Implementation Of A Parallel Light Path Tracking Rendering Engine For The Light Of Sunway TaihuLight

The digital media industry is considered to be a content innovation industry produced by the integration of digital technology and cultural entertainment industry[1].It has developed especially rapidly since twenty-first Century.It has gradually become one of the most active applications in the field of high performance computing.It has a large number of users.Meanwhile,the digital media industry is also a strategic emerging industry vigorously developed in China in recent years.With the rapid development of national cultural industry,the demand for image presentation accuracy,quality and scale is becoming more and more demanding in cultural and creative industries.Its large-scale computing task and data storage demand give new challenges to the realistic rendering of digital media and visual effect simulation.Since 2009,since the first P supercomputer has been successfully developed in China,the high performance computing technology has developed rapidly in our country.At present,the computing power of high performance computer has stepped into the age of E computing from the P level.The computing and storage capacity of the computer has reached a new height,and the core number of supercomputers also extends from hundreds of thousands to millions.Usually high-performance computers are mainly used in the need for a lot of computing work,such as weather forecasting,climate research,seismic simulation,molecular model,cryptanalysis and so on.There are not many attempts to use high-performance computers for highly realistic rendering,and the highly realistic rendering application of digital media for high-performance computers at home and abroad is not yet mature.Therefore,it is of high application value to develop a fully functional,stable and reliable rendering system based on high performance computer made in China,and how to fully utilize the high performance computing ability of the E level computing time to meet the large-scale and high realistic rendering demand is the key problem to be solved in this project.At present,most of the rendering systems are implemented in parallel rendering based on GPU.This paper is based on the domestic CPU heterogeneous public kernel architecture to achieve parallel rendering.It can promote the application of domestic high-performance computer in the field of graphics rendering,and promote the research and use of the international advanced level in the field of digital media.The work done in this paper is part of the national key project "High-Realism Parallel Rendering of Digital Media and High-Performance Application Software System ".This paper focuses on the development and deployment of a highly efficient and highly realistic rendering engine for high performance computer made in China,the high performance computer,Sunway TaihuLight,so that it can achieve good parallel efficiency on high performance computers.The main work of this article includes the following aspects:First,the scene file is processed,including the definition of the scene file specification and the scene file loading.This paper uses XSD to define the element data type and attribute in the scene file,and uses the Xerces-C++ to parse scene files to load the scene information in the scene file into memory.Secondly,a communication mechanism is established to achieve parallel rendering between master and slave nodes.Then the scene tasks are decomposed and distributed to each sub node to achieve load balancing.A three level scheduling mechanism is established,and multiple computing main nodes are set up on the second level to alleviate the communication pressure between the third tier slave nodes and the first tier main nodes.To establish fault tolerance mechanism,the main node monitors the heartbeat from the node,and finds out the node anomaly to transfer the rendering task to the normal node immediately,so as to ensure the normal rendering task.Finally,the deployment test of rendering engine for domestic high performance computing system in Sunway TaihuLight is presented.The ultimate goal of the whole engine is to be able to run efficiently and steadily on the home-made high performance computing system of the Sunway TaihuLight,but the deployment of the engine is relatively complex due to the use of domestic processors in Sunway TaihuLight and its unique compiler.After the deployment is successful,some parallel tests will be done,so that its parallel efficiency at the three hundred thousand core level will exceed 60%.