On-the-fly clustering for exascale molecular dynamics simulations.

In summary, we present two algorithms for connected component labeling. The first is our Formula One implementation, which is scalable and inspired by the image processing community. It has been successfully extended to MPI-level scalability, although achieving this optimization requires considerable complexity in the code.

The second algorithm is an iterative and alternative approach that is fully scalable. Even in the worst-case scenarios, it maintains scalability from a memory perspective. This algorithm is robust, producing deterministic connected component labeling regardless of the number of MPI processes used. Furthermore, the code is concise and will soon be portable to GPU platforms.

Now that we have explored both implementations of connected component labeling, we can discuss their applications. For instance, on the left, we illustrate the impact of liquids on surfaces and the resulting ejecta or aggregate explosions.