Solving quadratic assignment problem using iterated local search on GPU spatial memory.

The quadratic assignment problem is one of the most studied combinatorial optimization problems. Although many direct and heuristics methods exist for solving this problem, it takes a huge amount of time. Metaheuristics give an approximate solution in a feasible time. Therefore, we used a highly parallel iterated local search metaheuristic on a massively parallel graphics processing unit (GPU) to reduce the execution time. We utilized GPU spatial memory-global, local, constant, texture, and shared memory properties to further reduce the execution time. In this study, we found that using efficient utilization of GPU memory properties, and mixing different memory together we obtained better results from using only global memory. [ABSTRACT FROM AUTHOR]