Support for floating point

[dtolnay]

Can the same triple and quad digit lookup table technique be applied to f32 and f64? I would love to have something faster than dtoa.

[mmstick]

It should be able to work perfectly with floating point numbers too, but I don't know anything about how floating points are decoded. You would just need to get an integer representation of the whole number and the fractional number.

[dtolnay]

There are some macros in the dtoa crate to get you started.

#![allow(dead_code)]

#[macro_use] mod diyfp;

use std::{mem, ops};

diyfp! {
    floating_type: f64,
    significand_type: u64,
    exponent_type: isize,

    diy_significand_size: 64,
    significand_size: 52,
    exponent_bias: 0x3FF,
    mask_type: u64,
    exponent_mask: 0x7FF0000000000000,
    significand_mask: 0x000FFFFFFFFFFFFF,
    hidden_bit: 0x0010000000000000,
    cached_powers_f: [0; 0],
    cached_powers_e: [0; 0],
    min_power: (-348),
}

fn main() {
    let x = 2.7182818284f64;
    let fp = unsafe { diyfp::DiyFp::from(x) };
    println!("{:?}", fp);
}

The output is DiyFp { f: 6121026514735115, e: -51 }. That means the original number is f * 2^e.

$ python -c 'print 6121026514735115 * 2**-51'
2.7182818284