3D multicore CPU vs GPU on sparse patterns of Sleptsov net virtual machine

A Sleptsov net is a discrete-event system, capable of universal computations, applied as a graphical language of concurrent programming for HPC and embedded domains. Search for the first fireable transition in a sequence of transitions, reordered by their priorities, represents a challenge for mass-parallel devices and a bottleneck of the virtual machine. A reduction technique for sequential search implementation on GPU, with less than logarithmic time complexity concerning the sequence length, has been developed. Together with the conventional reduction of minimum and cooperative groups, the techniques yield about ten times speed-up on CPUs and GPUs.