High-level strategies for parallel shared-memory sparse matrix-vector multiplication
The Faculty of Informatics is pleased to announce a seminar given by Albert-Jan Yzelman
DATE: Wednesday, September 12th, 2012
PLACE: University of Lugano, room A24, Red building (Via G. Buffi 13)
The sparse matrix-vector multiplication is an important kernel, but is hard to efficiently execute even in the sequential case. The problems —namely low arithmetic intensity, inefficient cache use, and limited memory bandwidth— are magnified as the core count on shared-memory parallel architectures increases. Existing techniques are discussed in detail, and categorised chiefly based on their distribution types. Based on this new parallelisation techniques are proposed. The theoretical scalability and memory usage of the various strategies are analysed, and experiments on multiple NUMA architectures confirm the validity of the results. One of the newly proposed methods attains the best average result in experiments, in one of the experiments obtaining a parallel efficiency of 90 percent.
Albert-Jan Nicholas Yzelman was born in Zevenaar, the Netherlands. He was raised in Duiven, and after obtaining his Atheneum diploma there in 2002, he started his scientific career at Utrecht University. By 2007, he graduated with a B.Sc. in both Mathematics and Computing Sciences, and a M.Sc. in Scientific Computing; his final internship involved researching R-trees in application to oil reservoir simulation software, done in collaboration with Alten Nederland and Shell.
Later, he received his PhD from Utrecht University in 2011 for his thesis `Fast sparse matrix-vector multiplication by partitioning and reordering', written under supervision of Prof. dr. Rob H. Bisseling. Albert-Jan currently lives in Leuven, Belgium, and works at KU Leuven within the ExaScience Lab, part of Intel Labs Europe.
HOST: Prof. Olaf Schenk