A comparative study of GPU verify and GKLEE
Anmol Panda, Philipp Rümmer, Neena Goveas · 2016
Use of Graphics Processing Unit (GPU) software is increasing due to the need for data intensive operations and availability of GPUs. This has led to a need for effective GPU software verification tools. These tools have to satisfy requirements such as accuracy, reliability and ease of use. In this work, we have considered two such tools: GPU Verify and GKLEE. Our objectives were to learn about the common challenges developers faced in GPU programming, to understand the specific bugs that these two tools report and compare their scope and scalability aspects. We have also considered usability and learn ability aspects. In order to test the software, twenty-six benchmarks were selected from open-source applications. These benchmarks were then verified using the tools and the results documented and analysed. The conclusions have been included in the final section.