Heterogeneous Computing with Focus on Mechanical Engineering
Appears in the following Collection
- Matematisk institutt 
AbstractDuring the past few years there has been a revolution in the design of desktop computers. Most processors today include more than one processor core, allowing parallel execution of programs. Furthermore, most commodity computers include a graphical processor that outperforms the central processor by at least one order of magnitude. Tapping into this vast resource is commonly referred to as heterogeneous computing. The change in hardware invalidates old software-design truths. There is therefore need for new algorithms, and research into adapting existing algorithms to these architectures. Our main focus has been to accelerate algorithms relevant for mechanical engineering. In this dissertation we present four algorithms devoted to take advantage of the computational strengths of heterogeneous architectures. Each work is based on state-of-the-art hardware available at the time the research was performed.
First we describe an algorithm for high-quality visualization of parametric surfaces. This is useful in a CAD setting, were an accurate rendering is important for visual validation of model quality. We further describe simulation of shallow-water waves using a state-of-the-art numerical scheme. Our accelerated implementation gave a speedup of up to 40 times compared to an optimized reference implementation. Our implementation features real time simulation and visualization of semi-realistic nonlinear wave effects.
Finally we present two algorithms for shape simplification of 3D-models. The algorithms aim at reducing time spent on preparing models for finite element analysis. Finite element analysis is important to determine mechanical properties of objects prior to manufacture. Such analysis can be used to investigate thermal behavior and determine the strengths and weaknesses of physical components. Before the analysis can take place the models must undergo a preparation phase where shape simplification plays an important role. The first work we describe for shape simplification is a hybrid algorithm, using graphics hardware for the computationally demanding operations, and the main processor for maintaining the data structure. Our second work describes a shape simplification algorithm highly suitable for heterogeneous architectures and a reference implementation on the Cell BE.