#generic #array #numeric #simd

numeric-array

Wrapper around generic-array that adds efficient numeric trait implementations

19 releases

Uses old Rust 2015

0.5.2 Oct 7, 2020
0.5.0 Apr 11, 2020
0.4.2 Nov 29, 2019
0.4.1 Jul 5, 2019
0.1.6 Sep 22, 2017

#334 in Data structures

Download history 11/week @ 2021-08-10 27/week @ 2021-08-17 78/week @ 2021-08-24 107/week @ 2021-08-31 117/week @ 2021-09-07 121/week @ 2021-09-14 57/week @ 2021-09-21 73/week @ 2021-09-28 90/week @ 2021-10-05 77/week @ 2021-10-12 33/week @ 2021-10-19 55/week @ 2021-10-26 124/week @ 2021-11-02 62/week @ 2021-11-09 69/week @ 2021-11-16 20/week @ 2021-11-23

293 downloads per month
Used in dumbnet

MIT license

52KB
1.5K SLoC

numeric-array

numeric-array is a wrapper around generic-array that adds efficient numeric trait implementations, often times making use of autovectorized SIMD instructions and compile-time evaluations (not const-eval, but LLVM optimizations).

All stable core::ops traits are implemented for NumericArray itself, plus the thin NumericConstant type, which is required to differentiate constant values from NumericArray itself.

Additionally, most of num_traits are implemented, including Num itself. So you can even use a whole array as a generic number.

Example:

extern crate num_traits;
#[macro_use]
extern crate generic_array;
#[macro_use]
extern crate numeric_array;

use num_traits::Float;
use numeric_array::NumericArray;

fn main() {
    let a = narr![f32; 1, 2, 3, 4];
    let b = narr![f32; 5, 6, 7, 8];
    let c = narr![f32; 9, 1, 2, 3];

    // Compiles to a single vfmadd213ps instruction on my machine
    let d = a.mul_add(b, c);

    assert_eq!(d, narr![f32; 14, 13, 23, 35]);
}

When used with RUSTFLAGS = "-C opt-level=3 -C target-cpu=native", then Rust and LLVM are smart enough to autovectorize almost all operations into SIMD instructions, or even just evaluate them at compile time. The above example is actually evaluated at compile time, so if you were to view the assembly it would show the result only. Rust is pretty smart.

Therefore, this is ideal for situations where simple component-wise operations are required for arrays.

Dependencies

~270–415KB