SL_Library
The
PLAPACK
project investigates interfaces and implementations
of dense and sparse parallel linear algebra libraries.
This project is unique in that it concentrates on convenient
interfaces. It is our conclusion that this can only be
achieved by distributing the vectors in a linear
system first, which then induces the matrix distribution.
Justification for this lies in the fact that the physical problem
leads to a mathematical model, involving two spaces and an operator.
It is these spaces that represent the problem. Discretization
of the mathematical model leads to vectors, which hold the
solution and right hand side,
and the matrix, which is the discretized operator.
Physically Based Matrix Distribution is based on these
ideas. The interface that allows information from the
physical problem and/or mathematical model to be passed down
to the library is called the Physically Distributed Matrix Interface .
The SL_library is a prototype implementation for PLAPACK.
Here SL stands for Simple Library.
It was created in part
to investigate techniques required for the PLAPACK library,
and in part to provide an infrastructure for the course
"Parallel Techniques for Numerical Algorithms"
taught by
Robert van de Geijn at
the University of Texas at Austin.
SL_library: reference manual
John Gunnels now maintains this manual
SL_library: Users' Guide
In preparation. Notice that we are changing a number of
calling sequences. Due to the inconsistency between the currently
available library and this guide, which reflects the next release,
we have temporarily disabled this button. Check the above
reference manual instead. Feel free to contact
us at plapackers@cs.utexas.edu if you need access.
Related reading
Manuscripts in progress
-
Almadena Chtchelkanova,
Carter Edwards,
John Gunnels,
Robert van de Geijn,
"A Comprehensive Approach to Parallel Linear
Algebra Libraries," Draft Oct. 9, 1995.
Currently 75 pages.
Expect this document to be changed regularly.
How to read the above paper:
Go straight to Chapter 9, skipping everything else, and look
at the left-looking Cholesky factorization example.
Hopefully this will peak your interest enough
to read the rest of the document.
PLAPACK Working Notes
- C. Edwards, P. Geng, A. Patra, and R. van de Geijn,
"Parallel Matrix
Decompositions: have we been doing it all wrong?"
PLAPACK Working Note #1,
TR-95-39, Department of Computer Sciences, University of
Texas, Oct. 1995.
-
Robert van de Geijn and Jerrell Watts
"SUMMA: Scalable Universal Matrix Multiplication
Algorithm," PLAPACK Working Note #2,
Department of Computer Sciences, The Unversity of
Texas, TR-95-13, April 1995.
Also: LAPACK Working Note #96 , May 1995.
Journal version: Concurrency: Practice and Experience, to appear.
-
Brian Grayson and Robert van de Geijn
"A High Performance Parallel
Strassen Implementation," PLAPACK Working Note #3,
Parallel Processing Letters, to appear.
-
Almadena Chtchelkanova, John Gunnels, Greg Morrow,
James Overfelt,
Robert A. van de Geijn,
"Parallel Implementation of BLAS:
General Techniques for Level 3 BLAS," PLAPACK Working Note #4,
TR-95-40, Department of Computer Sciences, University of
Texas, Oct. 1995.
Journal version: Concurrency: Practice and Experience, to appear.
-
Almadena Chtchelkanova,
Carter Edwards,
John Gunnels,
Greg Morrow,
James Overfelt,
Robert van de Geijn,
"Towards Usable and Lean Parallel Linear Algebra Libraries," PLAPACK Working Note #5,
TR-96-09, Department of Computer Sciences, University of Texas,
May 1996.
Conference version: Supercomputing 96, submitted.
-
Overhead foils used at the
BLAS workshop in Tennessee
- Almadena Chtchelkanova, Carter Edwards, John Gunnels, and Robert van de Geijn,
A Comprehensive Approach to Parallel Linear Algebra Libraries, presented by Robert van de Geijn.
- Carter Edwards and Robert van de Geijn, Physically Distributed Matrix Interface presentated by Carter Edwards.
- Almadena Chtchelkanova, John Gunnels, Greg Morrow, James Overfelt,
Robert A. van de Geijn, Parallel Implementation of BLAS: General
Techniques for Level 3 BLAS,
presentated by Almadena Chtchelkanova.
Other Overhead foils
Sponsors
-
This project is being conducted as part of the
PRISM
(Parallel Research on Invariant Subspace Methods)
project, under ARPA grant P-95006.
-
It is also sponsored by
the NASA High Performance Computing and Communications
Program's Earth and Space Sciences Project under NRA Grant NAG5-2497.