Threshold pivoting for dense LU factorization on distributed memory multiprocessors
J. M. Malard · 1991
The cost of pivoting in L U factorization has be come non-negligible on MIMD computers due to the acceleration of floating point ari thmetic and compar atively slow communications. This study addresses the L U factorization of matrices stored by row. The importance of effici ently broadcasting pivot data is stressed. Multicasting is found preferable in this re spect to broadcasting along minimum spanning trees. Threshold pivoting is shown to effectively reduce the number of messages while preserving a good load bal ance.