Detecting bank conflict of GPU programs using symbolic execution

Abstract

Graphics Processing Units (GPUs) are gathering attention now and used in various fields. Accordingly, the demand for verification of GPU Programs is increasing. In this paper, we describe a method to detect bank conflict by using symbolic execution. Bank conflict is the bug happening when multiple processing units access the same bank in the shared memory at the same time. Symbolic execution is a means of analyzing programs with symbolic values. We implement a prototype model checker to detect bank conflict and evaluate it by tests in GkleeTests [4].