Multiprocessing a sparse matrix code on the Alliant FX/8 (Q1824981): Difference between revisions

Multiprocessing of a code for the direct solution of sparse linear equations on Alliant FX/8 is presented. As solution technique the multifrontal organization is used. Introducing the basic ideas of the multifrontal methods and the elimination trees, various issues concerning parallel implementation (data management, task spawning and granularity, management of work queue and performance code) are detailed. The presented experiments are performed with a modified version on the MA37 code of the author and \textit{J. K. Reid} [SIAM J. Sci. Stat. Comput. 5, 633-641 (1984; Zbl 0557.65017)]. Only the numerical factorization of the coefficient matrix is considered and the standard minimum degree ordering is used. For a portable interface for parallel programs the use of the SCHEDULE package from Argonne on the presented scheme is discussed. Then a very useful graphic output may be obtained to a SUN workstation. The presented tailored code attains a speedup by a factor about six on the eight processors of the Alliant.