RISC/DSP处理器数据转发机制设计

Design of Bypassing Mechanism of RISC?DSP Processor

针对一种RISC/DSP结构处理器MediaDSP3201(MD32),给出了一种分布式数据转发机制设计策略,有效地避免了MD32在执行过程中不必要的流水线停顿,并通过“数据转发链模型”实现.此策略在考虑转发效率的同时,通过电路优化避免转发电路对流水级时延的影响,以提高处理器整体性能.最后以MPEG解码程序为例,说明该策略以较小的硬件成本(占MD32资源的3.7%)有效地降低了CPI值,比集中式数据转发机制的处理性能提高了36%.

For a kind of RISC/DSP architecture processor： MediaDSP3201（MD32, for snort ）, a distributed bypassing unit strategy is proposed. The strategy adopts an effective mechanism to avoid unnecessary pipeline stall. A bypassing circuit chain model is used to realize the proposed strategy in MD32. A circuit optimization method for DBPU is used to avoid pipeline stage delay, so as to improve the overall performance for the processor. Finally, MPEG decoding program is taken as an example to show the improvement of the processor performance by reducing the CPI value： 36% compared to centralized method, and the hardware cost （accounting for 3.7 % of the processor resources） is small.