Melvin is a 2D, pseudo-spectral, adaptive-time implementation of
- Rayleigh-Benard convection (RBC)
- Double-diffusive convection (DDC)
The code has been parallelised on the CPU via OpenMP and FFTW and on the GPU via CUDA and cuFFT. Not all features and boundary conditions are currently implemented.
Melvin Stern was a pioneer in oceanographic fluid dynamics, particularly in the study of salt-fingering. This code is a personal monument to a great scientist.
- FFTW 3
- gcc with OpenMP support (tested with gcc 6 and 9)
Identical dependencies as CPU version plus:
- CUDA development and runtime libraries
The various run scripts available in run_scripts will automatically build and run the simulation. They are currently set up to run a number of different cases including
- Rayleigh-Benard convection
- Double-diffusive salt-fingering with and without periodic boundary conditions
- Rayleigh-Taylor instability
These scripts may be used to extend the simulations to other parameter choices, boundary conditions and initial conditions.
For manual building, running make will automatically build the OpenMP enabled CPU version of the code. make profile will include profiling information in the binary while make debug will enable assertions and include symbols in the binary.
To compile for GPU run make gpu. There is a corresponding make gpu-debug for debuggin purposes.
Currently the regression tests available are:
- Linear:
- Standard Rayleigh-Benard convection
- Salt-fingering
- Nonlinear:
- Standard Rayleigh-Benard convection
- Combined compositional + thermal convection
- Salt-fingering
The linear tests will run two linear simulations either side of the analytically calculated critical Rayleigh number.
The nonlinear RBC test will run to completion and compare the output to the known test found in Glatzmaier, page 46.
All tests can be found in the test folder.
The unit-testing framework currently used is the header-only library Catch.
To run the unit tests, run make test. The GPU tests can be run with make gpu-test.
For RBC only there is periodic and impermeable boundary conditions available in either direction. Due to constraints on the FFT algorithm, the periodic boundary conditions are faster for most resolutions.
Due to the nature of the fast-Fourier transform algorithm, the running performance is best when the horizontal real-space resolution is a factor of small primes. Powers of two are preferrable.
Glatzmaier: Introduction to Modeling Convection in Stars and Planets; Gary A. Glatzmaier; 2014