摘要
多GPU无合成绘制算法采用混合sort-first与sort-last的并行绘制模式,将绘制任务划分为多个子任务集合进行并行绘制并异步合成显示。原方法中对屏幕的分割方式过于平凡,未明确提出一种高效的屏幕分割策略,从无合成算法本身特性出发,提出一种自适应屏幕划分策略,该算法根据GPU数目及绘制分辨率参数,启发式计算屏幕初始划分方式,再采用统计性能反馈的方式动态选择最佳屏幕划分方式,较原有方法更易于实现子屏幕任务的负载平衡,实验表明,算法在带来很少额外计算开销的同时,能够有效平衡并行绘制任务负载,进一步发挥多GPU无合成并行绘制系统的绘制效率。
The task partition strategy of compositeless parallel rendering algorithm is based on simple hybrid sort-last and sort first rendering mode. This strategy is too ordinary to exploit the efficiency of compositeless method. The characters of compositeless method were analyzed, and an adaptive task partition strategy in screen-order was proposed. The initial tile styles were computed heuristically according to thread numbers and valid screen, then the rendering frame rates were read back dynamically and the tile layout was tuned up adaptively. Using this feed back method, the optimal tile layout could be detected quickly while the associated overhead could be ignored. The experiment shows that combined with the adaptive algorithm, the compositeless method could achieve much higher performance than ever.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2012年第1期99-103,共5页
Journal of System Simulation
基金
国家"973"项目(2009CB723803)
国家自然科学基金(61170157)
