Math Libraries

The table below is intended to serve as a guide to the capabilities of the various math libraries available on the JPL supercomputers. Different areas of computational mathematics are listed in the first column. The availability of routines for each area within each of the major math libraries is given in the remaining columns. The ratings (none, fair, good, best) refer ONLY to the capabilities, number, and variety of routines available within each catagory and library. These ratings DO NOT indicate the quality, power, efficiency (either in CPU time or memory), ease of use, or quality of documentation of the routines in any area or library.

In general, these routines are single-processor versions. Some libraries will allow some routines to run multi-threaded on a few (2 - 16) threads. The only library intended to run massively-parallel is the Parallel Iterative Methods library.

The libraries listed in the table are:

Cray SciLib:

This is a standard library developed by Cray over many years and is completely available on the Cray SV1. A version is also available on the SGI Origin 2000, called complib.sgimath

IMSL:

A commercial library of Fortran routines which is currently only on the Cray SV1.

LAPACK:

A publically-available library (in source code) from Oak Ridge National Labs/University of Tennessee at Knoxville (http://www.netlib.org) which has become the standard for linear algebra solvers. Each of the manufacturers has created a version of this library optimized for their hardware.

MATH77/mathc90:

A library of over 550 routines in Fortran (and their C equivalents) that was developed here at JPL over the past 30 years. This library is available on the Cray SV1 and the SGI Origin 2000. This library has the strong advantages of very good documentation available on the Web ( http://mathalacarte.com), of portability among all computers at JPL, and that one of its authors (Dr. Fred Krogh) is available locally for consultation to JPL supercomputer users at no charge to the user.

Numerical Algorithms Group:

This commercial library consists of 53 routines on the Cray SV1.

Parallel Iterative Methods:

A publically-available library (in source code) for solving large linear systems using any of 10 different iterative methods.

VODE/CVODE:

A publically-available library (in source code) for solving stiff and non-stiff initial value problems for systems of ordinary differential equations (ODEs).

A Comparison of Math Library Capabilities on the JPL Supercomputers

Related page

MATH77/mathc90 Libraries

Printer Friendly Version