Dongarra on LAPACK at Stanford - 5 Feb 91

[David Hough]

There are those who argue that LAPACK design emphasis should be the
current and forthcoming high-performance RISC/Unix desktops that most
scientists will be using rather than the increasingly select minority
depending on high-performance supercomputers; those will probably be
distinguished from the desktops primarily by massive parallelism and
enormous memory and I/O bandwidth.  Anyway the Feds still want to fund
the supercomputers, so:

*DAY     Tuesday, Feb 5, 1991
*EVENT   Computer Science Colloquium
*PLACE   Jordan 040
*TIME    4:15 pm
*PERSON  Jack Dongarra
*FROM    University of Tennessee and Oak Ridge National Laboratory
*TITLE   LAPACK: A Linear Algebra Library for High-Performance Computers


*** IMMEDIATELY BEFORE THE COLLOQUIUM, REFRESHMENTS WILL BE SERVED ***
                 
              IN THE THIRD FLOOR LOUNGE.  ALL ARE WELCOME!


*ABSTRACT

This talk outlines the proposed computational package called LAPACK.
LAPACK is planned to be a collection of Fortran 77 subroutines for the
analysis and solution of various systems of simultaneous linear
algebraic equations, linear least-squares problems, and matrix
eigenvalue problems.

The library is intended to provide a uniform set of subroutines to
solve the most common linear algebra problems and to run efficiently
on a wide range of architectures.  This library, which will be freely
accessible via computer network, will:

1) Facilitates the development of scientific codes on supercomputers.
   The availability of a highly efficient library for basic
   linear algebra problems on each major machine would free the programmer
   to work on more interesting parts of the code.
2) Increases the portability of scientific codes between different
   supercomputers. 
3) Provide access to state-of-the-art algorithms and software.
4) Provides tools to aid supercomputer performance evaluation.

The library will be based on the well-known and
widely used LINPACK and EISPACK packages for linear equation solving,
eigenvalue problems, and linear least squares.  
The target machines are high performance architectures with one or more
processors, usually with a vector-processing facility.

This talk will review the package, describe some of the new algorithms,
and discussion aspects of software design.