摘要
为了提高发射宽度,高端DSP普遍采用分簇结构设计.分簇结构的处理器依赖编译器在代码生成的时候指定每条指令所在的簇.针对传统分簇算法中存在参考信息太过局部的问题,提出一种依据资源压力将指令依赖图划分成多个子图,然后对子图进行分簇的算法.最后,验证了该算法能够提高分簇效率.
For more issue width, the clustered VLIW processor is a common design in advanced DSP. A clustered processor need compiler assign cluster to each instruction in code-generation. For the problem that traditional cluster assign algorithm using local reference information, the paper proposes an algorithm which divide dependence graph to multiple sub-graph first, then assign cluster to each sub-graph later. It is verified that the algorithm is effective .
出处
《微电子学与计算机》
CSCD
北大核心
2010年第8期49-52,56,共5页
Microelectronics & Computer
基金
核高基项目(2009ZX01034-001-001-002)
安徽省自然科学基金(090412068)
