Efficient data reduction and cache-coherent techniques toward real-time performance

Abstract

Figure 1: Application of our multi-resolution techniques and cache-efficient layouts to the rendering of the double eagle tanker (82 million triangles). By using these techniques, we are able to significantly improve the performance over the previous rasterization techniques and achieve an interactive performace on a commodity hardware. The performance improvement is mainly achieved by the significant reduction on the amount of necessary data and cache-coherent access patterns on the data. Course Notes: State-of-the art in Massive Model Visualization. Kasik et al. 44 Course Notes: State-of-the art in Massive Model Visualization. Kasik et al. 45limited memory . Algorithms for occlusion culling and out-of-core techniques also perform computations based on the view parameters. However, no known algorithms integrate conservative occlusion culling and out-of-core rendering with vertex hierarchies. Particularly, Yoon et al. [2004b; 2005b] proposed a clustered hierarchy of progressive meshes(CHPM) for interactive view-dependent rendering of massive models. The CHPM consists of two parts:• Cluster Hierarchy: The entire dataset is represented as a hierarchy of clusters, which are spatially localized mesh regions. Each cluster consists of a few thousand triangles. Conceptually, a cluster hierarchy is similar to a vertex hierarchy. However, every node of a cluster hierarchy represents a set of vertices and faces rather than a single vertex. The clusters provide the capability to perform coarse-grained view-dependent (or selective) refinement of the model. They are also used for visibility computations and out-of-core rendering.• Progressive Mesh: Then, each cluster is also represented as a linear sequence of edge collapses as a progressive mesh (PM). A PM is a mesh sequence built from an input mesh by a sequence of edge collapse operations. The inverse operation, a vertex split, restores the original vertices and replaces the removed triangles. The PMs are used for fine-grained local refinement and to compute an error-bounded simplification of each cluster at runtime. In practice, refining a PM is a very fast operation since PM can be stored in an array and can be refined without checking any dependency, which is typically required to guarantee correct LOD representation.CHPM representation provides two levels of refinement for interactive view-dependent rendering of massive models. First we perform a coarse-grained refinement at the cluster level. Next we refine the PMs of the selected clusters. The PM refinement provides smooth LOD transitions.This representation has been implemented as a system called Quick-VDR on a commodity PC. This system was demonstrated with several models including a complex CAD environment (12M triangles), scanned models (372M triangles), and an isosurface (100M triangles). Moreover, they were able to render these models at 10 − 35 frames per second using a limited memory footprint of 400 − 600MB.In this course note, we have discussed three orthogonal methods, 1) multi-resolution method to reduc...

Bounding Volume Hierarchies of Slab Cut Balls

“…On a higher algorithmic level, it also involves designing cache aware or cache oblivious data structures and algorithms (see e.g. [YM06, Kas07]). Also, multi‐core CPUs and parallel programming methods can be utilized to give our methods a further performance boost.…”