matrixmultiply_mt 0.2.1

Multithreaded Fork of bluss's matrixmultiply crate. General matrix multiplication of f32 and f64 matrices in Rust. Supports matrices with general strides. Uses a microkernel strategy, so that the implementation is easy to parallelize and optimize. `RUSTFLAGS="-C target-cpu=native"` is your friend here.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22


test mat_mul_f32::m004             ... bench:         142 ns/iter (+/- 3)
test mat_mul_f32::m005             ... bench:         200 ns/iter (+/- 2)
test mat_mul_f32::m006             ... bench:         215 ns/iter (+/- 5)
test mat_mul_f32::m007             ... bench:         242 ns/iter (+/- 10)
test mat_mul_f32::m008             ... bench:         251 ns/iter (+/- 15)
test mat_mul_f32::m009             ... bench:         457 ns/iter (+/- 8)
test mat_mul_f32::m012             ... bench:         606 ns/iter (+/- 7)
test mat_mul_f32::m016             ... bench:         910 ns/iter (+/- 20)
test mat_mul_f32::m032             ... bench:       4,595 ns/iter (+/- 280)
test mat_mul_f32::m064             ... bench:      28,104 ns/iter (+/- 530)
test mat_mul_f32::m127             ... bench:     189,393 ns/iter (+/- 4,303)
test mat_mul_f32::mix16x4          ... bench:       1,717 ns/iter (+/- 64)
test mat_mul_f32::mix32x2          ... bench:       1,462 ns/iter (+/- 29)
test mat_mul_f64::m004             ... bench:         145 ns/iter (+/- 17)
test mat_mul_f64::m007             ... bench:         257 ns/iter (+/- 6)
test mat_mul_f64::m008             ... bench:         276 ns/iter (+/- 11)
test mat_mul_f64::m012             ... bench:         678 ns/iter (+/- 22)
test mat_mul_f64::m016             ... bench:       1,065 ns/iter (+/- 19)
test mat_mul_f64::m032             ... bench:       6,024 ns/iter (+/- 1,709)
test mat_mul_f64::m064             ... bench:      39,642 ns/iter (+/- 6,456)
test mat_mul_f64::m127             ... bench:     278,104 ns/iter (+/- 8,016)