Identifying opportunities for automatic remote field cloning.

CASCON '04: Proceedings of the 2004 conference of the Centre for Advanced Studies on Collaborative research(2004)

引用 1|浏览42
暂无评分
摘要
Optimizing compilers can automatically re-arrange data objects to produce memory reference patterns with better reference locality while retaining correct application semantics. This paper describes a new technique, Remote Field Cloning (RFC), that consists of replicating the value from a field of a remote data object. This replication prevents long jumps in the data references of an application. RFC was motivated by work on cache-conscious re-engineering of algorithms. When hand-crafted into suitable algorithms, RFC produces speedups of up to 40% for a well-know refinement-based pathfinding problem. This paper presents a motivating example and a general problem statement of remote field cloning as an optimization problem. A new algorithm that discovers opportunities for RFC is applied to the SPEC2000 benchmark suite and discovers many such opportunities. However a correlation of the percentage of execution time spent in procedures that contain these opportunities reveals that the expected impact of RFC in the SPEC2000 suite is limited.
更多
查看译文
关键词
data reference,general problem statement,optimization problem,pathfinding problem,re-arrange data object,remote data,SPEC2000 benchmark suite,SPEC2000 suite,better reference locality,correct application semantics,Identifying opportunity,automatic remote field cloning
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要