LAPACK

From CBLFS
Revision as of 20:06, 21 December 2006 by Weibullguy (talk | contribs) (Added the instructions.)
Jump to navigationJump to search
Download Source: http://www.netlib.org/lapack/lapack-3.1.1.tgz

Introduction to LAPACK

LAPACK is a Linear Algebra PACKage that provides routines for solving systems of simultaneous linear equations, least-squares solutions of linear systems of equations, eigenvalue problems, and singular value problems. The associated matrix factorizations (LU, Cholesky, QR, SVD, Schur, generalized Schur) are also provided, as are related computations such as reordering of the Schur factorizations and estimating condition numbers. Dense and banded matrices are handled, but not general sparse matrices. In all areas, similar functionality is provided for real and complex matrices, in both single and double precision. LAPACK routines are written so that as much as possible of the computation is performed by calls to the BLAS.

LAPACK libraries can also be obtained from ATLAS or the microprocessor vendor's math library, such as AMD's ACML.

Dependencies

Required

One of the following is required to build LAPACK:

GCC-3.3 built with F77 support GCC built with gfortran support

Optional

Either BLAS or ACML are recommended otherwise LAPACK will install an unoptimized version of this library.

Configuration Information

There is no configure script for LAPACK. All options are set in the make.inc file. You may wish to modify the following options.

PLAT= appends the architecture to the library name. Beneficial if building LAPACK for more than one architecture.
OPTS= sets the options passed to the FORTRAN compiler.

Non-Multilib or Multilib

First we create a make.inc to use in our build.

cp -v make.inc.example make.inc

If you haven't built the F77 compiler provided by gcc-3.3, you need to tell LAPACK that you're using gfortran.

sed -i 's@g77@gfortran@g' make.inc

If you do not have a system BLAS library, tell LAPACK to build its own, unoptimized library:

sed -i 's@lib: lapacklib tmglib@lib: blaslib blas_testing lapacklib tmglib lapack_testing@' Makefile

If you installed the BLAS library provided by Atlas, tell LAPACK to use this library. Replace libf77blas.a with the name of your system BLAS FORTRAN library.

sed -i 's@../../blas$(PLAT).a@/usr/lib/libf77blas.a@' make.inc

If you installed the BLAS library provided by ACML, tell LAPACK to use this library. Replace </path/to/libacml.a> in the command with the actual path to your ACML library.

sed -i 's@../../blas$(PLAT).a@</path/to/libacml.a>@' make.inc

NOTE: If you want to build and execute the LAPACK test suite, it will fail if you're using the ATLAS BLAS library. From the LAPACK release notes, "The LAPACK test suite needs to be modified to allow for the inversion of diagonal elements, as ATLAS's implementation of xTRSM is valid." There is currently no patch available so you will need to edit the code manually or skip the test suite.

If not using the ATLAS BLAS library, tell LAPACK to build and execute the test suite.

sed -i 's@lib: lapacklib tmglib@lib: lapacklib tmglib lapack_testing@' Makefile

Compile the LAPACK libraries:

make lib

Move the libraries to a standard location:

cp -v lapack_LINUX.a /usr/lib/liblapack.a &&
cp -v tmglib_LINUX.a /usr/src/libtmg.a

Install the documentation that comes with the LAPACK library:

install -v -m755 -d /usr/share/doc/LAPACK-3.1.1/html &&
install -v INSTALL/lawn81.{pdf,ps} /usr/share/doc/LAPACK-3.1.1/ &&
for doc in html/*.html; do
  cp -v $doc /usr/share/doc/LAPACK-3.1.1/html
done

Contents

Installed Libraries: liblapack.a, libtmg.a

Short Description

liblapack.a Library containing the linear algebra functions.
libtmg.a Library containg the test matrix generators used in the LAPACK test suite.
Retrieved from "?title=LAPACK&oldid=6045"