Approach to Data Race Detection Based on Petri Nets with Additional Semantic Relations

Abstract

One of the most common and poorly detectable errors in parallel algorithms is data race condition. The article proposes an approach to detecting such states of data races based on simulation of a program using the mathematical apparatus of Petri nets with additional semantic relations. Based on chains of semantic relations between places belonging to different parallel threads, memory allocation places and other intermediate places, it is possible to detect data races associated with incorrect organization of access to a shared resource.