Example CUDA project for physics 141/241

Updated Feb 4 2019, David Moore.

This code as-is runs an O(n^2) CUDA simulation on 10,000 particles. Galaxy1.txt is a galaxy of 100,000 particles, and "nSkip" is set to 10 in kernel.cu, meaning every tenth particle is taken and its mass is multiplied by ten.

Compilation instructions

To run, extract all files into a directory. Open up terminal and cd into the directory. Then type "make run". This will:

1. Create the "out" directory.
2. Compile kernel.cu into an nbody executable
3. run ./nbody, which outputs 30 .bmp files
4. use convert to put the bitmaps into a .gif

How the source code works:

~50 lines of kernel.cu are CUDA code, the rest is C++. ImageUtil.cpp and ImageUtil.h are some low-quality utilities I wrote for outputting images. I do NOT recommend that you use them for your projects, but they can be useful for debugging. It's better to plot in matlab or python like usual.

The source code file kernel.cu was written while trying to simplify three separate resources: 1. GPU Gems 3 chapter 21 (the chapter/book is available online) 2. The CUDA nbody sample project, located on the lab computers at /Developer/NVIDIA/CUDA-10.0/samples/5_Simulations/nbody/. 3. The book CUDA By Example, which I highly recommend.

I also heavily commented the makefile detailing some of its inner workings, so please read it if makefiles seem intimidating!

galaxy1.txt is some galaxy with a dark matter halo generated by the Kuijken and Dubinski code. The first 50,000 particles are baryonic matter, the second 50,000 are dark matter. Blue particles are Baryonic, red are dark matter.