Introducing a parallel cache oblivious blocking approach for the lattice Boltzmann method (Q929255)

Summary: We propose a parallel cache oblivious spatial and temporal blocking algorithm for the lattice Boltzmann method in three spatial dimensions. The algorithm has originally been proposed by \textit{M. Frigo} et al. [ACM Trans. Algorithms 8, No. 1, Paper No. 4, 22 p. (2012; Zbl 1295.68236)] and divides the space-time domain of stencil-based methods in an optimal way, independently of any external parameters, e.g., cache size. In view of the increasing gap between processor speed and memory performance this approach offers a promising path to increase cache utilisation. We find that even a straightforward cache oblivious implementation can reduce memory traffic at least by a factor of two if compared to a highly optimised standard kernel and improves scalability for shared memory parallelisation. Due to the recursive structure of the algorithm we use an unconventional parallelisation scheme based on task queuing.