Indonesia Conference Directory

Membrane computing models are bio-inspired and used as intelligent algorithms. To exploit their inherent parallelism, previous studies have mapped one membrane to one thread block of a graphics processing unit (GPU). However, these previous approaches have not addressed the issue of GPU occupancy and the rate of the communication between thread blocks, which is a time consuming process. Here, we present a new classification algorithm to organize dependent objects and membranes based on the thread block communication rate and assign them to sub-matrices for execution by those same threads and thread blocks. This reduces the amount of communication required between threads and thread blocks, allowing the GPUs to maintain the highest possible occupancy. Our results indicate that for 48 objects per membrane, this algorithm facilitates a 93-fold increase in processing speed compared with a 1.6-fold increase associated with previous algorithms.