WebGPU GPU-Driven Occlusion Culling and Hi-Z Pipeline Architecture
Bypassing CPU bottlenecks by generating Hierarchical-Z mipmaps and executing compute-driven object visibility tests entirely within WebGPU workgroups.
Rendering sprawling architectural models or dense city environments in web browsers requires culling obscured geometry before submitting draw calls. Relying on CPU bounding-box tests introduces severe frame latency, making GPU-driven occlusion culling essential.
1. Constructing Hierarchical-Z (Hi-Z) Mipmaps
Iteratively downsampling depth buffers into max-reduction mipchain textures allows bounding boxes to be tested against depth values in a single texture lookup regardless of scale.
2. Compute Shader Object Visibility Dispatch
Running parallel WGSL compute passes evaluates instance bounding spheres against the Hi-Z buffer, filtering out occluded objects directly into indirect draw command buffers.
3. Indirect Multi-Draw Rendering in WebGPU
Feeding filtered storage buffers straight into `drawIndirect` or `drawIndexedIndirect` calls eliminates CPU-GPU synchronization stalls and minimizes draw call overhead.