Proposals:Sparse Linear Solvers: Difference between revisions
From KitwarePublic
Jump to navigationJump to search
| (One intermediate revision by the same user not shown) | |||
| Line 97: | Line 97: | ||
** [https://svn.boost.org/svn/boost/sandbox/numeric_bindings/ Trunk version of the automatically generated boost bindings] | ** [https://svn.boost.org/svn/boost/sandbox/numeric_bindings/ Trunk version of the automatically generated boost bindings] | ||
*** Things might be [http://permalink.gmane.org/gmane.comp.lib.boost.ublas/5573 moving towards a real release] | *** Things might be [http://permalink.gmane.org/gmane.comp.lib.boost.ublas/5573 moving towards a real release] | ||
* Alternatives to the numeric bindings for uBLAS could be to rely on: | |||
** Some [http://www.guwi17.de/ublas/examples/ user contributions]. This contains code for Cholesky decomposition, GMRES method, etc. | |||
** [http://devernay.free.fr/hacks/ublasJama.html ublasJama] a public domain port of [http://math.nist.gov/javanumerics/jama/ JAMA (Java Matrix Package)] for uBLAS | |||
== ARPACK++ == | == ARPACK++ == | ||
Latest revision as of 07:44, 24 June 2010
The Problem
ITK requires functionalities for solving large sparse linear systems.
The current solution obtained though VXL must be replaced because it relies internally on code whose Copyright is held by ACM, and is subject to a license that precludes commercial use.
Some interesting threads from the mailing lists about this issue:
- Replacing Sparse Matrix Linear Solver : due to Licensing issues
- Scope of VTK and it's potential as a common research language
- Scope of VTK and it's potential as a common research language (continued)
Previous Review
Tom Vercauteren previously gathered the following list of options.
ITK_Release_4.0#3rd_Party_Libraries
Options
Comparative Table
| Name | License | Size | Support Sparse | Support Dense | Language | Coverage | Valgrind | CMakeified | Dashboard |
|---|---|---|---|---|---|---|---|---|---|
| SuiteSparse | LGPL / GPL | Yes | C/C++ (with Fortran and Matlab interfaces) | Unknown | Unknown | ||||
| TAUCS google code page | MIT | Yes | C | Unknown | Unknown | ||||
| SuperLU | BSD | Yes | C | Unknown | Unknown | ||||
| Eigen | LGPL v3 (headers only) | 3.8 Mb | Yes | Yes | C++ | Unknown | Unknown | Yes | Yes |
| Boost uBLAS + Numeric Bindings or user contributions | Boost License but need to check binded libs | Yes | Yes | C++ | Unknown | Unknown | Under development | Yes for uBLAS | |
| ARPACK++ | BSD | Templates | Yes | C++ | Unknown | No | |||
| ARPACK | New BSD | 1.5 Mb | Yes | FORTRAN | Unknown | No | |||
| MUMPS | Public Domain | 11 Mb | Yes | FORTRAN 90 | Unknown | Unknown | |||
| lapack | BSD | 86 Mb | No | Yes | Fortran | ?? | Unknown | Yes | ?? |
| clapack | BSD | 22 Mb | No | Yes | C | 80% | Unknown | Yes | Yes |
| PLASMA | BSD | 13 Mb | No | Yes | C + FORTRAN | Yes | |||
| OpenNL | ?, BSD | 2.8 Mb | Yes | C, C++, CUDA | Unknown | Unknown | Yes | ||
| LSQR | CPL (& BSD?) | Yes | No | C, Fortran | Yes (already in ITK) | Part of ITK dashboard | |||
| OpenNL | BSD | 5.0Mb | Yes | ?? | C | 0% (no tests) | Unknown (no tests) | Yes |
SuiteSparse
- License: LGPL / GPL (depending on the submodules, e.g. CSparse and SuiteSparseQR)
- Home: http://www.cise.ufl.edu/research/sparse/SuiteSparse
TAUCS
- License: MIT
- Home: http://code.google.com/p/taucs/
SuperLU
- License: BSD
- Home: http://crd.lbl.gov/~xiaoye/SuperLU/
Eigen
- License: LGPL v3 headers only (The authors claim it is compatible with BSD software)
- Home: http://eigen.tuxfamily.org/index.php?title=Main_Page
- Cdash dashboard here
- Support for both sparse and dense matrices and algorithms
- Python bindings are available with BSD licence
MUMPS
- License:
- http://graal.ens-lyon.fr/MUMPS/
- Requires
- BLAS
Boost uBLAS + Numeric Bindings
- uBLAS hompage
- Numeric bindings
- License: seems to be BSD (but links to MUMPS, UMFPack, so better check with them directly)
- Packaged version of the old hand-written version of boost bindings
- Trunk version of the automatically generated boost bindings
- Things might be moving towards a real release
- Alternatives to the numeric bindings for uBLAS could be to rely on:
- Some user contributions. This contains code for Cholesky decomposition, GMRES method, etc.
- ublasJama a public domain port of JAMA (Java Matrix Package) for uBLAS
ARPACK++
- C++ Template API for sparse matrix containers and API to ARPACK FORTRAN library solvers
- License: BSD
- http://www.ime.unicamp.br/~chico/arpack++/
ARPACK
- to solve large scale eigenvalue problems
- License: New BSD
- http://www.caam.rice.edu/software/ARPACK/
lapack
- http://icl.cs.utk.edu/lapack-forum/viewtopic.php?f=2&t=1746
- Source Code
- Dashboard: http://my.cdash.org/index.php?project=LAPACK
- Main maintainer Site:
- Innovative Computing Laboratory
- University of Tennessee from Denver, Colorado
clapack
- Dashboard: http://my.cdash.org/index.php?project=CLAPACK
- License:
- Source code release: http://www.netlib.org/clapack/clapack-3.2.1-CMAKE.tgz
- Home page: https://icl.eecs.utk.edu/
- Repository:
- https://icl.cs.utk.edu/svn/lapack-dev/lapack/trunk/
- The code is generated from the FORTRAN version of lapack. The generation process is not under source control.
- Main maintainer Site:
- Innovative Computing Laboratory
- University of Tennessee from Denver, Colorado
PLASMA
- Home Page: http://icl.cs.utk.edu/plasma/index.html
- License: BSD
OpenNL
- Home Page: http://alice.loria.fr/index.php/software/4-library/23-opennl.html
- License: BSD
- (as specified in the header of source files)
- Size: 5.0 Mb
LSQR
- Home Page: http://www.stanford.edu/group/SOL/software/lsqr.html
- Already in ITK through the VXL-VNL-netlib chain
- License: CPL but looks to be also licensable under BSD
