We propose several graph partitioning algorithms for improving the performance of rigid body simulations. The algorithms operate on the graph formed by rigid bodies (nodes) and constraints (edges), producing non-overlapping and contiguous sub-systems that can be simulated in parallel by a domain decomposition technique. We demonstrate that certain partitioning algorithms reduce the computational time of the solver, and graph refinement techniques that reduce coupling between sub-systems, such as the Kernighan–Lin and Fiduccia–Mattheyses algorithms, give additional performance improvements.
BibTeX
@inproceedings{partition2022, author = {Liu, Yinchu and Andrews, Sheldon}, title = {Graph Partitioning Algorithms for Rigid Body Simulations}, booktitle = {Eurographics 2022 (Short Papers)}, year = {2022} }