摘要
针对3D引擎内如何合理调度、高效分配统一着色架构中的染色器资源问题,提出一种基于负载均衡的任务调度策略和硬件设计方案。首先,分析了统一着色架构GPU的图形绘制特点;在此基础上提出了一种基于染色器预先分配的避免图形绘制流水线死锁策略和一种基于队列统计的资源分配及负载均衡策略;最后,基于各类复杂渲染场景的仿真结果证明,论文提出的调度策略可有效地降低染色器资源的空闲时长和调度时间,提升了染色器资源的利用率和3D引擎的处理效率。
For the problem of how to logically schedule and efficiently allocation the shader resources of unified shader architecture inside of 3D engine, this paper proposes a task scheduling strategy and a hardware design scheme based on load balancing. Firstly, this paper analyses the characteristics of graphics rendering of unified architecture GPU. Secondly, based on the above analysis, two strategies are put forward. One is shader pre-allocating strategy which can avoid deadlock in graphing pipeline, another is resource scheduling and load balancing strategy based on queue statistics. Finally, the results of all simulation tests with complex rendering scene proves that the strategy proposed on this paper can reduce idle time and increase usage rate of shader resources as well as enhance processing efficient of 3D engine.
出处
《电子技术应用》
北大核心
2017年第5期55-59,共5页
Application of Electronic Technique
