Solving the advection PDE on the cell broadband engine

Rokos, G; Peteinatos, G; Kouveli, G; Goumas, G; Kourtis, K; Koziris, N

dc.contributor.author	Rokos, G	en
dc.contributor.author	Peteinatos, G	en
dc.contributor.author	Kouveli, G	en
dc.contributor.author	Goumas, G	en
dc.contributor.author	Kourtis, K	en
dc.contributor.author	Koziris, N	en
dc.date.accessioned	2014-03-01T02:46:59Z
dc.date.available	2014-03-01T02:46:59Z
dc.date.issued	2010	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/32979
dc.subject	Advection equation	en
dc.subject	Cell Broadband Engine	en
dc.subject	Explicit memory hierarchy	en
dc.subject	Instruction scheduling	en
dc.subject	Parallelization	en
dc.subject	Vectorization	en
dc.subject.other	Advection equations	en
dc.subject.other	Cell Broadband Engine	en
dc.subject.other	Instruction scheduling	en
dc.subject.other	Memory hierarchy	en
dc.subject.other	Parallelizations	en
dc.subject.other	Vectorization	en
dc.subject.other	Aircraft engines	en
dc.subject.other	Distributed parameter networks	en
dc.subject.other	Parallel architectures	en
dc.subject.other	Partial differential equations	en
dc.subject.other	Program compilers	en
dc.subject.other	Advection	en
dc.title	Solving the advection PDE on the cell broadband engine	en
heal.type	conferenceItem	en
heal.identifier.primary	10.1109/IPDPSW.2010.5470761	en
heal.identifier.secondary	http://dx.doi.org/10.1109/IPDPSW.2010.5470761	en
heal.identifier.secondary	5470761	en
heal.publicationDate	2010	en
heal.abstract	In this paper we present the venture of porting two different algorithms for solving the two-dimensional advection PDE on the CBE platform, an in-place and an outof- place one, and compare their computational performance, completion time and code productivity. Study of the advection equation reveals data dependencies which lead to limited performance and inefficient scaling to parallel architectures. We explore programming techniques and optimizations which maximize performance for these solver versions. The out-ofplace version is straightforward to implement and achieves greater raw performance than the in-place one, but requires more computational steps to converge. In both cases, achieving high computational performance relies heavily on manual source code optimization, due to compiler incapability to do data vectorization and efficient instruction scheduling. The latter proves to be a key factor in pursuit of high GFLOPS measurements. © 2010 IEEE.	en
heal.journalName	Proceedings of the 2010 IEEE International Symposium on Parallel and Distributed Processing, Workshops and Phd Forum, IPDPSW 2010	en
dc.identifier.doi	10.1109/IPDPSW.2010.5470761	en