Numerical Linear Algebra Libraries

Numerical Linear Algebra Libraries for HPC

If you are compiling your own numerical packages for R, Python, Perl, etc. then you will more than likely need to use the BLAS, LAPACK or ATLAS libraries. The purpose and relationships between these librareis are as follows


Basic Linear Algebra Subroutines

BLAS provides basic building blocks for performing basic scalar, vector, and matric operations, e.g. multiplication, addition, subtraction, etc. In particular

Level 1 BLAS perform scalar, vector and vector-vector operations,
Level 2 BLAS perform matrix-vector operations
Level 3 BLAS perform matrix-matrix operations

BLAS are very mature, efficient and portable. They are used in
essentially all high quality numerical code that performs linear algebra
See for documentation.

Default BLAS shared object libraries are at:



Linear Algebra PACKage

LAPACK largely replaced fortran libraries originally developed in the 1970s (LINPACK and EISPACK), and takes advantage of level 2 and level 3 BLAS for efficient operation on modern multi-core architectures that incorporate hierarcharical and shared memory.
See for documentation.

Default LAPACK shared object libraries are at:



Automatically Tuned Linear Algebra Software

The ATLAS libraries consist of the BLAS and a subset of the LAPACK routines.
ATLAS libraries are tuned to particular architectures and compilers in an HPC environment.

Default ATLAS shared object libraries are at:


See for documentation.