ParaSCAN: A Static Profiler to Help Parallelization

Date
2014-05-13
Authors
Upadhyaya, Ganesha
Rajan, Hridesh
Rajan, Hridesh
Sondag, Tyler
Journal Title
Journal ISSN
Volume Title
Publisher
Source URI
Altmetrics
Authors
Research Projects
Organizational Units
Computer Science
Organizational Unit
Journal Issue
Series
Abstract

Parallelizing software often starts by profiling to identify program paths that are worth parallelizing. Static profiling techniques, e.g. hot paths, can be used to identify parallelism opportunities for programs that lack representative inputs and in situations where dynamic techniques aren't applicable, e.g. parallelizing compilers and refactoring tools. Existing static techniques for identification of hot paths rely on path frequencies. Relying on path frequencies alone isn't sufficient for identifying parallelism opportunities. We propose a novel automated approach for static profiling that combines both path frequencies and computational weight of the paths. We apply our technique called ParaSCAN to parallelism recommendation, where it is highly effective. Our results demonstrate that ParaSCAN's recommendations cover all the parallelism manually identified by experts with 85% accuracy and in some cases also identifies parallelism missed by the experts.

Description
<p>Copyright © 2014, Ganesha Upadhyaya, Tyler Sondag, and Hridesh Rajan.</p>
Keywords
parallelism discovery and planning, software evolution
Citation
Collections