An automated approach to mapping DSP algorithms onto transputer arrays

There is an increasing demand for low cost, high throughput multiprocessors on which to implement real-time DSP applications. A general purpose solution can be offered using an MIMD architecture. MIMD architectures can come in a variety of different forms. Loosely-coupled, homogeneous architectures are examined, because they offer infinite expandability and a low cost:processing ratio. One potential disadvantage of loosely-coupled architectures arises at run-time, when program and data have to be dynamically scheduled onto processors. Run-time scheduling overheads tend to curtail performance. However, many real-time DSP applications have a deterministic structure. This allows scheduling to be performed at compile-time, in a static fashion, and so reduce run-time overheads. Static scheduling involves mapping the application (in the form of a task graph) onto the processor architecture (in the form of a processor graph)