Personal tools

FSPAI

From Sccswiki

Jump to: navigation, search

Contents

Factorized Sparse Approximate Inverses

This is a dummy textLatest release is 1.1

We developed a sequential and highly scalable parallel C/C++ implementation of the
known FSPAI (Factorized SParse Approximate Inverses) algorithm.

Note that there is a scalable implementation of Modified Sparse Approximate
Inverses (MSPAI)
as well.

Theory

FSPAI is a preconditioner for large sparse and ill-conditioned symmetric positive
definite systems of linear equations. It is the factorized version of the SPAI
algorithm. FSPAI is inherently parallel and generates a preconditioner which
approximates the inverse of the Cholesky factor of the system matrix, i.e.,

Fspai.png

Based on an initial chosen sparsity structure, FSPAI automatically updates its
sparsity structure and improves on a current approximation.

Features

  • Single source providing
    • highly scalable parallel implementation
    • sequential (MPI free) implementation
  • Support for real and complex valued problems
  • PCG solver available using HYPRE package
  • Arbitrary start patterns, i.e., possibility to compute a FSAI without
    updating the pattern
  • Support for sparse methods using CXSparse by Tim Davis
  • Caching and Hashing approach to avoid redundant Cholesky decompositions
  • Written in C/C++

We are currently working on a next release which will cover a highly scalable
block version of FSPAI, i.e., BFSPAI.

Download

FSPAI 1.1

In contrast to FSPAI 1.0, FSPAI 1.1 additionally requires CXSparse and UFconfig.
If you are not going to use the Fspai Caching/Hashing or the sparse methods,
you can use FSPAI 1.0 instead.

FSPAI 1.0

Tested environments

  • Shaheen from the KAUST Supercomputing Laboratory (KSL) at King
    Abdullah University of Science and Technology (KAUST):
    • 65536 compute cores: 16 racks of Blue Gene/P, each containing
      1024 quad-core, 32-bit 850 MHz PowerPC compute nodes
    • Architecture: PowerPC (PPC)
    • Compiler version: IBM XL C/C++ Advanced Edition for Blue Gene/P, V9.0
    • Libraries:
      • Lapack 3.2.1
      • MPI based on MPICH2
      • HYPRE 2.7
  • InfiniBand-Cluster of the Lehrstuhl Rechnertechnik und Rechnerorganisation/
    Parallelrechnerarchitektur of Technische Universtität München:
    • 128 compute cores: AMD Opteron™ Processor 850
    • Architecture: x86_64
    • Compiler version: gcc version 4.2.4 (Ubuntu 4.2.4-1ubuntu3)
    • Libraries:
      • Lapack 3.0
      • MPI based on MPICH 1.2.7
      • HYPRE 2.7.0b
  • i686-pc-linux-gnu:
    • Processor: Intel® Core™2 Duo CPU P8700 @ 2.53GHz
    • Architecture: i686
    • Compiler version: gcc version 4.3.4 (Gentoo 4.3.4 p1.1, pie-10.1.5)
    • Libraries:
      • Lapack 3.3.1
      • HYPRE 2.7.0b


FSPAI is provided "as is", without warranty and support of any kind, express or
implied, including but not limited to the warranties of merchantability, fitness for a
particular purpose, title and non-infringement. In no event shall the copyright holders
or anyone distributing FSPAI be liable for any damages or other liability, whether in contract, tort or otherwise, arising from, out of or in connection with FSPAI or the use
or other dealings in FSPAI.

Some References

Papers

  • T. Huckle: Factorized Sparse Approximate Inverses for Preconditioning [BibTeX].
    In Journal of Supercomputing, Volume 25(2), p. 109–117, 2003.
  • T. Huckle: Factorized Sparse Approximate Inverses for Preconditioning and Smoothing [BibTeX].
    In Selcuk Journal of Applied Mathematics, Volume 1(1) of Sonderheft zum 60. Geburtstag von Prof. Chr. Zenger, p. 63–70, 2000.
  • M. Grote and T. Huckle: Parallel preconditioning with sparse approximate inverses [BibTeX].
    In SIAM J. Sci. Comput., Volume 18(3), 1997.
  • A. Kallischko: Modified Sparse Approximate Inverses (MSPAI) for Parallel Preconditioning [] [BibTeX].
    Dissertation, Fakultät für Mathematik, Technische Universität München, March 2008.


Thesis

  • M. Sedlacek: Sparse Approximate Inverses for Preconditioning, Smoothing, and Regularization [] [BibTeX].
    Dissertation, Fakultät für Informatik, Technische Universität München. mediaTUM, München, October 2012.


Further

  • Short summary on sparse approximate inverses with core reference list: summary.pdf
  • Extended reference list: extended.pdf

Authors

License

FSPAI: Factorized Sparse Approximate Inverses
Copyright © 2012-2013, Matous Sedlacek
Research Unit Scientific Computing in Computer Science - Informatics V
Technische Universität München

This program is free software: you can redistribute it and/or modify it under the
terms of the GNU Lesser General Public License as published by the Free Software
Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License along with
this program. If not, see http://www.gnu.org/licenses/.

If you obtain any results with FSPAI we would appreciate that you refer to FSPAI.

Further work on Sparse Approximate Inverses