Abstract
Batched iterative solvers can be an attractive alternative to batched direct solvers if the linear systems allow for fast convergence. In non-batched settings, iterative solvers are often enhanced with sophisticated preconditioners to improve convergence. In this paper, we develop preconditioners for batched iterative solvers that improve the iterative solver convergence without incurring detrimental resource overhead and preserving much of the iterative solver flexibility. We detail the design and implementation considerations, present a user-friendly interface to the batched preconditioners, and demonstrate the convergence and runtime benefits over non-preconditioned batched iterative solvers on state-of-the-art GPUs for a variety of benchmark problems from finite difference stencil matrices, the Suitesparse matrix collection and a computational chemistry application.
This research was supported by the Exascale Computing Project (17-SC-20-SC), a collaborative effort of the U.S. Department of Energy Office of Science and the National Nuclear Security Administration.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Ginkgo features also batched versions of other Krylov solvers, BiCGStab is however the most lightweight Krylov solver for general problems available.
- 2.
This work was performed on the HoreKa supercomputer funded by the Ministry of Science, Research and the Arts Baden-Württemberg and by the Federal Ministry of Education and Research.
References
Abdelfattah, A., et al.: A set of batched basic linear algebra subprograms and LAPACK routines. ACM Trans. Math. Softw. 47(3), 21:1–21:23 (2021)
Aggarwal, I., Kashi, A., Nayak, P., Balos, C.J., Woodward, C.S., Anzt, H.: Batched sparse iterative solvers for computational chemistry simulations on GPUs. In: 2021 12th Workshop on Latest Advances in Scalable Algorithms for Large-Scale Systems (ScalA), pp. 35–43 (2021)
Anzt, H., Collins, G., Dongarra, J., Flegar, G., Quintana-Ortí, E.S.: Flexible batched sparse matrix-vector product on GPUs. In: Proceedings of the 8th Workshop on Latest Advances in Scalable Algorithms for Large-Scale Systems, ScalA 2017, pp. 1–8, New York, NY, USA, November 2017. Association for Computing Machinery (2017)
Anzt, H., Dongarra, J., Flegar, G., Quintana-Ortí, E.S.: Batched Gauss-Jordan elimination for Block-Jacobi preconditioner generation on GPUs. In: Proceedings of the 8th International Workshop on Programming Models and Applications for Multicores and Manycores, PMAM 2017, pp. 1–10, New York, NY, USA, February 2017. Association for Computing Machinery (2017)
Balos, C.J., Gardner, D.J., Woodward, C.S., Reynolds, D.R.: Enabling GPU accelerated computing in the SUNDIALS time integration library. Parallel Comput. 108, 102836 (2021)
Carroll, E., Gloster, A., Bustamante, M.D. and Náraigh, L.Ó.: A batched GPU methodology for numerical solutions of partial differential equations. arXiv, 2107.05395 (2021)
Chow, E., Patel, A.: Fine-grained parallel incomplete LU factorization. SIAM J. Sci. Comput. 37(2), C169–C193 (2015). Publisher: Society for Industrial and Applied Mathematics
Davis, T.A., Hu, Y.: The university of Florida sparse matrix collection. ACM Trans. Math. Softw. 38(1), 1:1–1:25 (2011)
Doi, S.: On parallelism and convergence of incomplete LU factorizations. Appl. Numer. Math. 7(5), 417–436 (1991)
Dong, T., Haidar, A., Luszczek, P., Harris, J.A., Tomov, S., Dongarra, J.: LU factorization of small matrices: accelerating batched DGETRF on the GPU. In: 2014 IEEE International Conference on High Performance Computing and Communications, 2014 IEEE 6th International Symposium on Cyberspace Safety and Security, 2014 IEEE 11th International Conference on Embedded Software and System (HPCC, CSS, ICESS), pp. 157–160, Paris, France, August 2014. IEEE
Dongarra, J., et al.: A proposed API for batched basic linear algebra subprograms, April 2016. https://eprints.ma.man.ac.uk/2464/
Evstigneev, N.M., Ryabkov, O.I., Tsatsorin, E.A.: On the inversion of multiple matrices on GPU in batched mode. Supercomput. Front. Innov. Int. J. 5(2), 23–42 (2018)
Gloster, A., Náraigh, L.Ó, Khang, E.P.: cupentbatch-a batched pentadiagonal solver for NVIDIA GPUs. Comput. Phys. Commun. 241, 113–121 (2019)
Göbel, F., Grützmacher, T., Ribizel, T., Anzt, H.: Mixed precision incomplete and factorized sparse approximate inverse preconditioning on GPUs. In: Sousa, L., Roma, N., Tomás, P. (eds.) Euro-Par 2021. LNCS, vol. 12820, pp. 550–564. Springer, Cham (2021). https://doi.org/10.1007/978-3-030-85665-6_34
Hager, R., Yoon, E.S., Ku, S., D’Azevedo, E.F., Worley, P.H., Chang, C.S.: A fully non-linear multi-species Fokker-Planck-Landau collision operator for simulation of fusion plasma. J. Comput. Phys. 315, 644–660 (2016)
Kashi, A., Nayak, P., Kulkarni, D., Scheinberg, A., Lin, P., Anzt, H.: Batched sparse iterative solvers on GPU for the collision operator for fusion plasma simulations. In: 2022 IEEE International Parallel and Distributed Processing Symposium (IPDPS) (2022)
Liu, W., Li, A., Hogg, J., Duff, I.S., Vinter, B.: A synchronization-free algorithm for parallel sparse triangular solves. In: Dutot, P.-F., Trystram, D. (eds.) Euro-Par 2016. LNCS, vol. 9833, pp. 617–630. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-43659-3_45
NVIDIA: cuSOLVER–gpu accelerated library for decompositions and linear system solutions on NVIDIA GPUs. https://docs.nvidia.com/cuda/cusolver/index.html. Accessed 24 Aug 2021
Valero-Lara, P., Martínez-Pérez, I., Sirvent, R., Martorell, X., Peña, A.J.: cuThomasBatch and cuThomasVBatch, CUDA routines to compute batch of tridiagonal systems on NVIDIA GPUs. Concurrency Comput. Pract. Exp. 30, e4909 (2018). https://doi.org/10.1002/cpe.4909
Van der Vorst, H.A.: Bi-CGSTAB: a fast and smoothly converging variant of Bi-CG for the solution of nonsymmetric linear systems. SIAM J. Sci. Stat. Comput. 13(2), 631–644 (1992). Publisher: Society for Industrial and Applied Mathematics
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Aggarwal, I., Nayak, P., Kashi, A., Anzt, H. (2022). Preconditioners for Batched Iterative Linear Solvers on GPUs. In: Doug, K., Al, G., Pophale, S., Liu, H., Parete-Koon, S. (eds) Accelerating Science and Engineering Discoveries Through Integrated Research Infrastructure for Experiment, Big Data, Modeling and Simulation. SMC 2022. Communications in Computer and Information Science, vol 1690. Springer, Cham. https://doi.org/10.1007/978-3-031-23606-8_3
Download citation
DOI: https://doi.org/10.1007/978-3-031-23606-8_3
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-23605-1
Online ISBN: 978-3-031-23606-8
eBook Packages: Computer ScienceComputer Science (R0)