RE: Sparse LU Decomposition

"König, Oliver" <[email protected]>
Newsgroups gmane.comp.lib.mtl.devel
Message-ID <[email protected]>
Hi,

We did implement a direct sparse solver in our Open Source Finite Element library FELyX. (http://felyx.sourceforge.net). FELyX uses MTL for all matrix operations, therefore the solver works with MTL matrices and vectors. In FELyX, we store stiffness matrices in the envelope sparse format (after doing node reordering with Sloan or Cuthill-McKee algorithm from Boost Graph library). We did quite some tuning on the solver, e.g. by introducing a BLAS routine from Atlas library for the most inner dot product (performance boost: ~30%). Further on, in the Finite Element library one can switch between float and double precision, choosing between "accuracy" and "speed"... 
All this together leads to a Finite Element Analysis performance that can match with commercial products as ANSYS, as long as the problems are relatively small (<100000 DOF's). 

If you are interested, I would recommend to download the CVS version from sourceforge, since the actual Release is not up to date anymore...

By the way, we are still looking for an efficient preconditioned gradient solver for FEM analyses... 

Greetings
ok

----------------------------------------------------
Oliver Koenig

Center of Structure Technologies
Institute of Mechanical Systems
Leonhardstr. 27, LEO C4
ETH Zurich
CH-8092 Zurich

Phone : +41 1 632 23 28
Fax :   +41 1 633 11 25
Mail :  [email protected]
WWW :   http://www.imes.ethz.ch/st/
----------------------------------------------------

 

-----Original Message-----
From: Björn Selent [mailto:[email protected]] 
Sent: Donnerstag, 28. August 2003 17:54
To: General Matrix Template Library (MTL) list
Subject: Re: MTL: Sparse LU Decomposition

hello,
thanks for the quick reply. Do you mind to send me the source code for direct 
solvers you have so far?
Bjoern

On Thursday 28 Aug 2003 2:17 pm, Andrew Lumsdaine wrote:
> We did do some initial prototypes of direct sparse solvers.  The
> challenge in creating a solver similar to SuperLU is that the structure
> of the matrix must be able to accommodate variable-sized dense blocks.
> If you are interested in writing such an algorithm, we can probably
> lend a hand.
>
> Related to this, we also have some fill-reducing sparse matrix ordering
> algorithms as part of the Boost Graph Library (www.boost.org), which
> you may also find useful.
>

-- 
In Dublin's fair city where girls are so pretty
Twas there that I first met sweet Molly Malone
As she wheeled her wheelbarrow
Through street broad and narrow
Crying, "Cockles and mussels, alive, alive oh"
_______________________________________________
This list is archived at http://www.osl.iu.edu/MailArchives/mtl-devel/
_______________________________________________
This list is archived at http://www.osl.iu.edu/MailArchives/mtl-devel/
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.