面向实时物理渲染图形流水线的体系结构关键技术研究

The physical rendering pipeline aims to create a highly realistic photo level image by modeling and simulating the interaction between light and scene through physical principles, and is expected to become the mainstream real-time graphics rendering method of the next generation. But its computational complexity is very high, and the traditional GPU design is oriented to the rasterization rendering algorithm, so the efficiency of physical rendering is low. With the continuous progress of integrated circuit technology, the rendering ability of most advanced GPU has more than ten times performance gap for real-time physical rendering. It is highly feasible to make up for this performance gap through customized design, which means that graphics processors will soon cross the threshold of real-time physical rendering, and the era of real-time physical rendering is coming. This project will conform to this trend and satisfy the requirement of real-time physical rendering graphics pipeline from the perspective of processor architecture and hardware design. It combines the advanced algorithm of graphics rendering, the new approximate computing, and PIM technology. It focuses on solving the problem of physical rendering programming model, large-scale light parallel processing, hardware support for dynamic generation of scene data and so on. The research work of this project will provide the key theory and technical support for the implementation of the future physical rendering-based graphics processor.

物理渲染图形流水线以真实感为目标，通过物理原理对光线和场景之间的交互行为进行建模仿真，从而绘制出高度逼真的照片级图像，有望成为下一代主流实时图形渲染方法。但其计算复杂度高，且传统GPU设计面向光栅化渲染算法，物理渲染效率低下。随着集成电路技术的不断进步，目前最先进GPU的物理渲染能力与实时渲染还有十几倍的性能差距。通过定制设计来弥补这一性能差距具有非常高的可行性，这就意味着图形处理器即将跨越实时物理渲染的门槛，实时物理渲染的时代即将到来。本项目将顺应这一趋势，从处理器体系结构和底层硬件设计角度出发，将图形学领域最新物理渲染算法与近似计算、PIM等体系结构先进技术相结合，重点解决物理渲染编程模型、大规模光线并行处理、场景数据动态生成等高效硬件支持的设计优化问题，实现满足实时要求的物理渲染图形流水线。本项目的研究工作将为基于物理渲染未来图形处理器的实现提供体系结构关键理论与技术支撑。

渲染是将三维场景的描述转换为二维图像的过程，主要包括光栅化和光线追踪两种方式。光线追踪采用全局光照模型，通过物理原理对光线和物体之间的交互行为进行建模。基于光线追踪的物理渲染因其渲染效果的真实性，在学术界和商业界引起广泛研究，有望成为下一代主流实时图形渲染方法。本项目从处理器体系结构和底层硬件设计角度出发，深入探索面向实时物理渲染图形流水线的体系结构关键技术，取得了一些技术突破和创新。.项目主要通过周期精确模拟器对物理渲染流水线和其中关键算法进行微体系结构层次的细粒度仿真分析，给出了面向实时物理渲染图形流水线的体系结构设计思路。其主要包括面向光线追踪硬件加速器的设计、图形流水线编程模型以及面向计算单元效率和存储架构优化等方面。基于近似计算，项目提出了硬件加速器中的计算单元结构，并且优化了基于光线混排的光线并行处理体系结构和场景动态构建加速器设计等。经过原型系统验证，本项目的研究工作可为基于物理渲染未来图形处理器的实现提供体系结构关键理论与技术支撑。