Interpolating from checkpoint

[jagoosw]

I've been running some simulations where I've been spinning up at lower resolution, checkpointing, and then interpolating to a higher resolution grid. To do this I wrote this function:

function interpolate_from_checkpoint!(model, checkpoint_path; field_names = propertynames(prognostic_fields(model)))
    model_fields = prognostic_fields(model)

    jldopen(checkpoint_path, "r") do file
        checkpoint_grid = file["grid"]

        for field_name in field_names
            field_info = file["$(field_name)"]

            checkpoint_field = Field{typeof.(field_info["location"])...}(checkpoint_grid, data = field_info["data"], indices = field_info["indices"], boundary_conditions = field_info["boundary_conditions"])
            
            interpolate!(model_fields[field_name], on_architecture(architecture(model), checkpoint_field))
        end

        @info "Interpolated prognostic variables in model $(size(model.grid)) to checkpoint at $(prettytime(file["clock"].time)) $(size(checkpoint_grid))"
    end

    return nothing
end

This seems like something that would be generally useful, should I add it to the source code?

I also wrote a shortcut:

checkpoint!(model; checkpointer = Checkpointer(model; schedule = IterationInterval(1), verbose = true)) =
    write_output!(checkpointer, model)

that is more convenient than adding the checkpointer to the simulation, instead I can just write checkpoint!(model) after the run!.

[glwagner]

I feel it might make more sense for Checkpointer to use checkpoint! rather than the other way around.