Reproduce “CuRast: Cuda-Based Software Rasterization for Billions of Triangles” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Markus Schütz, Lukas Lipp, Elias Kristmann, Michael Wimmer
CuRast is a CUDA-based software rasterization pipeline designed for rendering massive, unstructured triangle meshes without precomputed acceleration structures. It achieves up to 12 times faster rendering for instanced dense geometry and 2-5 times faster for unique geometry compared to Vulkan in specific high-triangle-count scenarios.
Key insights
- Carefully optimized brute-force GPGPU software rasterization with multi-stage processing can significantly outperform hardware rasterizers for massive, dense, and unstructured triangle meshes.
- Adapting the rasterization strategy based on triangle screen-space size is critical for achieving efficiency across a wide range of geometry within a unified pipeline.
- Using 64-bit atomicMin to interleave perspective-correct depth and a unique global triangle ID in a visibility buffer allows for efficient depth testing and deferred shading without complex synchronization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.