cavemanloverboy / cdugks

Implementation of CDUGKS

Massively Parallel Coupled Discrete Unified Gas Kinetic Scheme

Welcome to the MP-CDUGKS github repository. MP-CDUGKS is written in the Regent language, which uses the Legion Runtime System. I recommend using the control_replication branch, which as of writing this has better one-node performance for this code with the -dm:exact runtime flag.

The Regent implementation of CDUGKS (Liu et. al. 2018) can be found in regentsrc/. A nonparallelized version was written (long ago, and is not tested extensively) in C++ and can be found in src/.

Refer to the Legion repository for instructions on how to build the runtime system.

Quick Start Guide

There are O(~10) test problems currently implemented, with the testProblem id

1. Sod Shock Tube
2. Kelvin-Helmholtz (Sharp Interface, wavelength = 1)
3. Uniform Shearing Box
4. Ramped Kelvin-Helmholtz
5. Nonuniform Shearing Box
6. Maxwellian Relaxation (Nonstandard problem: Two blobs moving toward each other. Sine Wave Collapse is a better continuous/periodic version of this problem)
7. Cloud Crushing (Blob) Test. Note: 2D; no elongated box is used.
8. Thermoacoustic Wave
9. Gresho Vortex
10. Sine Wave Collapse

To run one of these problems, run path/to/regent/executable/regent.py Main.rg -p testProblem -c <subregions> -ll:cpu <cores/node> -ll:csize <mem/node>. It is recommended that subregions be equal to 2x the number of compute cores used.

If using the control_replication branch, also add the -dm:exact flag, which instructs the default mapper to map exact regions to cores when only using one node. Refer to the Legion Documentation for more information regarding the Machine Configuration and Runtime flags.

The conserved variables will be output at every timestep to the relative Data/ path unless the output boolean -o 1 (default) is set to zero. When the phase distribution flag -z 0 (default) is set to 1, the distribution function g will be output at every timestep. Warning: this is very I/O intensive and will take up a lot of disk space, especially for 2D problems.

Adding Test Problems

To add test problems, you will first need to set simulation parameters in the task TestProblem. Then, you will need to specify the initial conditions in InitializeW. For Non-Maxwellian initializations, you will need to modify InitializeGrid.

No external initial conditions are currently implemented, but it is certainly on the to-do list.

Planned Features

1. Comprehensive Unit Testing
2. HDF5 I/O
3. Support for externally produced initial conditions
4. Simple cooling models & Thermal Instability problem