I just completed my first open-source commit: the
sort! method for Julia’s official CUDA library, CUDA.jl. You can now sort arrays stored in GPU memory. It’s an implementation of Quicksort, which is a bit tricky to parallelize effectively. Here I’ll share a few of the insights which brought my pull request together.
This article balances code, algorithms, and hardware considerations like a whiteboard interview: lots of pictures and a little code when it helps. I’ve organized the content as follows:
sort!for fun and profit
Hi! I'm a software engineer, physicist, and enthusiast of numerical computing.