Interaction Between Data Parallel Compilation and Data Transfer and Storage Cost for Multimedia Applications

Real-time multi-media applications need large processing power and yet require a low-power implementation in an embedded context. Thus for programmable parallel processors, this poses new challenges for optimizing a given application for high-performance and low-power. In this paper, we present a case study of applying our low-power oriented DTSE (data transfer and storage exploration) methodology and coupling it with a state-of-the-art performance optimizing and parallelizing compiler. Experiments on two real-life applications show that this combined approach heavily reduces the memory access and bus-loading and hence power. At the same time a significant reduction in the total execution time is obtained. Decomposing the detailed parallelization and DTSE issues into two different stages is important to obtain the important benefits of both the stages without exploding the complexity of solving all the issues simultaneously. This will be demonstrated by the experimental results.