Optimization of an LU Factorization Routine Using Communication/Computation Overlap

Frédéric Desprez (1) , Stéphane Domas (1) , Bernard Tourancheau (1)

1 REMAP - Regularity and massive parallel computing

This report presents some works on the $LU$ factorization from the ScaLAPACK library. First, a complexity analysis is given. It allows to compute the optimal block size for the block scattered distribution used in ScaLAPACK. It also gives the communication phases that are interesting to overlap. Second, two optimizations based on computations/communications overlap are given with experimental results on Intel Paragon and IBM SP2 systems.

Mots clés: PARALLEL LINEAR ALGEBRA; LU FACTORIZATION; OVERLAP; SCALAPACK

Domaines: Autre [cs.OH]

Frédéric Desprez, Stéphane Domas, Bernard Tourancheau. Optimization of an LU Factorization Routine Using Communication/Computation Overlap. [Research Report] RR-3094, INRIA. 1997. ⟨inria-00073597⟩