faer
is a collection of crates that implement a low level API for linear algebra routines. It's somewhat similar to BLAS/Lapack, but gives more control to users by allowing parallelism to be specified on a per-call basis. it also exposes some of the low level knobs that users can tweak to maximize performance on their hardware, although the default settings should also give good performance on their own.
Benchmarks can be found on the official website, and code examples can be found there, as well as on the docs.rs pages for each crate