BOAST: Bringing Optimization through Automatic Source-to-Source Transformations

In this paper we present an automatic source-to-source transformation which optimizes loop structures in order to find the best performance configuration for a given multi-core platform. We focus on convolution operators, optimize them using loop unrolling and show that our approach can achieve up to 67% performance gain. Our experiments have been done with the BigDFT scientific application executed on the Tibidado high-performance and low-energy-consumption machine.