CAREER: Multithreaded Algorithms, Models, and Runtime System Tools for Multimedia Applications
PI: Ashfaq A. Khokhar
This project deals with the design and development of parallel algorithms, computation model, and runtime support tools for efficient execution of multimedia applications on multithreaded computing platforms. In particular, applications that employ irregular computation structures and data dependent communication patterns are studied and used as a vehicle to investigate algorithmic and implementation issues related to multithreaded computing. While existing computation models lack in capturing the architectural and computation characteristics of the multithreaded paradigms, a new model is investigated to be used as a platform for algorithms' development. Load adaptive parallel algorithms and implementations are developed by taking advantage of the computation and communication overlapping characteristics of the multithreaded computing paradigm. The executions of the programs based on the proposed algorithms are supported by dynamic schedulers and load balancers designed as runtime support tools. The application area focussed in this project consists of multimedia signal/image processing applications, particularly those based on non-linear signal processing techniques. This project has a wide-spread impact on real-time multimedia processing tasks in defense, scientific and commercial applications.
