System Analysis and Hardware/Software Based Approach Partitioning of Dataflow Programs

Abstract: Demand for more system functionality and computational resources are increasing exponentially. Available hardware technology can barely operate at higher frequencies and densities due to silicon and electron physical properties. This therefore drives the need to move towards parallel systems. Available programming languages are mainly designed with single CPU in mind and lose their competence when it comes to multicore systems that are already available. Dataflow programming and the CAL Actor Language (CAL) appears to be a promising solution. As part of the Ptolemy II project, CAL was invented for the specification of dataflow models. It increased the typical programming abstraction level to a higher abstraction level. This feature makes CAL an appropriate choice for system level specification and a method to achieve the desired computational performance in shorter time. There are existing CAL tools that can generate code for both hardware as well as software design. This feature makes CAL a potential solution for entire system development in the near future. It is notable that CAL has been recently adopted by ISO/MPEG as a specification language for MPEG-RVC (Moving Picture Experts Group - Reconfigurable Video Coding). The main aim of this thesis is to develop a method based on the existing tools that will enable CAL users to arbitrarily map functional units to hardware (FPGA fabric) or software (Processors) without any notion and knowledge of underlying interconnection and hardware design. Alongside this, the developed method is analyzed and a cost function and partitioning guidelines are presented at the end of this master thesis work.