Efficient Compression and Rasterization Algorithms for Graphics Hardware

Jon Hasselgren
Lund University


Despite rapid development, modern graphics hardware is still much too slow to render photo-realistic images in real time, and it will most likely remain so if we rely only on yearly growth in hardware performance. Therefore, better algorithms are constantly needed in order to advance the field.

In the first part of this thesis, we present new algorithms for hardware rasterization. We investigate different aspects of sampling, which leads to a new family of very inexpensive sampling schemes. Based on a perceptual error metric, the best performing patterns are presented. Our study of sampling also considers the use of conservative rasterization, and we present two novel algorithms designed to work on existing graphics hardware. Our final contribution for rasterization is a hardware algorithm for efficiently rasterizing multiple views of a three-dimensional scene. The current norm for multi-view rasterization is to process each view separately, but in this work, we exploit the inherent coherence by considering all views simultaneously. This is done by sorting the rasterization order among all views.

In the second part of this thesis, we consider compression algorithms for graphics hardware. Although compression does not provide anything new in terms of features, it is a good way of improving performance and lowering memory usage in a hardware system. Our contributions in this field are two new compression algorithms. The first is suited for real-time compression and decompression of depth values, and the second suited is for compression and decompression of high dynamic range textures.


[ Thesis ]