ndarray-linalg

Linear algebra package for Rust with ndarray based on external LAPACK implementations.

Examples

See examples directory.

Note: To run examples, you must specify which backend will be used (as described below). For example, you can execute the solve example with the OpenBLAS backend like this:

cargo run --example solve --features=openblas

and run all tests of ndarray-linalg with OpenBLAS

cargo test --features=openblas

BLAS/LAPACK Backend

Three BLAS/LAPACK implementations are supported:

• OpenBLAS
  • needs gfortran (or other Fortran compiler)
• Netlib
  • needs cmake and gfortran
• Intel MKL (non-free license, see the linked page)
  • needs curl

There are three features corresponding to the backend implementations (openblas / netlib / intel-mkl):

[dependencies]
ndarray = "0.12"
ndarray-linalg = { version = "0.10", features = ["openblas"] }

For librarian

If you creating a library depending on this crate, we encourage you not to link any backend:

[dependencies]
ndarray = "0.12"
ndarray-linalg = "0.10"

Link backend crate manually

For the sake of linking flexibility, you can provide LAPACKE implementation (as an extern crate) yourself. You should link a LAPACKE implementation to a final crate (like binary executable or dylib) only, not to a Rust library.

[dependencies]
ndarray = "0.12"
ndarray-linalg = "0.10"
openblas-src = "0.7" # or another backend of your choice

You must add extern crate to your code in this case:

extern crate ndarray;
extern crate ndarray_linalg;
extern crate openblas_src; // or another backend of your choice

Link statically

If you needs to link the backend (OpenBLAS/Netlib) statically, please add following to your Cargo.toml

[dependencies]
openblas-src = { version = "0.7.0", features = ["static"] }

Intel-MKL backend does not support static linking.