Simulation program to compute the movement of N bodies due to their gravitational interaction.
Clone the repository with
git clone https://github.com/corn2424/N-Body-Sim
The main functionality is provided by particle_simulation.h. To use this create your own .cc file which specifies the parameters for the simulation and the initial positions, velocities and masses of the particles you which to simulate. A few toy examples are provided in form of the .cc files and an according .gif, stored in /data/gif, which shows the result in animated form.
The included examples are:
- The earth moon system (earth_moon.cc)
- A hypothetical system containing the earth and 7 moons which don't orbit in one plane, which makes it highly unstable (earth_7moons.cc)
- A simple example of a slingshoot (aka. swing-by or gravity-assist) with the two bodies being the earth and the moon (slingshoot.cc)
- A highly simplified version of saturn's ring (saturn_ring.cc)
All .cc files have to be compiled with the flag that allow the usage of AVX2
intrinsics, so e.g. -march=native for a machine having native support for AVX2 or -mavx2 for other machines. This is due to the fact that
SimulateAVX2() makes use of AVX2 intrinsics.
However a alternative not utilizing those is available as well in SimulateCPU()
if support for AVX2 isn't available on a machine.
For multi threading and c++17 feature support the flags -std_c++17 -pthread are needed.
Tested with g++ version 10.2.0 .
An example:
g++ -std=c++17 -Wall -O3 -mavx2 -march=native -mtune=native -pthread earth_moon.cc -o a.out
The functionality is provided in particle_simulation_GPU.cu, which should be included instead of particle_simulation.h. The simulation on the device isnt (much) faster or even slower for a low amount of particles but starts to be MUCH faster at some point (for my 3900x and my gtx1070 at roughly N=).
The included example is the GPU version of saturn_ring.cc provided in saturn_ring.cu.
The .cu files have to be compiled with nvcc, which compiles the device code and passes the remaining c++ code to the host compiler. The in the previous section mentioned compiler flags for the host compiler must be passed through nvcc via the flag -Xcompiler.
An example:
nvcc -std=c++14 -Xcompiler -Wall -Xcompiler -std=c++14 -Xcompiler -O3 -Xcompiler -mavx2 -Xcompiler -march=native -Xcompiler -pthread saturn_ring.cu -o a.out
Constantin Pestka