Exploring ways to optimize compute shaders - Part 1. :: The GutHub — My humble blog

In this post we’ll be looking at compute shaders, profiling tools and failed experiments to improve performance. This is my journey learning compute shaders first through WebGPU and then Vulkan. My goal was to implement Conway’s Game Of Life algorithm in a brute force manner and to make it run as fast as possible on the GPU .

This will not be a tutorial on how to setup compute shaders. The easiest way to get started today is probably with WebGPU and this tutorial does a pretty good job at getting you through it. Once you’re familiar with the API, this example does everything you need to setup a proper Game Of Life.

Conway’s Game Of Life is a cellular automaton with very simple rules. A cell lives or dies depending solely on how many of its neighbours are alive or dead. In our case, we’ll be running the simulation in a 2D grid which means a cell has 8 neighbours. The rules are as follows:

Let’s look at a very straight forward way to implement this. Our simulation data from the previous simulation step is stored in a 1D array called srcGrid.state and the result should be stored in another 1D array called dstGrid.state. In those arrays, a dead cell is represented by 0 and a living cell is represented by 1. For now we’ll consider that somehow the GPU executes 1 thread per cell and that the variable currentCellIndex points to the right address in both srcGrid and dstGrid.

Leave a Comment