A Highly Parallel CUDA Implementation of the Mercury N-body Integrator

Note: Units are such that G=1, ie mass is in solar masses, distance is in AU, time is in years, etc.

Compiling

To compile, run make in the root directory. This will create a binary in the bin directory. Only supported on Hopper GPUs due to requiring distributed shared memory.

Running

Run the binary on a GPU node using ./bin/mint with a set of command line arguments:

• -c specifies the path to the JSON config file. For an example of what this should look like, check out the examples directory

Optional:

• -i prints simulation information
• -p prints positions at each timestep for each body in the simulation
• -t sets the number of timesteps to run the simulation for (default is 1)
• -d sets the GPU device to use (default is 0)