Efficiency improvements of iterative numerical algorithms on modern architectures [Elektronische Ressource] = Verbesserung der Effizienz für iterative numerische Algorithmen auf modernen Architekturen / vorgelegt von Jan Treibig

E ciency Improvements of Iterative NumericalAlgorithms on Modern ArchitecturesVerbesserung der E zienz fur iterative numerischeAlgorithmen auf modernen ArchitekturenDer Technischen Fakult at derUniversit at Erlangen-Nurn bergzur Erlangung des GradesDOKTOR{INGENIEURvorgelegt vonDipl.-Ing. Jan TreibigErlangen | 2008Als Dissertation genehmigt vonder Technischen Fakult at derUniversit at Erlangen-Nurn bergTag der Einreichung: 22.12.2008Tag der Promotion: 18.6.2009Dekan: Prof. Dr.-Ing. habil. J. HuberBerichterstatter: Prof. Dr. U. RudeProf. Dr. M. PhilippsenAbstractFor many numerical codes the transport of data from main memory to the registers is com-monly considered to be the main limiting factor to achieve high performance on present microarchitectures. This fact is referred to as the memory wall. A lot of research is targeting thispoint on di erent levels. This covers for example code transformations and architecture awaredata structures to achieve an optimal usage of the memory hierarchy found in all present microarchitectures. This work shows that on modern micro architectures it is necessary to also takethe requirements of the Single Instruction Multiple Data (SIMD) programming paradigm anddata prefetching into account to reach high e ciencies.