AFOSR BRI: Co-Design of Hardware/Software for Predicting MAV Aerodynamics
Abstract
While Moore's Law theoretically doubles processor performance every 24 months, much of the realizable performance remains untapped because the burden falls to the (less informed) domain scientist or engineer to exploit parallel hardware for performance gains. Even when such untapped hardware potential is fully realized, it is often not coupled with advances in algorithmic innovation, which can deliver further (multiplicative) speed-up beyond Moore's Law, as noted in the AFOSR BAA. In this project, we propose a formal co-design process for the structured grid and unstructured grid motifs found in computational fluid dynamics (CFD) in support of aerodynamic predictions for micro-air vehicles (MAVs). While many past efforts to develop such CFD codes on accelerated processors showed limited success, our hardware/software co-design approach created malleable algorithms that could be mapped and optimized onto the right type of processing core at the right time, and in turn, deliver an order-of-magnitude better performance than would have otherwise been possible by Moore's Law alone.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 27, 2016
- Accession Number
- AD1018148
Entities
People
- Wu-chun Feng
Organizations
- Virginia Tech