softposit

Crates.iosoftposit
lib.rssoftposit
version0.4.0
sourcesrc
created_at2019-05-11 13:19:40.130963
updated_at2022-12-14 10:03:21.505013
descriptionImplementation of Posit numbers
homepage
repositoryhttps://gitlab.com/burrbull/softposit-rs
max_upload_size
id133450
size502,171
Zgarbul Andrey (burrbull)

documentation

README

crates.io crates.io

softposit-rs

Rust port of SoftPosit repository.

Such types are supported now:

  • P32E2 or P32: 32-bit posit with exponent size ES=2;
  • P16E1 or P16: 16-bit posit with ES=1;
  • P8E0 or P8: 8-bit posit without exponent bits.

Initial support of generic types PxE1{N} and PxE2{N}. Requires rust 1.51.

Examples

Convert from f64, Add

use softposit::P32;

fn main() {
    let a = P32::from(12.3);
    let b = P32::from(154.);
    let c = a + b;
    println!("c = {0} or c = {0:?}", c);
}

Quire

Each Posit type has correspondent Quire type (Q32, Q16, Q8). These types support AddAssign and SubAssign operations. For example:

use softposit::{P32, Q32};

// c == 12.3*0.4 - 6.3*8.4 == -48;
fn main() {
    let mut q = Q32::init();
    q += (P32::from(12.3), P32::from(0.4));
    q -= (P32::from(6.3), P32::from(8.4));
    let c = q.to_posit();
    println!("c = {0} or c = {0:?}", c);
}

Math

Math functions, like trigonomentic, are partially implemented for P32E2 type. For complex computations use num::Complex type.

Linear algebra

You can also use matrix operations, defined in nalgebra crate by using linalg feature:

use softposit::P32;

use nalgebra::{Dynamic, Matrix, VecStorage};
type DMatrix = Matrix<P32, Dynamic, Dynamic, VecStorage<P32, Dynamic, Dynamic>>;

fn main() {
    let a = DMatrix::new_random(5, 5);
    println!("Matrix A = {}", a);
    let decomp = &a.clone().lu();
    if let Some(c) = decomp.try_inverse() {
        println!("Inverse matrix Aˉ¹ = {}", c);
        println!("Check Identity");
        let ones = a * c;
        println!("A × Aˉ¹ = {}", ones);
    } else {
        println!("Matrix A can't be inverted");
    }
}
cargo build --features="linalg,rand"

Benchmarking

cargo bench
Type Operation i3-2310M @ 2.10GHz i5-3470 @ 3.20GHz
P32E2 + 47 MPOPS 82 MPOPS
- 45 MPOPS 83 MPOPS
* 56 MPOPS 99 MPOPS
/ 25 MPOPS 42 MPOPS
√⎺⎺ 48 MPOPS 92 MPOPS
round 89 MPOPS 159 MPOPS
Commit count: 139

cargo fmt