4.1 Introduction to LAPACK

LAPACK ([5]) is a library of FORTRAN 77 subroutines for solving commonly occurring problems in numerical linear algebra. LAPACK components can solve systems of linear equations, linear least squares problems, eigenvalue problems and singular value problems. Dense and banded matrices are provided for, but not general sparse matrices. In all areas, similar functionality is provided for real and complex matrices.

LAPACK routines are written so that as much as possible of the computations is performed by calls to the BLAS. The efficiency of LAPACK routines depends, in large part, on the efficiency of the BLAS being called. Block algorithms are employed wherever possible to maximize the use of calls to level 3 BLAS, which generally run faster than lower level BLAS due to the high number of operations per memory access.

The performance of some of the LAPACK routines has been further improved by reworking the computational algorithms. Some of the LAPACK routines contained in ACML are therefore based on code that is different from the LAPACK sources available in the public domain. In all these cases the algorithmic and numerical properties of the original LAPACK routines have been strictly preserved. Furthermore, key LAPACK routines have been treated using OpenMP to take advantage of multiple processors when running on SMP machines. Your application will automatically benefit when you link with the OpenMP versions of ACML.