有效利用片上分块存储器

Efficient utilization of scratch-pad memory banks

为了提高嵌入式多媒体应用的实时性能,提出了一种最大化数据并行访问以便充分发挥CPU处理能力的片上存储器分配方法。CPU指令的并行数据访问以及CPU与直接存储器存取(DM A)的并行访问都可能导致冲突,片外存储器的慢速存取也会导致CPU流水线停止。根据CPU处理数据的需要分配片上存储器,采用DM A动态地将数据转移到片上,减小存取慢速片外存储器带来的延时;充分利用CPU多条数据总线并行访问多个存储器块的能力和双端口存储器(DARAM)一个周期两次访问的能力,减小存储器带宽的限制。实验结果表明:合理分配存储器,程序执行时间最多减少了48%。存储器分配该方法简单,易于实现。

A memory allocation algorithm was developed to maximize parallel data access and make full use of CPU processing ability to improve real time performance of embedded multimedia applications. A scratch-pad memory is assigned to meet the data processing demand with data dynamically transferred to on-chip memory by Direct Memory Access （DMA） to reduce the delays in slow off-chip memory access. Parallel access to multiple memory banks by several data buses and dual access through dual Access Random Access Memory （DARAM） reduces the memory bandwidth limit. Test results show that program run time is reduced by up to 48% using proper memory allocation. The memory assignment method is also easy to implement.