Polyhedral Model Based Data Locality Optimization for Embedded Applications

Abstract

The need for compilers of embedded systems to find effective ways of optimizing series of loop-nests is urgent. This is especially so for streaming applications such as M-Jpeg, H.264 etc. which are popular in embedded systems. The loop bounds and memory references of these applications are primarily affine functions of the outer loop counters and constant parameters. The polyhedral model provides powerful abstractions to optimize loop nests with such regular accesses. Affine transformations in this model capture a complex sequence of execution-reordering loop transformations. We propose a solution to the data locality optimization problem for the embedded systems by using the polyhedral model. Experiments show that our technique leads to 35% reduction in external memory accesses over best gcc optimization result.