面向并行图像处理的实时分布式操作系统的设计与实现

针对多片TMS320C64 DSP构成的多计算机体系结构的实时图像识别系统，设计并实现了面向并行图像处理的实时分布式操作系统PIPORTDOS（Parallel image processing-oriented real-time distributed operating system）．PIPORTDOS基于微内核体系结构，包括硬件抽象层、系统核心层、分布式消息通信机制和系统服务层四个层次．其多任务内核实现了基于优先级的抢先式调度、任务闻的同步和通信原语、实时的中断处理以及面向应用的缓存管理机制．为了实现对分布式并行图像处理的支持，PIPORTDOS采用了基于消息传递（Message Passing）的方式，并在实现中充分考虑了上层图像处理算法的应用需求以及DSP的硬件功能．相关性能指标表明，本文设计实现的PIPORTDOS完全可以满足系统的强实时性要求．在功能上也能适应算法对不同并行结构的需求．

Design and Implementation of a Real-Time Distributed Operating System for Parallel Image Processing

The PIPORTDOS (Parallel image processing-oriented real-time distributed operating system) was designed and implemented for a parallel image processing system, which is based on multicomputers architectures and consists of multi-DSPs (TMS320C64 DSP). PIPORTDOS is based on microkernel architecture, and consists of four layers: hardware abstract layer, system kernel layer, distributed message mechanism, and system servers layer. The multitasking kernel implements a priority-based preemptive scheduler, intertask synchronization and communication routines, real-time interrupt services and application-oriented memory management. Based on the message-passing model, PIPORTDOS implements a distributed message mechanism to support parallel image processing. To ensure the performance of the message mechanism, the demands of image processing algorithm and the hardware system capability are all taken into account. The performance figures show that PIPORTDOS can meet the demands of real-time response, and it also meet the demands of different parallel structures.