My research is an exploration of automatic task formation techniques targeted at the MLCA. The goal is to transform an existing sequential application into a set of tasks that will exhibit a high degree of parallelism. This will be done by exploiting characteristics common to many media applications that provide a natural and efficient task decomposition.
I am implementing these task formation techniques in the Open Research Compiler, an open-source variant of SGI's Pro64 compiler. The end result will be a development tool, aimed at easing the process of porting applications to the MLCA by suggesting task partitions and analyzing data-flow requirements.
Links:
" The Hyperprocessor: A Template System-onChip Architecture for Embedded Multimedia Applications," Faraydon Karim, Alain Mellan, Tarek Abdelrahman, Utku Aydonat, Bernd Stramm, Anh Nguyen, Proceedings of the Workshop on Application Specific Processors (WASP-2), San Diego, December 2003.