Early and Accurate Modeling of Streaming Embedded Applications

This paper presents automatic generation of fast and accurate timed models of streaming embedded applications, before the complete software-hardware platform is available. First, a measurement model is generated and executed, on the target processor, to predict the computation delays in the application. Next, the stochastic delays are annotated to the application code to generate a host-compiled model of the application. Our experiments show that such models can be generated and simulated in seconds to accurately predict the computation load offered by the application. Our results with large streaming media applications, such as music and voice codecs, show that the estimation errors are less than 3.3%, while providing very high simulation speed. Therefore, using our models, embedded system designers can perform early optimizations to the system architecture with high confidence.