Author: Fresno Javier
Publisher: Springer Publishing Company
ISSN: 0920-8542
Source: The Journal of Supercomputing, Vol.64, Iss.1, 2013-04, pp. : 59-68
Disclaimer: Any content in publications that violate the sovereignty, the constitution or regulations of the PRC is not accepted or approved by CNPIEC.
Abstract
Layout methods for dense and sparse data are often seen as two separate problems with their own particular techniques. However, they are based on the same basic concepts. This paper studies how to integrate automatic data-layout and partition techniques for both dense and sparse data structures. In particular, we show how to include support for sparse matrices or graphs in Hitmap, a library for hierarchical tiling and automatic mapping of arrays. The paper shows that it is possible to offer a unique interface to work with both dense and sparse data structures. Thus, the programmer can use a single and homogeneous programming style, reducing the development effort and simplifying the use of sparse data structures in parallel computations. Our experimental evaluation shows that this integration of techniques can be effectively done without compromising performance.