PowerList theory is well suited to express recursive, dataparallel algorithms. Its abstractness is very high and ensures simple and correct design of parallel programs. We try to reconcile this high level of abstraction with performance by introducing data-distributions into this theory. One advantage of formally introducing distributions is that it allows us to evaluate costs, depending on the number of available processors, which is considered as a parameter. The analysis of the possible distributions for a certain function may also lead to an improvement in the design decisions. Another important advantage is that after the introduction of data-distributions, mappings on real parallel architectures with limited number of processing elements can be analyzed. Case studies for Fast Fourier transform and rank-sorting are given. © Springer-Verlag Berlin Heidelberg 2007.
CITATION STYLE
Niculescu, V. (2007). Data-distributions in PowerList theory. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 4711 LNCS, pp. 396–409). Springer Verlag. https://doi.org/10.1007/978-3-540-75292-9_27
Mendeley helps you to discover research relevant for your work.