A fast (?) random sampling drawing library

inconvergent inconvergent Last update: Oct 27, 2023

Desert

Desert consists of two parts.

The main library is simply called Desert. It is a CUDA accelerated library for sandpainting: https://inconvergent.net/grains-of-sand/

The second part is called Erosion. A Redis-based client and worker that can accept and draw Desert primitives and commands encoded as JSON objects. That means that you can use the Erosion worker from any platform as long as you can construct JSON and send it to a Redis queue. Eg. if you want to program in a different language, while still having a fast drawing engine that benefits from CUDA.

I've written a little more about the library here: https://inconvergent.net/lost-in-the-desert/

img

Install

Use the install script:

./install.sh

This will use setuptools to install python libraries desert and erosion. As well as a terminal util called erosion. It will be available as ~/.local/bin/erosion if you installed with the install.sh script.

Examples

There are some examples in ./examples.

To use Desert via Python as a local library, see:

main.py

To see how Erosion works, you can run this command (from ./examples):

./erosion-send.py && ~/.local/bin/erosion worker --path ./ --show --vv

This will first send some Desert primitives to the Erosion (Redis) queue. Then it will run the Erosion worker, which draws those primitives. Finally it will save the resulting image.

To see how the Erosion terminal util works:

~/.local/bin/erosion -h

Dependencies

This code is developed on Ubuntu 16.04 LTS. I imagine you will be able to get it running on Mac as well.

The library depends on the CUDA toolkit (8.0), Redis (if you are using Erosion), and a few Python (3) packages. If you install using the install script, the python packages will be installed automatically.

On Use and Contributions

This code is a tool that I have written for my own use. I release it publicly in case people find it useful. It is not however intended as a collaboration/Open Source project. As such I am unlikely to accept PRs, reply to issues, or take requests.

Todo

Desert:

  • Circle: varying rad
  • Box: varying size

Notes

If cuda is not working try sudo ldconfig. and check $LD_LIBRARY_PATH

https://documen.tician.de/pycuda/tutorial.html#executing-a-kernel

http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#kernels

Subscribe to our newsletter