axpy/
axpy.rs

1/// Slightly modify the example on crate `simd` to make it work on stable channel 
2/// Reference link : https://github.com/rust-lang-nursery/simd
3
4extern crate ssimd;
5use ssimd::{f32x4, f32x8};
6
7#[inline(never)]
8pub fn axpy(z: &mut [f32], a: f32, x: &[f32], y: &[f32]) {
9    assert_eq!(x.len(), y.len());
10    assert_eq!(x.len(), z.len());
11
12    let len = std::cmp::min(std::cmp::min(x.len(), y.len()), z.len());
13
14    let mut i = 0;
15    while i < len & !3 {
16        let x = f32x4::load(x, i);
17        let y = f32x4::load(y, i);
18        (f32x4::splat(a) * x + y).store(z, i);
19        i += 4
20    }
21}
22
23#[inline(never)]
24pub fn axpy8(z: &mut [f32], a: f32, x: &[f32], y: &[f32]) {
25    assert_eq!(x.len(), y.len());
26    assert_eq!(x.len(), z.len());
27
28    let len = std::cmp::min(std::cmp::min(x.len(), y.len()), z.len());
29
30    let mut i = 0;
31    while i < len & !7 {
32        let x = f32x8::load(x, i);
33        let y = f32x8::load(y, i);
34        (f32x8::splat(a) * x + y).store(z, i);
35        i += 8
36    }
37}
38
39fn main() {
40    let mut z = vec![0.; 4];
41    axpy(&mut z, 2., &[1.0, 3.0, 5.0, 7.0], &[2.0, 4.0, 6.0, 8.0]);
42    println!("{:?}", z);
43    let mut z = vec![0.; 8];
44    axpy(&mut z, 3., &[1.0, 3.0, 6.0, 7.0, 10.0, 6.0, 3.0, 2.0],
45                       &[2.0, 4.0, 6.0, 8.0, 2.0, 4.0, 6.0, 8.0]);
46    println!("{:?}", z);
47
48    let mut z = vec![0.; 4];
49    axpy8(&mut z, 2., &[1.0, 3.0, 5.0, 7.0], &[2.0, 4.0, 6.0, 8.0]);
50    println!("{:?}", z);
51    let mut z = vec![0.; 8];
52    axpy8(&mut z, 3., &[1.0, 3.0, 6.0, 7.0, 10.0, 6.0, 3.0, 2.0],
53                       &[2.0, 4.0, 6.0, 8.0, 2.0, 4.0, 6.0, 8.0]);
54    println!("{:?}", z);
55}