|Source||CMES: Computer Modeling in Engineering & Sciences, Vol. 69, No. 2, pp. 143-166, 2010|
|Download||Full length paper in PDF format. Size = 578,920 bytes|
|Keywords||material point method, OpenMP, parallel, alternated grid updating, dynamic load balance|
|Abstract||Material point method(MPM) is a promising method in solving problems involving large deformations, especially explosion and penetration. In MPM, particles can move around the computing domain dynamically, which can result in load imbalance easily. In parallelizing MPM using OpenMP, data races will occur in the stage of grid node updating if we use loop-level parallelism for these loops. Huang et al. proposed a domain decomposition method to overcome data races [ ` 12 `
12 `$12 `&12 `#12 `^12 `_12 `%12 `~12 *phuang]. However, significant modifications of the original serial code are required. In this paper, we proposed a new alternated grid updating method to avoid data races in the stage of grid node updating, which only need small modifications of the original code and is much easier to achieve dynamic load balance. Test results show that our algorithm performs well.