Soundness of data flow analyses for weak memory models.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
, 7078 L
Modern multi-core microprocessors implement weak memory consistency models; programming for these architectures is a challenge. This paper solves a problem open for ten years, and originally posed by Rinard: we identify sufficient conditions for a data flow analysis to be sound w.r.t. weak memory models. We first identify a class of analyses that are sound, and provide a formal proof of soundness at the level of trace semantics. Then we discuss how analyses unsound with respect to weak memory models can be repaired via a fixed point iteration, and provide experimental data on the runtime overhead of this method. © 2011 Springer-Verlag.
|Title:||Soundness of data flow analyses for weak memory models|
|UCL classification:||UCL > School of BEAMS
UCL > School of BEAMS > Faculty of Engineering Science
Archive Staff Only