[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: ilu function
From: |
Marco Caliari |
Subject: |
Re: ilu function |
Date: |
Tue, 08 Jan 2013 11:50:40 +0100 (CET) |
User-agent: |
Alpine 1.00 (DEB 882 2007-12-20) |
Dear Giovanni,
thank you very much for considering the inclusion of your
implementation of ILU function in core Octave.
Of course your actual implementation (maybe rewritten in Octave style) is
fine for small matrices, and in fact you are considering to "translate" to
C language to improve the performance.
However, since the code will be used for sparse matrices, it is not just a
translation needed, but a complete rewriting using the CCS format.
Some months ago, I also tried to write the ilu(0) factorization and I
ended up with some files here enclosed. Basically, luincfull.m is your
nofill.m and luincsparse.m is the same done with vectors data, ridx, cidx
of CCS format. If you run testluincsparse.m, you discover that even in
Octave language the sparse version is faster.
To conclude, it is a translation into C of luincsparse.m and not of
luincfull.m that we need. The result would be very similar to subroutine
ilu0 in chapter 10 in Saad's book, except that we should use the CCS
format and not the CSR (if you are interested, I translated luincsparse.m
into Fortran).
At that time, I stopped working on it and contacted Saad, who knows and
uses Octave
http://www-users.cselabs.umn.edu/classes/Spring-2012/csci8314/matlab/index.html
and he, or some students of him, were quite interested into the project
(writing ilu for Octave), but since then (April 2012) I got no more
answer.
Cheers,
Marco
luinc.tar.gz
Description: Binary data
- Ilu function, Giovanni Matteo Fumarola, 2013/01/07
- Re: ilu function,
Marco Caliari <=