Help: parallel LU decomposition

Article View: sci.physics.computational.fluid-dynamics

Article #14

Back to sci.physics.computational.fluid-dynamics Previous Next

#14

From: Hin-Fan M. Lau
Date: Sat, 27 Jan 1996 00:00

14 lines
1347 bytes

I am working on developing an unstructured flow solver on the CM5 using the
SIMD (data parallel) programming model.  The solver uses an implicit time
stepping scheme.  Presently, I am trying to incorporate a preconditioning
to the iterative solver to increase the allowable CFL number.

The problem I am encountering is in the LU decomposition of the 
preconditioning.  The Gauss Elimination scheme from the serial code does
not work well for the parallel code due to its recursive nature and large
nested loops.

Does anyone know of any LU decomposition schemes which can be efficiently
executed on parallel computers in the data parallel mode?

Thank you very much in advance for any help given.

Technical Details

Message-ID: <4ebgtm$l8n@frodo.rutgers.edu>

Path: arc1-ca.pugleaf.net!unknown!not-for-mail

Show Full Headers

Complete Article Headers

X-Google-Language: ENGLISH,ASCII-7-bit
X-Google-Thread: 10acec,e4b2570511d4bb92,start
X-Google-Attributes: gid10acec,public
X-Google-Thread: 106aa4,e4b2570511d4bb92,start
X-Google-Attributes: gid106aa4,public
X-Google-Thread: 102daf,e4b2570511d4bb92,start
X-Google-Attributes: gid102daf,public
From: hlau@jove.rutgers.edu (Hin-Fan M. Lau)
Subject: Help: parallel LU decomposition
Date: 1996/01/27
Message-ID: <4ebgtm$l8n@frodo.rutgers.edu>
X-Deja-AN: 136571074
organization: Rutgers University
keywords: LU, parallel, preconditioning
newsgroups: sci.physics.computational.fluid-dynamics,comp.parallel,fj.comp.parallel