A fuzzy approach to automatic data locality optimization
Claudia S. Leopold · 1996
d i a ~m i n e t .u n i -j e n a .d e K e y w o r d s : fuzzy relations, fuzzy clustering, locality, m e m o r y hierarchies, compilation A B S T R A C T Many programs, both in sequential and parallel computation, can be significantly speeded up by increasing their degree of locality, i.e., by storing data that are used together in the same block, and by ordering the statements to mazimize reuse of local data.Based on the framework of a two-level memory in sequential computation, this paper suggests an approach to automatic locality optimization that operates at the level of statement instances, and can handle irregularity.The approach combines conservative data dependency constraints with fuzzy knowledge on the relatedness of individual data and operations.The optimization is an iterative process, where data distribution and statement ordering a r e alternatingly refined, and intermediate results are fuzzy.