Home

About

Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network
FAQHELPSITE MAPCONTACT US


  Advanced Search  

 
Benchmark Design for Robust Profile-Directed Optimization Paul Berube, Jose Nelson Amaral
 

Summary: Benchmark Design for Robust Profile-Directed Optimization
Paul Berube, Jos┤e Nelson Amaral
Dept. of Computing Science, University of Alberta
Edmonton, Alberta, T6G 2E8, Canada
Abstract
Profile-guided code transformations specialize program
code according to the profile provided by execution on train-
ing data. Consequently, the performance of the code gener-
ated usind this feedback is sensitive to the selection of train-
ing data. Used in this fashion, the principle behind profile-
guided optimization techniques is the same as off-line learn-
ing commonly used in the field of machine learning. How-
ever, scant use of proper validation techniques for profile-
guided optimizations have appeared in the literature. Given
the broad use of SPEC benchmarks in the computer archi-
tecture and optimizing compiler communities, SPEC is in
a position to influence the proper evaluation and valida-
tion of profile-guided optimizations. Thus, we propose an
evaluation methodology appropriate for profile-guided op-
timization based on cross-validation. Cross-validation is a

  

Source: Amaral, JosÚ Nelson - Department of Computing Science, University of Alberta

 

Collections: Computer Technologies and Information Sciences