Abstract
Purchase PDF (606 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (0)
Purchase PDF (748 K)
doi:10.1016/0020-0190(95)00155-6 
Copyright © 1995 Published by Elsevier Science B.V.
Deadlock-free absorption of barrier synchronisations*1
Received 9 November 1994;
References and further reading may be available for this article. To view references and further reading you must purchase this article.
Abstract
This paper validates and generalises an optimisation arising in the compilation of while loops in data-parallel languages for MIMD shared memory architectures. Dependencies in the source program force the introduction of global synchronisations in the target MIMD program; yet, an efficient compilation must decrease the number of these costly operations. In this context, Hatcher and Quinn have proposed an optimisation that consists in splitting the original loop in two consecutive loops: a computation loop without additional control dependencies and a waiting loop to ensure global termination. We prove its correctness in the axiomatic semantics of Owicki and Gries. We observe that the loops are independent and conclude that this property is more generally applicable than suggested by Hatcher and Quinn's method.
Author Keywords: Compilers; Parallel processing; Program correctness
