FastFlow: GPU Acceleration of Flow and Depression Routing for Landscape Simulation

Topographic variation does not have a significant impact on the performance of flow routing, so we benchmark flow routing solely on real terrains. We compare our algorithm against:

• • Barnes [Bar19], who provides a GPU algorithm for flow routing based on a pre-ordering of the computation graph.
• • Schott et al. [SPF⁺23], who use a GPU algorithm similar to ours, but without pointer jumping, which results in as many iterations as the node length of the longest river in the terrain ($\backslash sim\; \backslash sqrt\{n\}$).

Note that neither of these algorithms provides a solution for depression routing on the GPU. The results are illustrated in Figure 18, with an average speed-up of $10\backslash times$, $5\backslash times$, $3\backslash times$ and $3\backslash times$ against [Bar19] for each target resolution, respectively.

Figure 18: Performance of our flow routing against [Bar19] and [SPF⁺23].