Rendering with poxels. A software architecture for programmable shading systems

This thesis describes and discusses a flexible and general concept for the design of a programmable shading system. The architecture developed here identifies and attacks some of the limitations known to be present in other approaches. While the main focus is on flexibility, generality and far reaching openness, goals such as safety and efficiency shall not be compromised. Although these goals are not new, and often desirable, an original approach has been developed and successfully implemented that meets them, while being surprisingly simple to understand and use. The systemis centered around two basic concepts, which are derived and explained in the thesis: the poxel, which is a model for encapsulating the inner state of the ren-dering process, and the shader, which is used to manipulate this state. Atop these two concepts, a powerful mechanism is developed that can be used to construct complex rendering algorithms. By expressing both data manipulation and data flow control with the same concept - the shader - arbitrarily complex algorithms can be implemented by composing predefined building blocks and custom components, thus promoting reuse and prototyping. Together with an object oriented approach to modeling and rendering, the system is capable of serving a large variety of needs, including the cooperation of different rendering methods such as ray tracing and scanline oriented methods in the same image. The special properties of shaders allow for a wide applicability, which is explored in detail. A number of example shaders are described, ranging from well known basic algorithms to special applications, including constructs from the field of programming languages. A section is devoted to related work, especially RENDERMAN,which is the quasi standard in programmable shading. It is shown that the system can be extended to offer a RENDERMAN-like interface, so that the shading process can even be specified in terms of a completely different programming interface. One of the key features of the system described here is that it allows the shad-ing process to be optimized at run time using techniques borrowed from compiler construction and code optimization. It even supports scene-dependent optimization. Thus, using appropriate techniques, the shading process can be significantly sped-up. All these optimizations can be performed automatically without user intervention, although the user has the ability to interfere with the process, if so desired