Inhaltszusammenfassung:
Within this dissertation, a set of algorithmic optimizations are developed,
enabling significant performance improvements due to a much better
utilization of the available bandwidth. Additionally, new architectural
concepts circumventing the bottle-necks of currently available general
purpose graphics hardware are presented. In the field of polygon rendering,
a unique mechanism for hardware supported occlusion queries to cull geometry
prior to geometry transformation --- saving bandwidth on the front bus ---
is presented. As an orthogonal addition to this, a novel visibility driven
rasterization scheme is presented, saving processing cycles within the
pipeline by culling occluded geometry prior to rasterization. Thus, more
objects can be rendered or more cycles can be spent on multi-pass rendering
of the potentially visible objects. With respect to volume rendering, this
dissertation contributes the first side by side comparison of different
volume rendering algorithms identifying each algorithm's strengths and
weaknesses. Furthermore, new techniques for using polygon graphics hardware
and multi-pass rendering are presented, enabling the combination of shading
and classification of volume data. Additionally, minor modifications to the
data path are proposed such that multi-pass rendering can be avoided, thus
increasing the overall achievable frame-rate. Furthermore, we demonstrate
how to efficiently use general purpose hardware (a single-chip SIMD
architecture) for volume rendering, providing much more flexibility than
dedicated polygon graphics hardware. As a summary of the above described
work, a novel low-cost special purpose hardware architecture that achieves
superior image quality while providing an incomparable degree of flexibility
is presented.
Abstract:
Within this dissertation, a set of algorithmic optimizations are developed,
enabling significant performance improvements due to a much better
utilization of the available bandwidth. Additionally, new architectural
concepts circumventing the bottle-necks of currently available general
purpose graphics hardware are presented. In the field of polygon rendering,
a unique mechanism for hardware supported occlusion queries to cull geometry
prior to geometry transformation --- saving bandwidth on the front bus ---
is presented. As an orthogonal addition to this, a novel visibility driven
rasterization scheme is presented, saving processing cycles within the
pipeline by culling occluded geometry prior to rasterization. Thus, more
objects can be rendered or more cycles can be spent on multi-pass rendering
of the potentially visible objects. With respect to volume rendering, this
dissertation contributes the first side by side comparison of different
volume rendering algorithms identifying each algorithm's strengths and
weaknesses. Furthermore, new techniques for using polygon graphics hardware
and multi-pass rendering are presented, enabling the combination of shading
and classification of volume data. Additionally, minor modifications to the
data path are proposed such that multi-pass rendering can be avoided, thus
increasing the overall achievable frame-rate. Furthermore, we demonstrate
how to efficiently use general purpose hardware (a single-chip SIMD
architecture) for volume rendering, providing much more flexibility than
dedicated polygon graphics hardware. As a summary of the above described
work, a novel low-cost special purpose hardware architecture that achieves
superior image quality while providing an incomparable degree of flexibility
is presented.