This project is a study in efficient implementation of a particle simulation to run on a GPU. It was the bachelor project of Johannes Poulsen and Magnus Mouritzen.

The motivation for the project was to pave the ground for a GPU-based implementation of a PIC MCC particle simulation developed by a research group at DTU space. It explores the various scheduling paradigms given the challenges of a dynamic simulation and proposes optimised kernels for this task.

The version of the project that was submitted can be found at https://github.com/MagnusMouritzen/particle-simulation/tree/final_branch.

Information

See the report and/or slideshow from the presentation for additional information.

TO RUN THE PROGRAM

The program has been developed to run on the HPC provided by DTU. The following modules are needed to run the code:

module load cuda/12.2.2 nvhpc/23.7-nompi gcc/12.3.0-binutils-2.40

Compile

make

Run

The program run takes 8 arguments:

1. Mode (The scheduler to use. 30=Dynamic, 31=CPU Sync, 32=Naive, 33=Dynamic Old).
2. Verbose (0 for no verbosity. Another number determines how often the state of the system is printed. 1 for every mobility step, 2 for every second, etc.).
3. init n (The initial amount of particles).
4. max t (Poisson steps).
5. block size.
6. max n (Capacity for particles).
7. sleep time (Used for some barriers, but largely irrelevant).
8. poisson_timestep (Mobility steps for each poisson step).

run [Mode] [Verbose] [init n] [max t] [block size] [max n] [sleep time] [poisson_timestep]