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.