Optimization and scalability of tiled code generation

PIPS is a source-to-source compiler developed by CRI ParisTech performing loop tiling to enforce locality and parallelism.In this work we designed a new PIPS phase performing an invariant code optimization and parallel directive selection on the generated tiled code.We obtained scalable tiled code and minimized the parallel directive overhead.The current PIPS generated code outperforms the previous one and achieves comparable results to other state-of-art code optimizers in terms of speed-up.