Explores using spatial hashing as a memory-efficient alternative to 3D textures for storing raytraced ambient occlusion data. The technique quantizes 3D space into cells indexed by hash values derived from world position and surface normals, enabling caching and reuse of RTAO calculations for static scenes. Implementation details include hash functions (pcg, xxhash32), collision resolution via linear probing, distance-based cell sizing, and age-based cache eviction. Results show significant performance improvements (0.4ms vs 1.72ms) with reduced noise and better stability compared to traditional RTAO, though limited to static geometry.
Sort: