摘要
为了提高嵌入式多媒体应用的运行速度并降低功耗,本文提出一种高效利用片上存储器的方法.将数据矩阵划分成合理大小的子块,分阶段地将数据子块转移到片上,并尽可能重复利用已经转移到片上的数据,以便有效地减少片外存储器与片上存储器之间的数据转移.通过对汇编语言中存储器阵操作数适当分配,避免读写数据延迟.根据汇编语言代码写出不产生流水线停滞的各个矩阵操作数的存储器位置限制条件,根据限制条件,本文提出求解矩阵分配的方法.
To improve embedded multimedia application performance and reduce power consumption, an efficient method of exploiting scratchpad memory is proposed. Big data array are divided in small tiles and copied to on-chip memory by stages. Data in scratchpad memory should be reused as much as possible to reduce data transfer between off-chip memory and scratchpad memory. Memory access delay can be avoided by proper assignment of memory operand in assemble instructions. Some operand memory constraint conditions can be listed out to minimize oioeline stall. A method of array assignment is proposed in these conditions.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2005年第11期1937-1940,共4页
Acta Electronica Sinica
基金
国家自然科学基金(No.60272020)