bend-n / umath

rust ffast-math

Home Page:https://docs.rs/umath

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

umath: ffast-math, for rust.

MSRV DOCS

Want to make your math faster? *t&c apply

Want to order a float?

You can do all of that, with umath!

use umath::FFloat;
// wrap a non NAN and non INF f32/f64 (we will also *never* make this number nan).
let mut f = unsafe { FFloat::new(4.0f32) };
f *= 3; // multiply by 3
// this check will be removed by the optimizer!
assert!(!f.is_nan());
# use std::collections::BinaryHeap;
// use a ORD type! this is allowed, as FFloat is not allowed to be NAN | INF.
let mut b = BinaryHeap::new();
b.push(unsafe { FFloat::new(2.0) });
b.push(unsafe { FFloat::new(1.0) });
b.push(unsafe { FFloat::new(3.0) });
b.push(f);
assert_eq!(b.pop(), Some(unsafe { FFloat::new(24.0) }));

A note on safety

When you make your first FFLoat, you must promise that you will never create a NAN | INF FFLoat. Hence, *f = NAN is (delayed) UB.

Nightlyness

umath is nightly because it makes use of core intrinsics, like fadd_fast(), which require the core_intrinsics feature to use.

About

rust ffast-math

https://docs.rs/umath

License:MIT License


Languages

Language:Rust 100.0%