Quinton Grant

What I built

• GPU compute pipeline: agent updates + trail writes on GPU to avoid CPU bottlenecks.
• Minimized GPU ⇄ CPU synchronization: kept state on GPU to reduce stalls and readbacks.
• Stable performance at scale: profiled dispatch sizes, buffer usage, and per-frame work to keep frame times consistent.
• Shippable cross-platform Unity project: integrated Steam + iOS features while keeping core simulation portable.

What’s interesting here

• Compute-driven simulation: spore updates run on GPU; CPU mostly updates settings + dispatches work.
• Ray marching optimization: two-pass approach using Jump Flood for coarse SDF then a fine SDF for detail.
• Shader “header injection”: shared struct definitions between C++ and GLSL to avoid duplicated maintenance.

Platform note: Linux + Windows supported. macOS not supported (OpenGL 4.3 compute shaders).

Pipeline (high level)

• Silhouette seeds → Jump Flood → distance field generated each frame.
• Ray-marched lighting: stochastic sampling + sphere tracing against the distance field.
• Temporal accumulation: exponential moving average to reduce noise.
• Performance knobs: downscale, samples/adaptive samples, blur, time constant.