Parallelization of Adaptive Mesh Refinement Using OpenMP and CUDA

• Ronan A. Pereira Jr UFSCar
• Ayla Y. Nishino UFSCar
• Vinícius R. Kury UFSCar
• Hélio C. Guardia UFSCar

DOI: https://doi.org/10.5753/eradsp.2026.30744

Resumo

This work presents parallel implementations of Adaptive Mesh Refinement (AMR) using OpenMP and CUDA. AMR dynamically refines mesh resolution in regions requiring higher precision, optimizing computational resources. We compare CPU parallelization with OpenMP and GPU acceleration with CUDA against a sequential baseline. Results show that OpenMP achieves super-linear scaling up to 32 threads (32.40× speedup, 101.3% efficiency) on a 40-core Intel Xeon E7-4870 architecture distributed across 4 sockets, peaking at a 54.40× speedup with 256 overcommitted threads due to cumulative L3 cache size and aggregated memory bandwidth. Consequently, the multi-socket CPU execution outperforms the CUDA implementation, which demonstrates a 28.65× speedup on a Tesla T4 GPU, remaining constrained by warp divergence and irregular memory patterns inherent to AMR.

Referências

Berger, M. J. and Colella, P. (1989). Local adaptive mesh refinement for shock hydrodynamics. Journal of Computational Physics, 82(1):64–84.

Burstedde, C., Wilcox, L. C., and Ghattas, O. (2011). p4est: Scalable algorithms for parallel adaptive mesh refinement on forests of octrees. SIAM Journal on Scientific Computing, 33(3):1103–1133.

Holke, J., Knapp, D., et al. (2021). t8code: Modular adaptive mesh refinement in the exascale era. arXiv preprint arXiv:2109.02729.

Morton, G. M. (1966). A computer oriented geodetic data base and a new technique in file sequencing. Technical report, IBM Ltd.

Qin, X., LeVeque, R. J., and Motley, M. R. (2019). Accelerating an adaptive mesh refinement code for depth-averaged flows using gpus. Journal of Advances in Modeling Earth Systems, 11(8):2606–2629.