An ILP-based approach to locality optimization

Guilin Chen, Ozean Ozturk, Mahmut Kandemir

Research output: Contribution to journalConference article

1 Scopus citations

Abstract

One of the most important factors that determine performance of data-intensive applications is data locality. A program with high data locality makes better use of fast, on-chip memories and can avoid large main memory latencies. Although previous compiler research investigated numerous techniques for enhancing locality, we lack of formal techniques, against which the existing heuristics can be compared. Motivated by this observation, this paper presents a fresh look at locality optimization based on integer linear programming (ILP). We formulate the conditions for data locality, and present a system of constraints whose solution gives optimal computation re-ordering and data-to-memory assignment under our objective function and cost model. Our experimental results using three data-intensive applications clearly indicate that the ILP-based approach generates very good results and outperforms a previously proposed heuristic solution to locality.

Original languageEnglish (US)
Pages (from-to)149-163
Number of pages15
JournalLecture Notes in Computer Science
Volume3602
StatePublished - Oct 19 2005
Event17th International Workshop on Languages and Compilers for High Performance Computing, LCPC 2004 - West Lafayette, IN, United States
Duration: Sep 22 2004Sep 24 2004

    Fingerprint

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this