Processing Algorithmic Skeletons at Compile-Time

EasyChair Preprint 2702

Abstract

When developing software in Operational Research (OR), one usually aims at getting the shortest execution time. This can be done by parallelizing the execution, however it requires coding effort and specific knowledge.

We proposed a library that ease the building of parallel implementations of OR algorithms through algorithmic skeletons. These skeletons allow the library to have knowledge of the algorithms structures at compile-time. The library uses this to generate parallel code without runtime overhead.

We present an intermediary library that makes the processing of this compile-time information similar to usual runtime programming for an end developer. It enables the processing of skeletons as trees or vectors and applying well-known transformations on these compile-time structures. We show some use cases for this tool while explaining how it works.

Keyphrases: Algorithmic Skeletons, Metaheuristics, compile-time, metaprogramming

@booklet{EasyChair:2702,
  author    = {Alexis Pereda and David R.C. Hill and Claude Mazel and Loïc Yon and Bruno Bachelet},
  title     = {Processing Algorithmic Skeletons at Compile-Time},
  howpublished = {EasyChair Preprint 2702},
  year      = {EasyChair, 2020}}