1000 SDF Spheres Colliding at 200 Frames Per Second

Spheres are not a stress test, but it does run in 5 milliseconds. Also, as it turns out, the step in the algorithm where the n-d spheres are filtered should be single threaded. this is quite counter intuitive, and might be caused by the simplicity of the SDFs used, but it did degrade performance by 20%. Using Vec3A (SIMD optimized Vec3) had a negligible impact on performance, but this might also be caused by the simplicity of spheres. I plan to test more complex and/or expensive shapes like fractals, and add unit tests for false positives/negatives using solved geometry. Still, I am very impressed by the results. Unfortunately, raymarching is quite slow, so the complexity of the demos I can make is still very limited.