|
|
Terrain Processing on Modern GPU
Margold, Roman ; Pelikán, Josef (advisor) ; Maršálek, Lukáš (referee)
Recent development in graphics hardware opened the possibility for new terrain rendering techniques. Former techniques are based on rapid geometry reduction performed on CPU. Recent approaches are moving the load from CPU to GPU, thus keeping CPU available for other tasks, which is especially important for game development. A majority of these approaches is restricted to static datasets or limited data size. We introduce a novel approach which is capable of modifying data during runtime and is easily applicable to potentially infinite landscapes. It is implemented entirely on GPU, except view frustum culling, which is still performed by CPU. However, thanks to the proposed terrain sampling scheme it is trivial and extremely efficient. Employed two-level data representation offers simple implementation without loss of functionality such as sudden change of visual angle or view direction. Those are common problems of other approaches. We also adopted block compression to keep memory consumptions low. Further, a general loading mechanism has been designed and implemented in order to allow asynchronous data read from external medium. This system has been optimized for reading with serial access, although data demands are processed in parallel. The system has been used for continuous terrain data retrieval and...
|
|
|
Terrain Processing on Modern GPU
Margold, Roman ; Maršálek, Lukáš (referee) ; Pelikán, Josef (advisor)
Recent development in graphics hardware opened the possibility for new terrain rendering techniques. Former techniques are based on rapid geometry reduction performed on CPU. Recent approaches are moving the load from CPU to GPU, thus keeping CPU available for other tasks, which is especially important for game development. A majority of these approaches is restricted to static datasets or limited data size. We introduce a novel approach which is capable of modifying data during runtime and is easily applicable to potentially infinite landscapes. It is implemented entirely on GPU, except view frustum culling, which is still performed by CPU. However, thanks to the proposed terrain sampling scheme it is trivial and extremely efficient. Employed two-level data representation offers simple implementation without loss of functionality such as sudden change of visual angle or view direction. Those are common problems of other approaches. We also adopted block compression to keep memory consumptions low. Further, a general loading mechanism has been designed and implemented in order to allow asynchronous data read from external medium. This system has been optimized for reading with serial access, although data demands are processed in parallel. The system has been used for continuous terrain data retrieval and...
|
guest ::
