High performance code generation from symbolic computing

Abstract

Upcoming disruptive changes to computer architectures offer many new opportunities for developing high performance applications, but it also demanding disruptive changes in software to achieve the full potential of the new hardware. Therefore, the question now is: how we achieve an acceptable degree of performance portability across different (and rapidly evolving) architectures? There is in general a sharp trade-off between easy to maintain, V WORKSHOP DE PESQUISA TECNOLOGIA E INOVAÇÃO (PTI) I SIMPÓSIO INTERNACIONAL DE INOVAÇÃO E TECNOLOGIA (SIINTEC) 10 e 11 de setembro de 2015 / Salvador, Bahia, Brasil 577 extensible portable software written using high-level languages, and highly optimized parallel code for a target architecture. The solution proposed by the OPESCI (Open portable Seismic Imaging) framework is to leverage domain specific languages (DSL) and code generation software technologies to introduce multiple layers of software abstraction. At the highest level of abstraction application developers will be able to write algorithms in a clear and concise manner akin to how the algorithm might be written mathematically on paper. While at the lower levels source-to-source compilers will explore a rich implementation space to transform this DSL code into highly optimized code that can be compiled for a target platform to run at near-to-peak performance. It will provide layers that decouple domain experts from code tuning specialists, where different optimized code generator back ends can be replaced, and the high level code attains its portability. The result is a separation of concerns where new numerical approaches are readily evaluated and are capable of matching or outperforming hand tuned code.