About this event
Abstract The phenomenal growth in the capabilities of modern software and its influence on the world is based on two simple facts: first, modern software can be developed for a simple sequential von Neumann machine abstraction; second, the software can expect to double its performance every 18 months due to Moore's law. While Moore's law is still in effect, it has stopped providing the same performance gains for sequential processors. Thus, all hardware vendors have switched to multi-core processors. Soon the software community will face an unpleasant choice: either learn to live with minimal performance improvements or learn to run the applications on multiple cores and get the performance back on track with Moore's law. While the second option seems more attractive, obtaining pervasive and scalable parallelism has been the holy grail for many computer scientists since the days of ILLIAC-IV; despite much effort, the dream has yet to be fully realized.
In this talk I will examine possible paths that can potentially break this logjam. First, I will look at how novel languages can break the von Neumann barrier without unduly burdening the programmer. Second, I will discuss how compilers and tools can help the parallelization process. Finally, I will present novel architectural opportunities created in the shift to multi-cores that can potentially eliminate some of the difficulties and costs of parallel execution.If you would like to arrange a meeting with Dr. Amarasinghe, please contact Angelica Brewer or Dawn Patterson at ext. 8910.