自优化绘制流水线研究

This project addresses the optimization challenges of nowadays real-time rendering applications. With the rapid development of software and hardware technology, real-time rendering applications have integrated more and more rendering algorithms, the rendering process has become more and more complicated, and the programmable rendering pipeline has evolved with more and more flexibility. As a result, the manual optimization of rendering application rises as a great challenge. In order to solve this problem, this project attempts to introduce the program auto-tuning technology to build new rendering framework and pipeline that support automatic optimization. Specifically, parameters and code of rendering algorithms are analyzed to form proper optimization space for automated optimization. Then, specific search strategies and algorithms are designed to efficiently explore such a space and find the optimal rendering that adaptively assigns computations, reduces overheads and optimally optimizes resources. Through the research of this project, on the one hand, we can greatly reduce the dependence on the manual optimization and achieve better performance of real-time graphics applications. Moreover, it will better promote the development of real-time rendering technology as well as enlarge the scope of many graphics applications, such as visualization, games and virtual reality. On the other hand, the proposed auto-tuning rendering pipeline will inspire and support new theoretical and architectural design of the future real-time rendering framework and pipeline.

随着软硬件技术的飞速发展，实时绘制应用使用的绘制特效算法越来越多，包含的绘制过程越来越复杂，依赖的绘制流水线自定义程度越来越高，这使得对绘制应用进行人工优化变得越来越低效、越来越困难。针对这一挑战，本项目尝试引入程序自优化技术，构建能够支持自动优化的绘制计算框架与绘制流水线，通过对绘制算法、绘制过程参数与代码的解析与分析，创建用于自动优化的优化空间，利用自动优化算法在优化空间中的搜索，实现对绘制过程的自适应计算、性能加速与资源优化，从而解决实时绘制难于优化的难题，更好地提升实时真实感绘制技术的使用范围与应用场合。通过本项目的研究，一方面可以让实时绘制应用大大减少对人工优化的依赖，有效提升性能，进一步促进实时绘制技术的发展，深化并推动包含可视化、游戏以及虚拟现实技术在人们生活中的应用；另一方面将为实时绘制的计算框架与未来的实时绘制流水线设计提供新的理论与架构支撑。

本项目围绕当前实时绘制计算面临的优化难题开展研究，针对性研究了多种绘制算法，并通过引入程序自优化技术，构建自动优化的绘制计算框架，用于提升绘制性能。项目研究中，探索了自优化绘制理论与计算框架，研究了多种可适用于自优化的绘制优化目标与度量方法，重点分析了不同绘制算法的特征与优化空间构建方法，提出了一系列适用于绘制算法优化的自优化方法与优化工具，并构建了面向自优化计算框架的绘制流水线与绘制语言，最后将所研发的自优化绘制技术集成到了云端融合的实时绘制引擎之上，实现了所研发的绘制技术与应用集成。本课题已顺利完成计划工作的各项预期目标，在绘制理论、绘制算法、绘制优化实现、实时绘制引擎等方面取得了重要突破，形成了多项创新成果，达到了同类研究的国际领先水平。共发表（含录用）研究论文20 篇，包含4篇ACM TOG论文，5篇IEEE TVCG论文，1篇Cell子刊iScience，2篇ACM SIGGRAPH/SIGGRAPH Asia会议论文，2篇IEEE VR会议论文，共计14篇CCF-A的期刊和会议论文。已授权国家发明专利10项，注册软件著作权2项。培养博士后1名，博士4名，硕士生16名。项目相关的成果获2021年浙江省技术发明奖一等奖。