Accelerated P systems with active membranes on a graphics processing unit

P systems have parallel structures; therefore, parallel tools like graphic processing units (GPU) are needed to extract this parallelism. In previous approaches each membrane was assigned to one thread block. This was inefficient when the number of objects inside each membrane was low. In the proposed approach objects of the membranes are represented as entries of a matrix. Then this matrix can be divided in sub blocks (that can include more than one membrane) to run on thread blocks of GPU efficiently. With this approach, the number of active threads in each thread block can be balanced and processing speed enhanced. For example, using previous approaches, for two objects in each membrane since one membrane is assigned to one thread block the multiprocessor occupancy is low and the speedup is 0.6 times, whereas the proposed approach multiprocessor occupancy is high and achieves a 33.7-times speedup with respect to sequential implementation.