Optimization of N-Queens Solvers on Graphics Processors

While graphics processing units (GPUs) show high performance for problems with regular structures, they do not perform well for irregular tasks due to the mismatches between irregular problem structures and SIMD-like GPU architectures. In this paper, we explore software approaches for improving the performance of irregular parallel computation on graphics processors. We propose general approaches that can eliminate the branch divergence and allow runtime load balancing. We evaluate the optimization rules and approaches with the n-queens problem benchmark. The experimental results show that the proposed approaches can substantially improve the performance of irregular computation on GPUs. These general approaches could be easily applied to many other irregular problems to improve their performance.