|
|
|
Home > Help Files > Software Applications > LaPack > LaPack Documenation
LAPACK Documentation
LAPACK - Linear Algebra PACKage
LAPACK (Linear Algebra PACKage) is a library of subroutines for
solving common problems in numberical linear algebra. One of the
design goals of LAPACK is to be efficient on a wide range of
high-performance computers. It is public domain and available
free of charge from Oak Ridge National Laboratory via xnetlib
(see "help f77 libraries netlib").
The library includes about 150 "driver" routines which solve
problem in the areas of:
Linear Equations
Linear Least Squares (LLS) Problems
Standard Eigenvalue and Singular Value Problems
Generalized Eigenvalue Problems
These drivers are based on about 400 computational routines in
the areas of:
Linear Equations
Orthogonal Factorizations and Linear Least Squares Problems
Symmetric Eigneproblems
Nonsymmetric Eigenproblems
Singular Value Decomposition
Generalized Symmetric Definite Eigenproblems
The routines have been coded for many different types of matrices
including general, band, tridiagonal, Hermitian, Hessenberg,
orthogonal, symmetric, and complex unitary matrices.
Most procedures are coded in four versions: single precision,
double precision, complex, and complex*16. The latter is not
part of the Fortran standard and may not be available on with
some Fortran compilers.
Whenever possible, LAPACK uses the BLAS (Basic Linear Algebra
Subroutines) to do most of the intensive computation. If possi-
ble, use a vendor-supplied optimized version of the BLAS with
LAPACK for high performance. When this is not possible, use the
unoptimized BLAS from the LAPACK distribution.
LAPACK is designed to replace the EISPACK and LINPACK libraries.
These are quite old; the routines in LAPACK are generally both
faster and use better algorithms than those of the older li-
braries.
The public distribution of LAPACK version 1.0a is installed on
garnet. To use it, enter:
f77 prog.f -llapack -lblas
The LAPACK and BLAS single precision and complex routines are
distributed as part of the default Fortran libraries on the Cray.
To use them, enter:
cf77 prog.f
Due to the long Cray word size, the double precision routines are
not installed. As the default complex data type on the Cray uses
16 bytes per complex value, the special complex*16 routines are
not installed on the Cray.
Under UNICOS 6.1, this is a slightly earlier version of LAPACK.
It has been optimized for the Cray and uses Cray's highly optim-
ized library of BLAS routines. However, there may be a few in-
compatibilities between the Cray's version of LAPACK and the ver-
sion 1.0a distribution. This should be fixed with the installa-
tion of UNICOS 7.0 on the Cray.
REFERENCES
"LAPACK User's Guide", E. Anderson, Z. Bai, C. Bischof, J. Dem-
mel, J. Dongarra, J. Du Croz, A. Greenbaum, S. Hammarling, A.
McKenney, S. Ostrouchov, and D. Sorensen, SIAM (Society for In-
dustrial and Applied Mathematics), 1992.
|
|
© |