from na news - ScaLapack

[David G. Hough at validgh]

From: Jack Dongarra <dongarraacs.utk.edu>
Date: Fri, 28 Jan 94 12:49:55 -0500
Subject: Availability of ScaLapack

As part of the ScaLapack project, several new software items are now
available on netlib.  Be aware that these are preliminary version
of the package.  Major changes will occur over time.  The new items
that have been introduced are:  
 
1) Distributed memory version of the core routines from LAPACK
2) Fully parallel package to solve a symmetric positive definite sparse linear
   system on a message passing multiprocessor using Cholesky factorization.
3) A package based on Arnoldi's method for solving large scale nonsymmetric,
   symmetric, and generalized algebraic eigenvalue problems. 
4) C version of LAPACK
5) LAPACK++ a C++ implementation of some of the LAPACK.  
6) Templates for sparse iterative methods for non-symmetric Ax=b.
 
For more information on the availability of each of these
packages, consult the scalapack, clapack, c++, or linalg indexes on netlib.

echo "send index from scalapack" | mail netlibaornl.gov
echo "send index from clapack" | mail netlibaornl.gov
echo "send index from c++/lapack++" | mail netlibaornl.gov
echo "send index from linalg" | mail netlibaornl.gov


1) Distributed memory version of the core routines from LAPACK

   Beta version 1.0 of this part of the package includes factorization
   and solve routines for LU, QR, and Cholesky; decomposition routines to
   Hessenberg form, tridiagonal form, and bidiagonal form; and,
   preliminary versions of QR with column pivoting, triangular inversion,
   and a parallel implementation of the SIGN function, which uses
   deflation to calculate eigenvalues.  Condition estimation and iterative
   refinement routines are also provided for LU and Cholesky.  The current
   version of ScaLapack is in double precision real.  Future releases of
   ScaLapack will include complex versions of routines as well as the
   single precision equivalents.  At the present time, ScaLapack has been
   ported to the Intel Gamma, Delta, and Paragon, Thinking Machines CM-5,
   and PVM clusters.  We are in the process of porting the BLACS to the
   IBM SP-1.

   A second release of PUMMA (Parallel Universal Matrix Multiply Algorithm)
   is included with the ScaLapack software.  Both a PICL implementation and
   a BLACS implementation of PUMMA are provided.
   

2) Fully parallel package to solve a symmetric positive definite sparse linear
--More--
   system on a message passing multiprocessor using Cholesky factorization.
 
   CAPSS (CArtesian Parallel Sparse Solver) is a fully parallel package to
   solve a symmetric positive definite sparse linear system on a message
   passing multiprocessor using Cholesky factorization.  All phases of the
   computation, from ordering through numerical solution, are performed in
   parallel.  The ordering uses Cartesian nested dissection based on an
   embedding of the problem in Euclidean space.  This first release is
   meant for Intel iPSC/860 machines; the code has been compiled and
   tested on an Intel iPSC/860 with 128 processors.  The code is written
   in C with message passing extensions provided by PICL (Portable
   Instrumented Communications Library), which is also available from
   netlib.  CAPSS also uses a few native iPSC/860 functions.
 

3) A package based on Arnoldi's method for solving large scale nonsymmetric,
   symmetric, and generalized algebraic eigenvalue problems. 
 
   ARPACK is a Fortran 77 software package for solving large scale
   eigenvalue problems.   The package is designed to compute a few 
   eigenvalues and corresponding eigenvectors of a large (sparse) matrix.   
   The package provides a communication interface(RCI) to user applications. 
   RCI allows maximal flexibility with respect to user needs and allows 
   (and requires) a user to define its own matrix-vector multiply 
   and/or linear solver routines for the ARPACK supported modes
   (simple REGULAR, simple SHIFT-AND-INVERT, generalized REGULAR, 
   generalized SHIFT-AND-INVERT and CAYLEY mode are supported). 
   A symmetric ARPACK Intel Touchstone Delta parallel implementation 
   is also available on netlib (see arnoldi-delta/SRC/ex-sym.doc). 
   ARPACK depends on standard BLAS (Levels 1 , 2 and 3) and LAPACK libraries 
   which exist in object form on the Delta.
 
 
4) C version of LAPACK

   CLAPACK is an automated f2c conversion of Fortran 77 LAPACK into ANSI C.
   Be aware that since this is an f2c conversion of existing column-oriented
   Fortran 77 LAPACK code, all CLAPACK code is column-oriented NOT
   row-oriented.
 

5) LAPACK++ a C++ implementation of some of the LAPACK.  

   LAPACK++ is the C++ version of LAPACK.  This version includes support
   for solving linear systems using LU, Cholesky, and QR matrix factorizations.    LAPACK++ supports various matrix classes for vectors, non-symmetric
   matrices, SPD matrices, symmetric matrices, banded, triangular,
   and tridiagonal matrices; however, Version 0.9 does not include all
   of the capabilities of original f77 LAPACK.  Emphasis is given to
   routines for solving linear systems consisting of non-symmetric matrices,
   symmetric positive definite systems, and solving linear least-square systems.   Support for eigenvalue problems and singular value decompositions are
   not included in this prototype release.  Future versions of LAPACK++
   will support this as well as distributed matrix classes for parallel
   computer architectures.


6) Templates for sparse iterative methods for non-symmetric Ax=b.
   We have put together a book on iterative method for large sparse 
   nonsymmetric systems of linear equations.  The book is available in 
   postscript form on netlib or can be ordered from SIAM.
   Using concept of templates, we presents the algorithms using the 
   same notation in a straight forward manner permitting the user to 
   inspect, modify, or ignore any desired level of implementation detail. 
   Hints on parallelization, use, and other practical aspects are provided.
 
 
   In addition to the algorithmic description in the book we have
   provided a set of software in Fortran and in Matlab for the 
   following methods:

       Bi-conjugate Gradient 
       Bi-conjugate Gradient stabilized 
       Chebyshev 
       Conjugate Gradient 
       Conjugate Gradient squared 
       Generalized Minimal Residual 
       Jacobi 
       Quasi-Minimal residual 
       Successive Over-Relaxation 


The ScaLapack group:
   Oak Ridge National Laboratory
   Rice University
   University of California, Berkeley
   University of Illinois
   University of Tennessee

Comments and questions can be sent to scalapackacs.utk.edu.