Parallel tiled QR factorization for multicore architectures

• DOI: 10.1002/cpe.1301
• Publication Date: 2008-06-03T16:46:20Z
• Authors: Alfredo Buttari; Julien Langou; Jakub Kurzak; Jack Dongarra
• Keywords: QR decomposition; Multi-core processor; Implementation; Linear algebra

Abstract As multicore systems continue to gain ground in the high‐performance computing world, linear algebra algorithms have be reformulated or new developed order take advantage of architectural features on these processors. Fine‐grain parallelism becomes a major requirement and introduces necessity loose synchronization parallel execution an operation. This paper presents algorithm for QR factorization where operations can represented as sequence small tasks that operate square blocks data (referred ‘tiles’). These dynamically scheduled based dependencies among them availability computational resources. may result out‐of‐order will completely hide presence intrinsically sequential factorization. Performance comparisons are presented with LAPACK exploited only at level BLAS vendor implementations. Copyright © 2008 John Wiley & Sons, Ltd.