An Empirical Study of Parallelizing Test Execution Using CUDA Unified Memory and OpenMP GPU Offloading

Abstract

The execution of software testing is costly and time-consuming. To accelerate the test execution, researchers have applied several methods to run the testing in parallel. One method of parallelizing the test execution is by using a GPU to distribute test inputs among several threads running in parallel. In this paper, we investigate three programming models CUDA Unified Memory, CUDA Non-Unified Memory, and OpenMP GPU offloading to parallelize the test execution and discuss the challenges using these programming models. We use eleven benchmarks and parallelize their test suites by using these models. Our study shows some limitations (e.g. cache size, branch divergence, and load imbalance) when using GPUs to execute the testing in parallel.