Dynamic Scheduling And Mesh Layout Optimization Of Massive Complex Scene

Real-time rendering is an important and highly active area in Computer Graphics. Nowadays with the tremendous advances in both hardware capabilities and rendering algorithms, rendering performance is steadily increasing. However, scene complexity seems to be rising even faster than rendering performance due to more advanced model acquisition methods, such as 3D scanner and massive collabrotive modeling software.To visualize massively complex scene real-time efficiently, algorithms specially designed to alleviate the bottleneck between internal and external memory attract abroad interests. This thesis proposed an Out-of-Core framework to implement cache and on demand strategy based on Memory Map File and multi-threading technology. Meanwhile, during the preprocess period, we optimized mesh layout of shapes to maxmize the usage of GPU vertex cache, and optimized inter-shapes layout off-line to implement automatic prefetch and caching when rendering.The first chapter introduces previous work in real-time rendering, especially Out-of-Core algorithms, then an overview of Tyr system with the focus on scene data representation, and an overview of Out-of-Core design in Tyr. The second chapter first explained the tiled memory management module based on Memory Map File, and its deliberate design to support asynchronous loading, prefecthing and eviction in upper modules. Then goes the elaborate introduction of Out-of-Core framework, including data preprocessing, loading, multi-threading, etc. The third chapter focuses on the optimization of geometry data layout during preprocessing, including vertex layout optimization of shapes (specially for GPU vertex cache), and inter-shape layout optimization(specially for the internal and external caching strategy). The fourth chapter evaluates the performace of our system, then concludes and ends with an outlook on future work.