Crates.io | softposit |
lib.rs | softposit |
version | 0.4.0 |
source | src |
created_at | 2019-05-11 13:19:40.130963 |
updated_at | 2022-12-14 10:03:21.505013 |
description | Implementation of Posit numbers |
homepage | |
repository | https://gitlab.com/burrbull/softposit-rs |
max_upload_size | |
id | 133450 |
size | 502,171 |
softposit-rs
Rust port of SoftPosit repository.
Such types are supported now:
Initial support of generic types PxE1{N} and PxE2{N}. Requires rust 1.51.
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);
}
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 functions, like trigonomentic, are partially implemented for P32E2 type.
For complex computations use num::Complex
type.
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"
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 |