A case study of the task-based parallel wavefront pattern

This paper analyzes the applicability of the task programming model in the parallelization of the wavefront pattern. Computations on this type of problem are characterized by a data dependency pattern across a data space, which can produce a variable number of independent tasks through the traversal of such a space. We explore several implementations of this pattern, based on the current state-of-the-art threading libraries that support tasks. For each implementation, we discuss the particularities from a programmers point of view, highlighting the advantageous features in each case. We conduct several experiments to identify the factors that can limit the performance in each implementation. Moreover, we propose and evaluate some optimizations (task recycling, prioritization of tasks based on locality hints and tiling) that the programmer can exploit to reduce the overhead in some cases.