Data-distributions in powerlist theory

Authors:
Virginia Niculescu
Affiliations:
Faculty of Mathematics and Computer Science, Babeş-Bolyai University, Cluj-Napoca
Venue:
ICTAC'07 Proceedings of the 4th international conference on Theoretical aspects of computing
Year:
2007

Citing 7
Cited 0

Powerlist: a structure for parallel recursion

ACM Transactions on Programming Languages and Systems (TOPLAS)
A semantics for shape

ESOP '94 Selected papers of ESOP '94, the 5th European symposium on Programming
Models and languages for parallel computation

ACM Computing Surveys (CSUR)
Parallel programming: techniques and applications using networked workstations and parallel computers

Parallel programming: techniques and applications using networked workstations and parallel computers
Costing parallel programs as a function of shapes

Science of Computer Programming - Special issue on mathematics of program construction
Data structures for parallel recursion

Data structures for parallel recursion
On Data Distributions in the Construction of Parallel Programs

The Journal of Supercomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

PowerList theory is well suited to express recursive, data-parallel 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.