Generation and implementation of DSP parallel programs from a signal processing design environment

摘要

Multiprocessor digital signal processing systems are an effective medium for implementing high-speed real-time signal processing algorithms. While multiprocessor systems are becoming widely available, software for generating multiprocessor code is not as easily obtainable. Although the general problem of generating multiprocessor schedules is a difficult one, parallel schedules for digital signal processing algorithms can be generated from sequential code when that code is written in a specific format. The preprocessor, scheduler and postprocessor that constitute the Software Development System (SDS) for generating parallel schedules are described. Restrictions on their application and degree of processor dependency in the SDS are given. Schedules for various practical signal processing algorithms are obtained and implemented on a message-passing multiprocessor system called ARIA. Performance gains are reported.