Relevant Technology Study On Programmable Graphics Hardware

The computation power of the Graphics Processing Unit (GPU) in current commodity graphics hardware is increasing at a much faster rate than that of the Central Processing Unit (CPU) in computer systems. The projected time to double in efficiency for the GPU is quoted to be roughly 12 months by the leading graphics card manufacturers. A recent major breakthrough in graphics hardware technology has been the introduction of programmability; this allows the user to replace portions of the fixed graphics pipeline with customized shader programs exposing the ability of GPU to function more like a general processing unit. In spite of all the rendering power, it is not possible or meaningful to use algorithms designed with CPU in mind on graphics hardware. The essential difference is that GPU provides a highly parallel Single Instruction Multiple Data Set (SIMD) architecture. The key to harnessing this resource is reengineering the computationally expensive algorithms to take advantage of this architecture as well as making use of rendering optimizations built into the programmable graphics pipeline. This thesis presents several novel graphics approaches which utilize programmable graphics hardware to obtain both real-time frame rate performance and high quality result.Our research works in this thesis mainly focus on the following aspects:1. Real-time Voxelization for Complex ModelsWe present an efficient voxelization algorithm for complex polygonal models by exploiting newest programmable graphics hardware. We first convert the model into three discrete voxel spaces according to its surface orientation. The resultant voxels are encoded as 2D textures and stored in three intermediate sheet buffers called directional sheet buffers. These buffers are finally synthesized into one worksheet, which records the volumetric representation of the target. The whole algorithm traverses the geometric model only once and is accomplished entirely in GPU, achieving real-time frame rate for models with up to 2 million triangles. The algorithm is simple to implement and can be integrated easily into diverse applications such as volume based modeling, transparent rendering and collision detection.2. High Quality Real-time Rendering of Large Scale Point ModelHere we introduce an adaptive rendering algorithm for large scale point models. The algorithm first subdivide the target model into multiple patches in preprocess. A hierarchical structure is built for each patch and then converted into a linear binary tree. During rendering, the model is processed patch by patch. Fast visibility decision is made to cull invisible patches. Visible patches are displayed in GPU by choosing appropriate rendering mode, i.e, a distance-dependent strategy. Our algorithm takes full advantage of GPU and effectively balances the workload between CPU and GPU. We also propose a fast compression/decompression technique which achieves 8 times compression ratio. The results demonstrate high performance and image quality rendering for large scale point models in consumer PC.3. Real-time shadow mappingShadow mapping is an image-based shadowing technique. It is particularly amenable to hardware implementation because it makes use of hardware functionality- texturing and depth buffering existed. Here we present the implementation process of two real-time shadow mapping methods in detail: common GPU-based shadow mapping and hardware shadow mapping.Finally, I summarize my own research experience of programmable graphics pipeline and propose some potential research topics in the future.