A generic functional simulation of heterogeneous systems

Abstract

Virtual Prototypes are often used for software development before the actual hardware configuration of the finished product is available. Today’s platforms often provide different kinds of processors forming a heterogeneous system. For example, ADAS applications require dedicated realtime processors, parallel accelerators like graphics cards and general purpose CPUs. This paper presents an approach for creating a simulation system for a heterogeneous system by using already available processor models. The approach is intended to be flexible and to support different kinds of models to fulfill the requirements of a heterogeneous system. Simulators should easily be exchangeable by simulators with the same architecture support. It was possible to identify the SystemC connection of the considered general purpose CPU models as a bottleneck for the simulation speed. The connection to the realtime core suffers from a necessary connection via the network which is evaluated in more detail. Combining the GPU emulator with the rest of the system reduces the simulation speed of the CUDA kernels in a negligible manner.