Rahul P. Sathe

Rigid body collision detection on the GPU

We propose a technique for collision detection of rigid bodies using cube-maps. This technique runs entirely on todays GPUs. Conventional rigid body collision detection algorithms run on the CPU. This technique is highly parallel in nature and highly scalable. At the core of the algorithm is the idea of creating a cube-map for each unique object. This cube-map stores the distances to the surface of the object from its centroid. The algorithm has two steps - a preprocessing step, and a run-time step. Preprocessing is done only once per unique object and the run-time step is evaluated for every instance of the object every frame.

Pixel Merge Unit

Multi-sample anti-aliasing is a popular technique for reducing geometric aliasing (jagged edges) and is supported in all modern graphics processors. With multi-sampling anti-aliasing, visibility and depth are sampled more than once per pixel, while shading is done only once per pixel per primitive. Although this significantly reduces the appearance of jagged edges around object boundaries, the image quality improvement in non-silhouette regions is hardly noticeable. We propose a hardware unit, called the pixel merge unit, which is located just after the early depth test unit but before the pixel shader. Our unit attempts to reduce the shading rate to once per pixel per group of connected primitives covering a pixel using a novel merging strategy. We demonstrate up to 15% reduction in pixel shader executions. Given the simple implementation that we propose, this is a substantial reduction. (Less)

Post-Tessellation Geometry Caches

Current 3D rendering architectures support adaptive tessellation of patches, allowing for increased geometric detail. Patches are specified independently, giving the implementation freedom to exploit parallel execution. However, this independence leads to redundant shading computations at patch corners and along edges. In this paper, we present post-tessellation geometry caches, the edge cache and corner cache, that can reduce redundant shading along patch edges and corners, respectively. We demonstrate the two caches in a software-simulated D3D11 rendering pipeline, and show that for current tessellation workloads our approach saves up to 37% of post-tessellation vertex shading using caches with as few as 8 entries.