tlm_adjoint.checkpoint_schedules.binomial

Module Contents

class tlm_adjoint.checkpoint_schedules.binomial.MultistageCheckpointSchedule(max_n, snapshots_in_ram, snapshots_on_disk, *, trajectory='maximum')

A binomial checkpointing schedule using the approach described in

  • Andreas Griewank and Andrea Walther, ‘Algorithm 799: revolve: an implementation of checkpointing for the reverse or adjoint mode of computational differentiation’, ACM Transactions on Mathematical Software, 26(1), pp. 19–45, 2000, doi: 10.1145/347837.347846

hereafter referred to as GW2000.

Uses a ‘MultiStage’ distribution of checkpoints between RAM and disk equivalent to that described in

  • Philipp Stumm and Andrea Walther, ‘MultiStage approaches for optimal offline checkpointing’, SIAM Journal on Scientific Computing, 31(3), pp. 1946–1967, 2009, doi: 10.1137/080718036

The distribution between RAM and disk is determined using an initial run of the schedule.

Offline, one adjoint calculation permitted.

Parameters:

  • max_n – The number of forward steps in the initial forward calculation.

  • snapshots_in_ram – The maximum number of forward restart checkpoints to store in memory.

  • snapshots_on_disk – The maximum number of forward restart checkpoints to store on disk.

  • trajectory –

    When advancing n forward steps with s checkpointing units available there are in general multiple solutions to the problem of determining the number of forward steps to advance before storing a new forward restart checkpoint – see Fig. 4 of GW2000. This argument selects a solution:

    • ’revolve’: The standard revolve solution, as specified in the equation at the bottom of p. 34 of GW2000.

    • ’maximum’: The maximum possible number of steps, corresponding to the maximum step size compatible with the optimal region in Fig. 4 of GW2000.

The argument names snaps_in_ram and snaps_on_disk originate from the corresponding arguments for the dolfin_adjoint.solving.adj_checkpointing function in dolfin-adjoint (see e.g. version 2017.1.0).

property is_exhausted

Whether the schedule has concluded. Note that some schedules permit multiple adjoint calculation, and may never conclude.

property uses_disk_storage

Whether the schedule may use disk storage. If False then no disk storage is required.

iter()

A generator which should be overridden in derived classes in order to define a checkpointing schedule.

class tlm_adjoint.checkpoint_schedules.binomial.TwoLevelCheckpointSchedule(period, binomial_snapshots, *, binomial_storage='disk', binomial_trajectory='maximum')

A two-level mixed periodic/binomial checkpointing schedule using the approach described in

  • Gavin J. Pringle, Daniel C. Jones, Sudipta Goswami, Sri Hari Krishna Narayanan, and Daniel Goldberg, ‘Providing the ARCHER community with adjoint modelling tools for high-performance oceanographic and cryospheric computation’, version 1.1, EPCC, 2016

and in the supporting information for

  • D. N. Goldberg, T. A. Smith, S. H. K. Narayanan, P. Heimbach, and M. Morlighem, ‘Bathymetric influences on Antarctic ice-shelf melt rates’, Journal of Geophysical Research: Oceans, 125(11), e2020JC016370, 2020, doi: 10.1029/2020JC016370

Online, unlimited adjoint calculations permitted.

Parameters:

  • period – Forward restart checkpoints are stored to disk every period forward steps in the initial forward calculation.

  • binomial_snapshots – The maximum number of additional forward restart checkpointing units to use when advancing the adjoint between periodic disk checkpoints.

  • binomial_storage – The storage to use for the additional forward restart checkpoints generated when advancing the adjoint between periodic disk checkpoints. Either ‘RAM’ or ‘disk’.

  • binomial_trajectory – See the trajectory constructor argument for MultistageCheckpointSchedule.

property is_exhausted

Whether the schedule has concluded. Note that some schedules permit multiple adjoint calculation, and may never conclude.

property uses_disk_storage

Whether the schedule may use disk storage. If False then no disk storage is required.

iter()

A generator which should be overridden in derived classes in order to define a checkpointing schedule.