Gravity

This is an attempt to compare computation time for different implementation of a (triangular) matrix multiplication task.

The task is a brute force simulation of the nbody problem, thus O(n^2) in time complexity.

3 implementations are compared:

• nbody_c: implemention in C.
• nbody_cuda: implementation in Cuda (disclaimer: I'm not a Cuda expert, don't expect this to be well optimized).
• nbody_pytorch: implementation in Pytorch. GPU acceleration can be deactivated by adding the flag CUDA_VISIBLE_DEVICES= to the commandline.

Rendering is also available.

Install

python setup.py install

Run

python run.py --backend [nbody_c,nbody_cuda,nbody_pytorch,nbody_triton]

Rendering

Computed trajectories are stored in result.data. They can be rendered by adding the flag --render then rendered.

To render previously saved trajectories, run:

python run.py --render path/to/another.data