BLAS usage examples

The documents and presentations of the tutorial can be found in the doc directory.

Compilation

To compile these examples, you will need access to a BLAS library. Netlib BLAS is the reference implementation of BLAS and is distributed together with the reference implementation of LAPACK. Clone the official repository of Netlib LAPACK. You can the checkout the latest version (v3.12.1), build, and install BLAS and LAPACK.

Building BLAS and LAPACK

BLAS and LAPACK are built with a single configuration. Configure the compilation with the following command:

cmake -S path/to/lapack -B /tmp/build_lapack -G Ninja -DBUILD_SHARED_LIBS:BOOL=ON -DCBLAS:BOOL=ON -DCMAKE_BUILD_TYPE:STRING=Release -DCMAKE_INSTALL_PREFIX:PATH=${HOME}/.local/opt/netlib

This configures the software to be built in the local volume of the node and to install the resulting libraries in your home directory. Build BLAS and LAPACK with,

cmake --build /tmp/build_lapack --target all

and install with:

cmake --build /tmp/build_lapack --target install

Building the matrix garden I/O library

This tutorial is also using an auxiliary library for reading and writing matrices in files in the matrix garden formats. Download the Matrix Market Exchange Formats repository and checkout the v1.0 tag. The library can be configured with:

cmake -S . --preset default-config

Build the software with,

cmake --build build/Release --target all

and install with:

cmake --build build/Release --target install

The library is installed in ${HOME}/.local/opt/matrix-market by default.

Collection of matrix examples

The Matrix Market repository of test data for use in performance analysis of algorithms for numerical linear algebra. It contains multiple collections of test data, such as the Harwell-Boeing and SPARSKIT collections. The data are provided in multiple text file formats, including Harwell-Boeing Exchange Format and the Matrix Market Exchange Format. We are using the Matrix Market Exchange Format due to its simplicity. More information regarding the format can be found in its official documentation.

In this example the matrices user are:

• Harwell-Boeing/bcsstruc1/bcsstk01: BCS Structural Engineering Matrices (eigenvalue matrices)
  • Type: real symmetric positive definite
  • Size: 48 x 48, 224 entries