Abstract
We consider adjoint checkpointing strategies that minimize the number of recomputations needed when using multistage timestepping. We demonstrate that we can improve on the seminal work based on the Revolve algorithm. The new approach provides better performance for a small number of time steps or checkpointing storage. Numerical results illustrate that the proposed algorithm can deliver up to two times speedup compared with that of Revolve and avoid recomputation completely when there is sufficient memory for checkpointing. Moreover, we discuss a tailored implementation that is arguably better suited for mature scientific computing libraries by avoiding central control assumed in the original checkpointing strategy. The proposed algorithm has been included in the PETSc library.
This work was supported in part by the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research, Scientific Discovery through Advanced Computing (SciDAC) program through the FASTMath Institute under Contract DE-AC02-06CH11357 at Argonne National Laboratory.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Aupy, G., Herrmann, J.: Periodicity in optimal hierarchical checkpointing schemes for adjoint computations. Optim. Methods Softw. 32(3), 594–624 (2017). https://doi.org/10.1080/10556788.2016.1230612
Aupy, G., Herrmann, J., Hovland, P., Robert, Y.: Optimal multistage algorithm for adjoint computation. SIAM J. Sci. Comput. 38(3), C232–C255 (2016)
Balay, S., et al.: PETSc Users Manual. Technical report ANL-95/11 - Revision 3.7, Argonne National Laboratory (2016)
Griewank, A., Walther, A.: Algorithm 799: revolve: an implementation of checkpointing for the reverse or adjoint mode of computational differentiation. ACM Trans. Math. Softw. 26(1), 19–45 (2000). https://doi.org/10.1145/347837.347846
Nagel, W.E., Walter, W.V., Lehner, W. (eds.): Euro-Par 2006. LNCS, vol. 4128. Springer, Heidelberg (2006). https://doi.org/10.1007/11823285
Hundsdorfer, W., Ruuth, S.J.: IMEX extensions of linear multistep methods with general monotonicity and boundedness properties. J. Comput. Phys. 225(2007), 2016–2042 (2007). https://doi.org/10.1016/j.jcp.2007.03.003
Schanen, M., Marin, O., Zhang, H., Anitescu, M.: Asynchronous two-level checkpointing scheme for large-scale adjoints in the spectral-element solver Nek5000. Procedia Comput. Sci. 80, 1147–1158 (2016). https://doi.org/10.1016/j.procs.2016.05.444
Stumm, P., Walther, A.: MultiStage approaches for optimal offline checkpointing. SIAM J. Sci. Comput. 31(3), 1946–1967 (2009). https://doi.org/10.1137/080718036
Stumm, P., Walther, A.: New algorithms for optimal online checkpointing. SIAM J. Sci. Comput. 32(2), 836–854 (2010). https://doi.org/10.1137/080742439
Wang, Q., Moin, P., Iaccarino, G.: Minimal repetition dynamic checkpointing algorithm for unsteady adjoint calculation. SIAM J. Sci. Comput. 31(4), 2549–2567 (2009). https://doi.org/10.1137/080727890
Zhang, H., Constantinescu, E.M., Smith, B.F.: PETSc TSAdjoint: a discrete adjoint ODE solver for first-order and second-order sensitivity analysis. CoRR abs/1912.07696 (2020), http://arxiv.org/abs/1912.07696
Zhang, H., Sandu, A.: FATODE: a library for forward, adjoint, and tangent linear integration of ODEs. SIAM J. Sci. Comput. 36(5), C504–C523 (2014). https://doi.org/10.1137/130912335
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2021 Springer Nature Switzerland AG
About this paper
Cite this paper
Zhang, H., Constantinescu, E. (2021). Revolve-Based Adjoint Checkpointing for Multistage Time Integration. In: Paszynski, M., Kranzlmüller, D., Krzhizhanovskaya, V.V., Dongarra, J.J., Sloot, P.M.A. (eds) Computational Science – ICCS 2021. ICCS 2021. Lecture Notes in Computer Science(), vol 12742. Springer, Cham. https://doi.org/10.1007/978-3-030-77961-0_37
Download citation
DOI: https://doi.org/10.1007/978-3-030-77961-0_37
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-77960-3
Online ISBN: 978-3-030-77961-0
eBook Packages: Computer ScienceComputer Science (R0)