Difference between revisions of "The BLAS and LAPACK libraries"

Latest revision as of 09:59, 6 October 2026

The BLAS (Basic Linaer Algebra) and LAPACK (Linear Algebra PACKage) libraries are very common libraries useful for doing linear algebra operations involving matrices and vectors, .e.g. solving linear equation systems, doing SVD factorization, etc. In many case they are already installed and available Linux systems, but not always... and then you will have to download and install them yourself.

Using package managers etc.

In many cases, obtaining and installing the BLAS and LAPACK libraries can be done via the system's package manager; e.g. under Ubuntu-Linux with:

sudo apt-get install libblas-dev liblapack-dev

If you don't have root-like (sudo) privileges, or your system manager doesn't help, then you can always install the libraries from scratch as detailed below.

Obtaining the source code of the BLAS and LAPACK libraries

The Fortran source codes are available on Netlib.org : http://www.netlib.org/blas/ and http://www.netlib.org/lapack/

Compiling the libraries

Prerequisites

All you need are a Fortran compiler and the gmake utility

Compiling the BLAS library

Assuming you dowloaded the package in a "BLAS" directory, e.g.:

wget http://www.netlib.org/blas/blas-3.11.0.tgz
tar xvzf blas-3.11.0.tgz
cd BLAS-3.11.0/

Edit the make.inc to fit your settings (e.g. compiler name and options), if needed and the simply run

make

which should produce the static library blas_LINUX.a. It is good practice to usually use a different naming convention and call the library libblas.a so you can make a symbolic link to have both available:

ln -s blas_LINUX.a libblas.a

Compiling the LAPACK library

Assuming you dowloaded the package in a "LAPACK" directory, e.g.:

wget https://github.com/Reference-LAPACK/lapack/archive/refs/tags/v3.11.0.tar.gz
tar xvzf v3.11.0.tar.gz
cd lapack-3.11.0

copy the provided make.inc.example as make.inc and adapt it to fit your settings (e.g. compiler name and options), if needed, and then run:

make

which should produce the static library liblapack.a. Note that recent versions of LAPACK actually incorporate the BLAS library but with a different name: librefblas.a.

Linking to the libraries

One simply needs to specify options

-Lpath/to/the/LAPACK/library -llapack -Lpath/to/the/BLAS/library -lblas

at compile time. Note the ordering: as some LAPACK routines rely on BLAS routines one must specify the LAPACK library first

A simple example to test your Lapack installation

If you want to double check that all is well, you can for instance test with this small Fortran program:

program test_lapack_dgesv

! test solving linear system with Lapack's DGESV
! compile with:
! gfortran test_lapack_dgesv.F90 -o test_lapack_dgesv -llapack
implicit none

! Solve system A*X=B
integer,parameter :: ssize=2
double precision :: a(ssize,ssize) 
double precision :: b(ssize)
double precision :: pivot(ssize)
integer  :: rc       ! Return code.

! build system
a = reshape([ 2., 3., 1., 1. ], [ 2, 2 ])
b = [ 5., 6. ]

write(*,*) "a(1,1)=",a(1,1)," a(1,2)=",a(1,2)
write(*,*) "a(2,1)=",a(2,1)," a(2,2)=",a(2,2)
write(*,*) ""
write(*,*) "b(1)=",b(1)," b(2)=",b(2)

call dgesv(ssize, 1, a, ssize, pivot, b, ssize, rc)
! nb: b is overwritten with solution
if (rc==0) then
  write(*,*) "call to dgesv successfull!"
endif
write(*,*) "x(1)=",b(1)," x(2)=",b(2)

end program test_lapack_dgesv

which, when run, should give the following output:

 a(1,1)=   2.0000000000000000       a(1,2)=   1.0000000000000000     
 a(2,1)=   3.0000000000000000       a(2,2)=   1.0000000000000000     
 
 b(1)=   5.0000000000000000       b(2)=   6.0000000000000000     
 call to dgesv successfull!
 x(1)=   1.0000000000000002       x(2)=   2.9999999999999996