Petri Net-based Thread Composition for Improved System Level Simulation
Nick Savoiu, Sandeep K. Shukla, Rajesh K. Gupta · 2003
Shrinking time-to-market requirements and growing system complexity place an ever-increasing pressure on design teams to deliver quality products on time. As simulation often accounts for a considerable amount of the development cycle time reducing it can either allow more time for design improvements or directly reduce the time-to-market. We have shown that the implicit concurrency in a hardware model can be used to improve its simulation performance if we transform the threading model from that imposed by the hardware model structure to one more suitable for efficient simulation. However, as these transformations tend to be nontrivial changes to the structure of a simulation model, it is important to have a rigorous way of applying them. In this paper, we present a Petri net(PN)-based methodology for the application of such transformations using a set of structural reduction rules. The PNs are annotated with code from the system model and transformed as to extract complete code cycles from them. These cycles have better simulation performance due to lower simulation overheads and can also be mapped to threads for further simulation runtime improvements as shown by our experiments.