Exploiting the huge logic resources in current embedded devices has led to a plethora of on-chip multi-processor architec- tures. However, besides instantiating more and more soft-core processors on a chip, developing applications suited for such architectures still remains a hard task. A further step in the evolution of embedded multi-processing might be the so called Invasive Programming. In this paradigm, an application may be switched from sequential to parallel execution at runtime. A task may then dynamically invade currently unused processor resources in a multi-processor system to resume in parallel execution mode. This hardens existing problems, however, because not only the development of suited software, but also the creation of multi-processor architectures supporting this paradigm is needed. Therefore, this work presents a concise methodology to enable Invasive Programming properties on an embedded Multi-Processor System-on-Chip (MPSoC). This is achieved by combining a designer-guided code parallelization approach with a virtualizable, generic, and scalable embedded MPSoC architecture. To resolve data dependencies during task invasion, a processor-independent task-based communication scheme for the MPSoC is proposed. Moreover, a tool framework dedicated to the generic creation of virtualizable MPSoC is provided. The approach is demonstrated by the generation of an MPSoC featuring eight processors executing an application which dynamically switches at runtime between sequential and parallel execution.
展开▼
