Software developer blog

Game of Life on the GPU

Conway's Game of Life is one of the most popular code katas, and the one we will exercise on our next code retreat this Saturday. I have done this kata, quite a number of times, but this time there was a twist: I used the GPU on my Nvidia GTX 560 to calculate the steps of evolution, and to draw them as well. Here is how the result looks like:

Puffer train in Game of Life

Puffer train in Game of Life. Living cells are green, they increase the blue chanel by 1 in each generation, and leave a faiding red trail after they die.

The igniter of this exercise was that I read a book on CUDA programing, and it was quite a disappointment. The examples were really far from anything I would call clean code, or at least easy to understand: it was spaghetti with huge meatballs in it. That kind of examples require a lot of explanation, which basically took up most of the books volume, and there was little left for the useful stuff. So I got curious: is it CUDA that promotes bad code, or the writers of the book didn't care enough? I had a preconception that there was nothing wrong with CUDA  - and actually it's a pretty neat language - so I set out on a quest to code Game of Life in CUDA, and try to make it as clean as possible.  I don't claim, that my code is perfect, but I hope it's going to be an easy read for anyone who tries.

You can download the result of my quest by clicking here, or just browse it as HTML by clicking here. The rest of the blog entry - basically a retrospective of the project, and a few remarks on CUDA features - might not make much sense without looking at parts of the code. If you'd like to compile, than you will need the nvcc compiler for CUDA, the CUDA Toolkit, GLUT and GoogleTest to run my unit tests. Please note, that the code is not production quality: it may well throw exceptions at you, it's only tested on Ubuntu, and it is not performance optimized in any way. (Even the algorithm is the most naive one.)

@ //