Space-efficient first race detection in shared memory programs with nested parallelism

Data races are errors caused by uncoordinated access in parallel programs, resulting in non-deterministic program execution. Therefore, the main focus of the current paper is to create efficient on-the-fly detection of data races based on minimizing the shared data structures, thereby reducing the space overhead required to maintain the access history and concurrency information during an execution. Accordingly, a space efficient method is proposed for detecting first races, since their detection can eliminate other races. To reduce the storage requirements, the proposed method uses a sequential monitoring technique and decomposition tree, checks the logical concurrency among the threads, and examines the root events. The programs considered in this paper have a series-parallel graph with a fork-join. The resulting space complexity is O(VN), where V is the number of shared variables, T is the maximum parallelism, and N is the nesting depth of the program.