Parallel Rendering Method Of View Shot For Large-Scale Scenes By High Performance Computer

The digital creative industry is based on modern information technology and incorporates cultural creativity and design services into its emerging industries.Its main applications are in virtual reality,product visualization,animation and film special effects.At present,the development of the digital creative industry is mainly to develop in a more realistic direction.Therefore,the three-dimensional scene grows rapidly in scale,and the illumination model is closer to nature,thereby causing the complexity of the scene to be larger and larger,and the scale of the scene data.It is possible to reach the TB level.Such rendering puts new demands on the storage and visual effects of large-scale scene data.Two important indicators for measuring high-reality rendering are rendering quality and rendering speed.International research on high-reality rendering is also focused on these two aspects.The research on rendering quality mainly focuses on the simulation of physical optics.Since the global illumination model accurately simulates the illumination of all possible situations in the scene,the global illumination algorithm can generate photo-realistic images,which are realistic rendering core.Optical path traceing is one of the basic methods to achieve global illumination rendering.It can calculate most of the global illumination effects.However,due to the large computational complexity of the rendering algorithm,parallel acceleration is one of the necessary research work.At the same time,in order to improve the rendering quality and create special effects,the degree of detailing of the scene and the scale of the scene are also expanding,and the global illumination calculation in the highly realistic rendering based on physics requires the use of the overall scene data,so the large-scale scene drawing will have a memory bottleneck problem.In rendering speed,parallel rendering is the primary means of accelerating the rendering of highly realistic.At present,domestic and foreign research mainly focuses on parallel rendering based on GPU or Intel integrated many-core architecture,lacking universal processing for heterogeneous many-core architecture,and at the same time,the scalability is relatively low,and the scale of scenes that can be processed is limited.Therefore,researching efficient and scalable high-reality parallel rendering acceleration is of great significance to the digital creative industry.Aiming at the data intensive and computationally intensive problems of large-scale scene lens rendering,this paper mainly proposes multi-level task partitioning and corresponding multi-level task scheduling,and optimizes according to the data correlation of tasks between successive frame renderings.Perform dynamic and static division to accelerate lens rendering.In this paper,distributed storage scene data is used to divide the rendering computing tasks into multiple levels for multi-level task scheduling.At the same time,in order to make full use of the computing resources of Sunway Taihu Light,the platform-specific accelerated thread library is used for calculation acceleration.For the continuous frames of the rendered shot,there is static scene data consistent between each frame,so the node partitioning and loading scene data are optimized according to this featureAs the scale of the scene is enlarged,the number of nodes drawn increases,and the stability of the node begins to decrease with time.Therefore,the parallel rendering of large-scale scene shots necessarily requires fault-tolerant processing,which needs to provide further guarantee for the correctness of the rendering results.This paper investigates the related fault-tolerant processing in the hope that errors will be discovered when a node fails,and the task progress can be continued after restarting the task.