_std example:

use num_bigfloat::BigFloat;
use num_bigfloat::ONE;
use num_bigfloat::PI;
 
// compute pi: pi = 6*arctan(1/sqrt(3))
let six: BigFloat = BigFloat::from_u8(6);
let three: BigFloat = BigFloat::from_u8(3);
let pi = six.mul(&ONE.div(&three.sqrt()).atan());
let epsilon = BigFloat::from_f64(1.0e-38);
 
assert!(pi.sub(&PI).abs().cmp(&epsilon).unwrap() < 0);

The use of additional digits of the manitissa in calculations allows minimizing the error and rounding the results correctly (i.e. as if the result was a rounded infinitely precise number). The precision of the results may be lost under certain conditions, such as when the argument is a subnormal number, or when the function is periodic, such as sine or cosine, when the argument is much larger than pi.