Evaluation of platform architecture performance using abstract instruction-level workload models

Evaluation of platform performance is critical in the optimisation and validation of integrated application domain-specific multiprocessor systems. This work describes a method for creating abstract instruction-level workload models from source code, and a method for modelling multi-processor platforms. The approaches are validated by simulating complex use cases in a multiprocessor platform and comparing some of the results to measurements obtained from a prototype product. The approach is targeted at defining architecture parameters and to feature feasibility studies at product concept creation phase.