Static domain analysis for compiling commutative loop nests
Abstract
In the field of scientific computation, many users wish to use the sequential programming model, even though they aim to execute their program on a Distributed Memory Parallel Computer (DMPC). To meet this demand, some prototypes of compilers have been designed to "distribute" sequential programs onto DMPCs. In this paper, we present a static domain analysis which leads to the generation of efficient code for these machines. This analysis relies on the enumeration of the points of a polyhedron which is based on linear and integer programming.