This paper addresses two issues in linear algebra algorithms for multicomputers: first, how to unify parallel implementations of the same algorithm in a decomposition-independent way;and second, how to optimize naive ...
详细信息
This paper addresses two issues in linear algebra algorithms for multicomputers: first, how to unify parallel implementations of the same algorithm in a decomposition-independent way;and second, how to optimize naive parallelprograms maintaining the decomposition independence. Several matrix decompositions are viewed as instances of a more general allocation function called Subcube Matrix Decomposition. By this meta-decomposition, we define a programming environment characterized by general primitives that allow us to design meta-algorithms independent from a particular decomposition. For example, we apply such a framework to the parallel solution of dense matrices, thus demonstrating that most of the existing algorithms can be derived by suitably setting the primitives used in the meta-algorithm. A further original application of this programming style concerns the optimization of parallel algorithms. The idea to overlap communication and computation has been extended from 1-D decompositions to 2-D decompositions;thus, we provide a first attempt toward a decomposition-independent definition of such optimization strategies.
暂无评论