kszenes/dg_code

Discontinuous Galerkin solver in cartesian and spherical geometry

DG_code

Overview

This repository contains a Discontinuous Galerkin (DG) solver for various consevations laws that I implemented during my 4-month internship at the Swiss Supercomputing Center (CSCS). The solver makes use of GT4Py, a Domain-Specific-Language (DSL) for weather and climate simulations, to produce a high-performance implementation. GT4Py provides the user with a high-level Python interface for stencil computations and generates high-performance implementations for a target architecture, including CPUs and GPUs.

This work builds upon a previous intenship at CSCS conducted by Niccolo Discacciati in 2019. In particular, his Matlab code has been adapted and used as a reference implementation.

The repository has the following filestructure:

DG_code
├── docs                   <-- Report containing documentation and benchmarks
├── gt4py                  <-- GT4Py implementation
    ├── linear_advection   <-- Linear Advection Solver
    ├── sph_swe_3d         <-- Shallow Water Equation on the Sphere Solver
    ├── swe_3d             <-- Planar Shallow Water Equation Solver
└── matlab                 <-- Reference Matlab implementation

Documentation

• All the directories are complemented with dedicated README files.
• In addition, my final report contains the mathematical theory behind the DG method, implementation details as well as performance benchmarks carried out on the Piz Daint supercomputer.

Installation

• First install GT4Py. Currently, this must be done through my personal fork of GT4Py however it will be merged into the production code soon.
• Then install the additional dependencies contained in the requirements.txt file using pip: pip install -r requirements.txt.