Investigating performance variations of an optimized GPU-ported granulometry algorithm