A Many-Core Energy and Latency Estimator

Many-core processors can now be used as an alternative hardware platform for implementing embedded media devices. However, a lack of generic tools for application development may hamper their rate of adoption by industry. This work has contributed towards the solution by providing an abstraction from the many design constraints facing application developers. A Many-core Energy and Latency Estimator (MELE) has been designed to improve the programmer’s ability to iteratively map data flow applications to a target machine. A set of models included in the tool transforms an application mapped to an abstract machine into its Intermediate Representation. An Abstract Interpreter which runs on the Ptolemy II modeling platform is used to return feedback from the Intermediate Representation to the programmer. A case study has been used to showcase the use of MELE in analyzing the mapping of data flow applications. The case study has also been used to explain how a rank based system can arrive at the most suitable mapping of an application to the processor based on energy and latency costs. Results from the case study shows that the use of a greater number of cores in the processing does not necessarily result in the highest ranked mapping. Also, some mappings take too long to arrive at their steady state processing cost value. This may result in a lower Quality of Service (QOS) for the application. Based on these findings, suggestions have been made for further work