Parallelism for free

Abstract

In this paper we s h o w h o w to construct optimal bitvector analysis algorithms for parallel programs with shared memory that are as e cient as their purely sequential counterparts, and which can easily be implemented. Whereas the complexity result is rather obvious, our optimality result is a consequence of a new Kam/Ullman-style Coincidence Theorem. Thus, the important merits of sequential bitvector analyses survive the introduction of parallel statements.

Type Systems Based Data Race Detector

“…Analyzing may concentrate on whole program, or focuses on each program point. There are many aspects of analyzing multi-threaded programs: pointer analysis (El-Zawawy, 2012a;El-Zawawy, 2011;El-Zawawy, 2011a;El-Zawawy, 2011b;Rugina & Rinard, 2003), optimization uses (El-Zawawy & Nayel, 2011;Knoop & Steffen, 1999;Knoop, Steffen, & Vollmer, 1996;Lee, Midkiff, & Padua, 1998;Lee, Padua, & Midkiff, 1999), data race detection (Cheng, Feng, Leiserson, Randall, & Stark, 1998;Rugina & Rinard, 2005 ), and deadlock detection (Blieberger, Burgstaller, & Scholz, 2000;Koskinen & Herlihy, 2008;Wang, Kelly, Kudlur, Lafortune, & Mahlke, 2008).…”

Soundness of Data Flow Analyses for Weak Memory Models

“…With the re-Analysis Soundness w.r.t. weak memory models Knoop et al [33] yes (separable) Chugh et al [34] yes (if no datarace) Steensgaard [35] yes (flow-insensitive) Miné [16] yes Rugina and Rinard [24] yes Jeannet [14] no Ferrara [15] yes on Java Memory Model Farzan and Kincaid [25] yes (separable) Khedker and Dhamdhere [36] separable: yes; non-separable: not in general Constant propagation [32] yes (non-relational) Fig. 6.…”

“…Therefore the approaches described by Knoop et al [33] and Farzan and Kincaid [25], who present methods of adapting a unidirectional bit-vector analysis designed with sequential programs in mind for use with multi-threaded programs, are immediately sound for weak memory models if sound for SC.…”

Static slicing of threaded programs

“…The technique to calculate the edges is beyond the scope of the papers, they can be calculated with standard algorithms [8]. A simple version would assume the existence of a boolean function parallel(i, j ) which returns true if it is possible for nodes i and j to execute in parallel (see [12] for an overview of ways to calculate this function).…”