使用GPU绘制分形图的混合精度方法研究  

Mixed Precision Method for GPU-Based Fractal Rendering

作　　者：刘伟峰 唐先明[1] 李媛媛[1] 曹邦功 

机构地区：[1]中国石油化工股份有限公司石油勘探开发研究院信息技术研究所,北京100083

出　　处：《工程图学学报》2009年第6期46-52,共7页Journal of Engineering Graphics

摘　　要：从Mandelbrot集分形算法在GPU上的计算速度和绘制效果角度出发,提出了在追求计算速度的单精度算法和追求绘制效果的双精度算法之外的第三种选择——混合精度算法。首先,分析了Mandelbrot集分形算法;其次,对GPU体系结构和CUDA编程模型进行了介绍;再次,在GPU上实现了单、双精度浮点数算法,基于实验结果提出了精度问题;最后,实现了混合精度算法并进行分析。结果显示,GPU上的混合精度优化是计算速度和绘制效果之间的折中,为分形算法的实现提供了新的选择。Focusing on the trade-off between frame rate and visual quality of GPU-based fractal rendering, mixed precision method is proposed. Parallelism within Mandelbrot sets, GPU architecture and CUDA programming model are analyzed firstly. Then single and double precision method of fractal rendering on GPU is implemented and floating point accuracy problem is detected. In order to solve the problem, the method of fractal rendering using mixed precision is presented. Experimental results demonstrate that the trade-off is superior to visual quality under single precision method and frame rate under double precision method on GPU-based fractal rendering. It is hopeful that mixed precision method would result in new perspectives for generalization of fractal graphics using GPU.

关 键 词：计算机图形学 分形图形绘制 混合精度方法 图形处理器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]