Automatic Refactoring for Parallelization

Amrein, Christoph (2017) Automatic Refactoring for Parallelization. Masters thesis, HSR Hochschule für Technik Rapperswil.

[img]
Preview
Text
paper.pdf - Supplemental Material

Download (546kB) | Preview

Abstract

In modern software, adaptions like the parallelization are necessary to fully leverage the CPU's capabilities. However, the parallelization introduces a new range of possible software faults. Thus, assisting utilities like static code analyzers are desirable. For instance, ones that inform engineers about the code fragments that can be safely adapted. This thesis focuses on loops and introduces a conservative approach to verify that these can be parallelized. More specifically, it allows proofing that array accesses do not conflict between iterations. The procedure is a data flow analysis, which proofs the absence of conflicts by employing rules for a selection of binary expressions. Furthermore, its design allows it to profit from various code optimizations automatically. Experimental evaluations show that both, the prototype and the data flow analysis itself, do not incorrectly identify loops as parallelizable. Moreover, it pinpoints that the analysis can correctly identify most of the parallelizable loops, and only a negligible amount requires a more mature approach.

Item Type: Thesis (Masters)
Subjects: Topics > Software > Performance
Topics > Software > Optimization
Area of Application > Development Tools
Area of Application > Academic and Education
Technologies > Programming Languages > C#
Technologies > Frameworks and Libraries > .NET
Technologies > Parallel Computing
Divisions: Master of Science in Engineering (MRU Software and Systems)
Creators:
CreatorsEmail
Amrein, ChristophUNSPECIFIED
Contributors:
ContributionNameEmail
Thesis advisorBlaeser, LucUNSPECIFIED
Depositing User: Christian Spielmann
Date Deposited: 23 Apr 2019 09:22
Last Modified: 23 Apr 2019 09:22
URI: http://eprints.hsr.ch/id/eprint/760

Actions (login required)

View Item View Item