SHF:Medium:Collaborative Research: Decentralized On-Chip Infrastructure for Robustness and Portability in Heterogeneous Multicores

Princeton University, Princeton NJ

Investigators

Abstract

When new hardware options emerge in modern computer systems, software code must be rewritten or retailored to work on the new systems. The result is that application code-bases do not seamlessly port from one generation to another. Even worse, they often cannot fully or nimbly adjust to dynamic variations during execution even on a given system. This work will mitigate the complexity, portability, and robustness challenges of heterogeneous platforms, while also continuing to garner high performance. Leveraging the research team's experience in open-source software release and curriculum development, this project will distribute the tools developed in these research thrusts as well as teach these new tools via an existing course on design and programming of heterogeneous architectures. Addressing the research challenges of this project requires moving away from a processor-centric viewpoint, and towards a broader perspective aimed at managing communication issues. The research team will introduce the concept of "hardware shims" that can be employed for a range of uses such as acting as prefetchers, translating between different communication protocols, or assisting with dynamic verification of properties specified by the designer. The project will automate the design and synthesis of static or dynamic shims that can verify the memory consistency of an arbitrary heterogeneous multicore for which the designer provides a set of ordering specifications. The proposed research activities will demonstrate gains in performance, functional robustness, and code portability as the application is compiled and executed on systems that differ significantly in terms of the number and type of processors and specialized hardware accelerators that they use. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.