Speaker

Robert van de Geijn, Institute for Computational Engineering and Sciences, University of Texas, Austin, USA

Contents

With the arrival of multicore, many-core, and multi-accelerator architectures, possibly in conjunction with heterogeneous clusters, a flexible alternative to (Sca)LAPACK deserves consideration. As part of the FLAME project, we have developed an alternative to LAPACK, libflame. It differentiates itself in a number of ways from LAPACK: it is written in C (and requires no fortran compiler); it supports matrices stored in column and/or row-major order (and general row and column stride); it incorporates families of algorithms for each operation so that the best algorithm can be chosen for a given architecture; it can be easily configured to use multicore and multiGPU via a runtime system that schedules algorithms-by-blocks. Best of all, it achieves competitive performance, often exceeding that of LAPACK and/or vendor libraries and includes an optional LAPACK interface. Similarly, for massively parallel architectures as well as distributed memory many-core architectures, an alternative to ScaLAPACK called Elemental has been developed. This new library is coded in C++ and builds on many of the insights from the PLAPACK and FLAME projects. In this talk, these and other FLAME-related developments are showcased.

Time

Thursday, 12 April 2012, 14:00

Venue

Jülich Supercomputing Centre, Hörsaal, building 16.3, room 006

Announcement as pdf: