Efficient Verification of First Tangled Races to Occur in Programs with Nested Parallelism

Abstract

Since data races result in unintended nondeterministic executions of the programs, detecting the races is important for debugging shared memory programs with nested parallelism. Particularly, the first races to occur in an execution of a program must be detected, because they can potentially affect other races that occur later. Previous on-the-fly techniques are inefficient or can not guarantee to verify the existence of the first tangle of data races to occur. This paper presents an efficient two-pass on-the-fly technique in such a kind of programs. This technique is still efficient with regard to the execution time and memory space by keeping a constant number of accesses involved in the first races for each shared variable during the execution. We empirically compare our technique with previous techniques using a set of synthetic programs with OpenMP directives.