Acceleration for Heterogeneous Systems Cosimulation

Heterogeneous systems simulation requires executing models using different simulators. One of the main problems consists in synchronizing all the kernels together, which significantly increases the simulation time. In this work we are interested in two currently prevailing abstractions levels: Transaction Level Modeling (TLM) and Register Transfer Level (RTL). They both require different semantics for simulating model. Acceleration for both TLM and RTL simulation models can be achieved by using transformations to build a new efficient simulation model respecting the semantics of the initial multi-language model. Our objective is to create an internal representation from descriptions in different languages, and then generate a more efficient model. To highlight the effectiveness of our approach, we show how to cosimulate ESys. Net and SystemC models with a performance gain over a shared memory cosimulation environment.