Currently, I'm working on commutativity.
My PhD topic is Parallelism discovery and mapping to heterogeneous multi-core DSPs. DSPs - Digital Signal Processors - are aptly named and are specialized in manipulating audio, video, networking, and other digital signals. They are especially good at performing vector instructions - Single Instruction Multiple Data kind of instructions. With the hit of the Power Wall, heterogeneous multi-core DSPs become a possible next step in the evolution.
Parallelism discovery is an old topic. However, it has not been applied to DSPs extensively, as it is not so common to have multi-core ones. Thus the specific knowledge of the target platform can be leveraged to achieve automatic parallelisation superior to the one that can be reached for general purpose processors. With the aid of profiling, vectorization possibilities can be revealed and exploited.
Mapping involves determining the best partition and scheduling of computation and data to the target processor. This requires a higher-level analysis of the workload in order to identify whether tasks are heterogeneous or homogeneous in complexity. This knowledge defines the mapping to the target processor.
December 2016: Free Rider: A Source-Level Transformation Tool for Retargeting Platform-Specific Intrinsic Functions, Journal ACM Transactions on Embedded Computing Systems (TECS'16) extension to the LCTES'15 paper
April 2015: Free Rider: A Tool for Retargeting Platform-Specific Intrinsic Functions, Proceedings of the 16th ACM SIGPLAN/SIGBED Conference on Languages, Compilers and Tools for Embedded Systems 2015 (LCTES'15)
October 2013: Started PhD
2009 to 2013: BSc Computer Science and Mathematics at The Univerity of Edinburgh
School of Informatics
Room 1.05, Informatics Forum
10 Crichton Street Edinburgh