Abstract

FPG As once had a unique position in computing because they could be massively parallel on a single chip. This unique parallel identity is fading thanks to competition from GPUs and multi-core. The FPGA ’s uniqueness now relies more on its customisable architecture, direct high speed I/O and perhaps lower power. Some argue that the extra development effort required for custom architecture FPGA applications is too high to compete in this new environment. They predict the demise of FPG As in all but a narrow niche unless something is done about the development process (Some advocate an open source FPGA as a motivator for faster adoption of new tools).On the other hand those in the fixed architecture community are facing a similar crisis as the reality of parallel application development emerges. This presentation describes research results that explore issues in this controversy. The topics covered include:

1. rule based abstractions for hiding the complexity in synchronization in concurrent applications,
2. the primacy of spatial abstractions in future parallel computing,
3. proposals to standardize application development on FPG As for signal processing and image processing
4. a partial answer to the question: “What should a “compiler” for FPGA based applications do?” and
5. the relationship between hardware and software research in parallel computing for FPG As and the search for better abstractions

About the presenter

David is Associate Professor of Computer Science at the University of South Australia in Adelaide. He has published 108 refereed journal and conference papers, been awarded more than $A1 million in competitive grants and has supervised to graduation 12 higher degree by research students. He has previously worked on asynchronous architectures but in the last 15 years has concentrated in the area of reconfigurable computing with FPG As. He is especially interested in languages and operating systems for reconfigurable computing in the domain of image and signal processing.