SSPRK33 Subcycling with the Fortran Interface

[Andris-R]

Dear AMReX developers,

I would like to implement SSPRK33 with subcycling using only the Fortran interface. Does the Fortran interface provide access to the functionality needed to reproduce the FillPatcher RK workflow, specifically fillRK and storeRKCoarseData?

If so, could you point me to the corresponding Fortran procedures? I am particularly interested in whether amrex_fillpatch and amrex_fillcoarsepatch can be used for this purpose, or whether a different approach is required.

Thank you.

[WeiqunZhang]

We don't have the Fortran interface for it. We could try to add it.

[Andris-R]

The addition would be very welcome. Regarding the deployment timeline, are we talking about weeks or months? What would be your conservative estimate?

Going back to my original question, it seems that the functionality provided by storeRKCoarseData and fillRK should be reproducible on the application side using the current Fortran interface. As I understand it, the algorithm consists of three main steps:

1. Define additional MultiFabs to store the coarse-level state u0 and the stage fluxes k1, k2, and k3.
2. Perform temporal interpolation on the coarse level to reconstruct u(lev - 1) at the fine-level time using u0, k1, k2, and k3.
3. Use the reconstructed u(lev - 1) to fill the guard cells required for advancing u(lev).

What gives me pause is that fillRK calls FillPatchInterp (see code snippet below), which is not exposed through the Fortran interface. As far as I can tell, the only available alternative is to use amrex_fillpatch_two.

Could you comment on the viability of this approach? In particular, would amrex_fillpatch_two be sufficient to reproduce the behavior of FillPatchInterp in this context, or am I overlooking an important detail?

    FillPatchInterp(*m_cf_fine_data, 0, *m_cf_crse_data_tmp, 0,
                    m_ncomp, IntVect(0), m_cgeom, m_fgeom,
                    amrex::grow(amrex::convert(m_fgeom.Domain(),
                                               mf.ixType()),m_nghost),
                    m_ratio, m_interp, bcs, 0);

[WeiqunZhang]

Give #5486 a try.

[Andris-R]

Thank you for the very quick implementation and for adding this functionality.

The lack of FillPatcher support in the Fortran interface had become a significant source of uncertainty, and I am glad that it was resolved. I will give the implementation a try as soon as I can, though it will take at least a couple of weeks until my application is stable enough.

What are the next steps for getting this pull request merged into the main development branch and eventually included in a tagged release?

Thanks again.

[WeiqunZhang]

If a capability exists in the C++ side, but is not yet available through the Fortran interface, contributions from the community to add it are welcome.

[Andris-R]

I'll gladly contribute where I can. My knowledge of Fortran, including its more recent standards, is quite strong, whereas my C++ experience is more limited.

Since I'm still fairly new to AMReX, I can realistically only contribute to the existing Fortran interface for now. The FillPatcher addition, for example, was beyond what I could have realistically implemented myself.

On a related note, how many users does the Fortran interface currently have?