Voxel octree raycaster written using SIMD Intrinsics. All math is fixed-point. Uses AVX2, casts 8 rays simultaniously.
Raycaster code at src/game.c:343
Made this purely to practice manually vectorizing complicated semi-branchy algorithms, and to see just how slow this kind of raycasting is on a CPU, when all possible tricks are used.
Raycasting is done all with fixed-point integers. No division is used either as AVX2 lacks that insturction. Fuzziness and certain artifacts are due to these optimizations and workarounds. The algorithm appears to be severly memory bandwidth limited, it does not scale at all past 4 parallel rays.
Uses my Voxplat engine as a base, thus has a lot of inherited code. Uses OpenGL for text rendering and to present and scale the raycasted image. Traversal algorithm is original, 2D JS demo used to develop it at https://kosshi.net/projects/quadtree-caster/
Processor used is AMD Ryzen 1600. All cores used.
And when things go wrong ..
