Bilkent University
Department of Computer Engineering
S E M I N A R
Employing Hypergraph based SpMxV routines on Intel Xeon Phi Coprocessors
Mehmet Başaran
MSc Student
Computer Engineering Department
Bilkent UniversitySparse matrix vector multiplication (SpMxV) is a kernel operation in linear solvers in which the sparse matrix is multiplied with a dense vector many times until result vector converges to particular values. However, process becomes really tedious as variable count in linear equations increases. Sometimes the size of sparse matrix increases so much that even today it can barely fit into memory let alone cache. As a result, scientists still invest a lot of effort into optimizing these linear solvers.
With our SpMxV routines we can attain considerable speed-up by capturing the data access patern and by relaxing communication volume for parallel platforms. In this work, we have targeted Intel's brand new Xeon Phi High Performance Computing Platform and ported our algorithm to make the most out of this architecture.
DATE: 23 December, 2013, Monday @ 16:15
PLACE: EA-409