An approach to solve the cache thrashing problem

摘要

Cache or local memory thrashing problem arises very often inparallel processing architectures where each processor has its localcache or memory and a write-back protocol is employed for cachecoherence. To solve the problem of large amount of data moving back andforth between the caches or local memories in different processors,techniques associated with parallel compiler need to be developed. Basedon the relations between array element accesses and enclosed loopindices in a nested parallel construct, the authors present someapproaches to reduce the data movement between the caches or localmemories for parallel programs. By analyzing the array subscriptexpressions, the compilers let the processor execute the correspondingiterations of parallel loops in terms of the data in its cache or localmemory at execution time. It benefits, particularly, the parallelprograms in which a parallel loop is enclosed by a sequential loop andarray elements are repeatedly used in different iterations of theparallel loop