cblas-sys VS matrixmultiply

I've switched to just using the AMD BLIS library, and linking through cblas-sys. One day I would like to rewrite a matmul and convolution library with packed-simd-2 or portable-simd when they and const generics are finished.

The matrixmultiply crate from the ndarray author (https://github.com/bluss/matrixmultiply) is one such implementation. It uses the same algorithm as the BLIS project (https://www.cs.utexas.edu/users/flame/pubs/TOMS-BLIS-Analytical.pdf) to partition the problem and exploit the cache hierarchy. It isn't as well tuned as eg. Intel MKL or BLIS, but the results are very respectable.

Do you plan to support integers as native types? I know there is an issue for the crate matrixmultiply for that, it seems it can be problematic because of overflow.

There's a famous crate [matrixmultiply](https://github.com/bluss/matrixmultiply) for matrix-matrix multiplication in Rust. But it's a bit slow for me.

In my benchmarks, Nim is faster than Rust:

- multithreading runtime (i.e Rayon vs Weave https://github.com/mratsim/weave)

- Cryptography: https://hackmd.io/@gnark/eccbench#Pairing

- Scientific computing / matrix multiplication: https://github.com/bluss/matrixmultiply/issues/34#issuecomme...

There is no inherent reason why a Nim program would be slower than Rust.