microgemm 0.3.1

General matrix multiplication with custom configuration in Rust. Supports no_std and no_alloc environments.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

use microgemm::{kernels::GenericKernel8x8, Kernel as _, MatMut, MatRef, PackSizes};

#[test]
fn test_main() {
    let kernel = GenericKernel8x8::<f32>::new();
    assert_eq!(kernel.mr(), 8);
    assert_eq!(kernel.nr(), 8);

    let pack_sizes = PackSizes {
        mc: 5 * kernel.mr(), // MC must be divisible by MR
        kc: 190,
        nc: 9 * kernel.nr(), // NC must be divisible by NR
    };
    let mut packing_buf = vec![0.0; pack_sizes.buf_len()];

    let (alpha, beta) = (2.0, -3.0);
    let (m, k, n) = (100, 380, 250);

    let a = vec![2.0; m * k];
    let b = vec![3.0; k * n];
    let mut c = vec![4.0; m * n];

    let a = MatRef::row_major(m, k, &a);
    let b = MatRef::row_major(k, n, &b);
    let mut c = MatMut::row_major(m, n, &mut c);

    // c <- alpha a b + beta c
    kernel.gemm(alpha, a, b, beta, &mut c, pack_sizes, &mut packing_buf);
    println!("{:?}", c.as_slice());
}