A Tale of Three Runtimes

TL;DR

This paper presents a method for automatically generating event-driven, tuple-space programs from sequential C code, enabling efficient execution across multiple runtimes like CnC, SWARM, and OCR, with experimental validation.

Contribution

It introduces a hierarchical mapping approach using auto-parallelizing compiler technology to target various EDT-based runtimes from a sequential C specification.

Findings

Automatic code generation improves runtime efficiency.

Hierarchical async-finish support enhances runtime flexibility.

Cross-runtime comparisons demonstrate performance benefits.

Abstract

This contribution discusses the automatic generation of event-driven, tuple-space based programs for task-oriented execution models from a sequential C specification. We developed a hierarchical mapping solution using auto-parallelizing compiler technology to target three different runtimes relying on event-driven tasks (EDTs). Our solution benefits from the important observation that loop types encode short, transitive relations among EDTs that are compact and efficiently evaluated at runtime. In this context, permutable loops are of particular importance as they translate immediately into conservative point-to-point synchronizations of distance 1. Our solution generates calls into a runtime-agnostic C++ layer, which we have retargeted to Intel's Concurrent Collections (CnC), ETI's SWARM, and the Open Community Runtime (OCR). Experience with other runtime systems motivates our introduction of support for hierarchical async-finishes in CnC. Experimental data is provided to show the benefit of automatically generated code for EDT-based runtimes as well as comparisons across runtimes.