A Study Of Program Locality Oriented Parallel And Optimization Technology

Parallel program development is now facing with two important problem. First, along with the appearance of multiprocessor and its popularization, the requirement for the parrallel programming is more taller. Secondly, with the progress of semiconductor and manufacture techniques and the fast development of processor architectures, the speed of processors has far exceeded the speed of memories, and the gap between memory and processor speeds has continued to widen. which leads to the occurrence of Memory Wall. Actually, the traditional parallel partitioning technique can make use of characteristic of multiprocessor, but it do not consider the program locality, which causing the bad program performence.This paper reserches a set of program locality oriented parallel optimization methods which aim at a kind program of loop nest. first, we propose a new parallel partitioning technique. This technique considers the reuse of inner-thread(U-U), which is used to increase the program locality. Secondly, on the basis of above technique, we also propose a optimization technology. This technique uses the reuse of inter-thread(T-T) to increase the program locality further. Further more, we find the phenormenon of locality which called'excursion'in the storage hierarchy in our reserch.At last, the experimental results show that this set of methods can reduce the program runtime, increase the program performence observably. we show the compiler implement frame, summarize and view the future of our methods.