Additional key features required for different directives based porting approaches

TL;DR

This paper discusses the key features needed for OpenMP and OpenACC directives to enhance performance portability in weather prediction models, based on experiences from the ESCAPE project.

Contribution

It provides recommendations for directive features and highlights the need for improved interoperability and debugging support in mixed-directive programming environments.

Findings

OpenMP 5.0 includes deep copy and multi-level memory features.

OpenACC TR16 discusses features beyond version 2.6.

Coordination of runtime and debugging for mixed OpenMP and OpenACC use is needed.

Abstract

This document is one of the deliverable reports created for the ESCAPE project. ESCAPE stands for Energy-efficient Scalable Algorithms for Weather Prediction at Exascale. The project develops world-class, extreme-scale computing capabilities for European operational numerical weather prediction and future climate models. This is done by identifying Weather & Climate dwarfs which are key patterns in terms of computation and communication (in the spirit of the Berkeley dwarfs). These dwarfs are then optimised for different hardware architectures (single and multi-node) and alternative algorithms are explored. Performance portability is addressed through the use of domain specific languages. This report summarizes key features required for OpenMP and OpenACC directives based on experience in the ESCAPE project. For OpenMP, the latest public draft standard, 5.0, contains the deep copy and multi-level memory features desired; for OpenACC, Technical Report 16 summarizes ongoing discussions beyond the Standard version 2.6. This document includes a summary of our recommendations on this approach. Additional work is also desirable in coordinating the runtime and debugging when both OpenACC and OpenMP directives are used; in particular the interoperability of the new OPDT debugging interface for OpenMP and its semantics in a mixed-directive program.