User guide to Roundabout
Roundabout is a GPU-native ocean solver written in modern Fortran. It solves the hydrostatic, Boussinesq primitive equations in layer form on a structured Arakawa C-grid, with continuity-PPM layer transport, a split-explicit time integration that sub-cycles the fast barotropic mode, and ALE vertical coordinates. It targets regional and global hydrostatic configurations down to submesoscale-resolving resolution.
Two things distinguish it. The first is that layer thickness is a
prognostic variable: continuity is a transport equation rather than a
constraint, so there is no Poisson solve and no elliptic stage anywhere in
the dynamical core. The second is that the parallelism is expressed in
standard Fortran — all data-parallel loops are do concurrent, with
OpenACC used only for data movement and reductions — so one source builds
for NVIDIA GPUs, for CPU multicore, and for plain serial CPU across NVHPC,
gfortran and ifx.
Roundabout is named for Canberra, the one large Australian city with no ocean, and its roundabouts, which look a lot like eddies.
The numerical methods in Roundabout were developed at NOAA’s Geophysical
Fluid Dynamics Laboratory (NOAA-GFDL). Roundabout adapts them to GPUs. The
ocean dynamical core follows MOM6 and
the sea-ice component follows SIS2.
Roundabout’s own work is rewriting those methods around do concurrent and
OpenACC. Full credit for the numerics goes to the MOM6 and SIS2 developers.
MOM6 is also Roundabout’s physics oracle: where Roundabout implements an
algorithm MOM6 also has, MOM6’s behaviour is the reference it is checked
against, and deliberate divergences are documented. MOM6 and SIS2 are
licensed under the Apache License 2.0, and the ported code is used and
modified under it. See Acknowledgements and references for the full acknowledgement,
licensing and the papers the physics comes from.
Roundabout is a hobby project, written for fun and as the author’s learning project. The author is part of MOM6-GPU, the effort to port MOM6 to GPUs. Roundabout runs in parallel to, and separately from, that work and the GPU porting efforts at NOAA-GFDL and ACCESS-NRI, and it is not part of them.
Authorship. Most of Roundabout’s code was written by Claude, Anthropic’s AI model, working through Claude Code. Claude was chosen simply because it is what the author had and pays for. The baseline design is the author’s: the architecture, the API, the memory model and the parallelism model, all designed by Jorge Luis Gálvez Vallejo.
Roundabout makes no claim to be a novel implementation of a general circulation model. It is an LLM-driven, human-supervised experiment in taking a general circulation model to GPUs.
In the author’s words:
I (Jorge) acknowledge that no work here is claiming to be novel. It is a porting endeavor, with four aims:
to explore the role of agents in GPU porting;
to show Fortran as a viable, modern language for GPU-accelerated, portable scientific code;
to provide example codebases in modern Fortran for LFortran compiler development;
to support the wider Fortran community.
If a reference is missing, that is not malicious in any way, shape or form. It is human error, and it will be addressed when someone brings it up.
Because of this, the licensing of this code is MIT.
The parts ported from MOM6 and SIS2 are used and modified under the Apache
License 2.0, as set out in NOTICE.
The API documentation for the code itself — per-module and per-procedure reference generated from the source — is hosted separately: https://jorgeg94.github.io/roundabout/