光学学报, 2015, 35 (2): 0209001, 网络出版: 2015-01-15
一种基于新型查表方法的统一计算设备架构并行计算全息算法
A New Look-up Table Method of Holographic Algorithms Based on Compute Unified Device Architecture Parallel Computing
全息 三角函数查表算法 查表法 统一计算设备架构 并行计算 holography trigonometric look- up table method look- up table method compute unified device architecture parallel computing
摘要
为解决点源法计算全息速度较慢的问题,提出了一种新的查表算法,命名为三角函数查表法(T-LUT算法)。该算法是基于点源法基本的数学公式,通过一系列数学近似与恒等变换,生成了一种纯相位查找表,该查找表具有三维特性,并具有生成速度快、精度高、占用内存少等特点,克服了点源法重复计算相位的缺点。同时采用统一计算设备架构(CUDA)并行计算在图形处理器(GPU)上加以实现,并进行了三次并行优化。在算法的验证与对比实验中,采用单显卡(GPU显卡)实现T-LUT算法,在不牺牲全息图再现像质量的前提下,成功地将点源法计算全息的速度大幅度提升。实验发现在不同的物空间采样点数量的情况下,速度相对于点源法GPU 运算提升30倍至近千倍不等。
Abstract
We propose a new look- up table algorithm, named trigonometric look- up tables (T- LUT) method to increase the speed of coherent ray trace (CRT)algorithm in generating holograms. The proposed algorithm is based on the original ray tracing method. By using a series of mathematical approximation changes and identical transformations, we generate a pure phase look- up table. This look- up table has three- dimensional features, fast generating speed, high precision, less memory capacity and so on. The new method overcomes CRT′ s shortcomings in computing the phase repeatedly. We implement T- LUT on graphics processing unit (GPU) using parallel computing by compute unified device architecture (CUDA), and optimize it three times. After algorithm optimizing, a series of experiments are carried out to verify the results by using a single GPU card. Our results indicate that the T- LUT algorithm can effectively shorten the computing time without sacrificing the quality of hologram reconstruction. Due to the different quantities of spatial samples, the speed has increased 30 to nearly a thousand times compared with original ray tracing algorithm implemented on GPU.
蒋晓瑜, 丛彬, 裴闯, 闫兴鹏, 赵锴. 一种基于新型查表方法的统一计算设备架构并行计算全息算法[J]. 光学学报, 2015, 35(2): 0209001. Jiang Xiaoyu, Cong Bin, Pei Chuang, Yan Xingpeng, Zhao Kai. A New Look-up Table Method of Holographic Algorithms Based on Compute Unified Device Architecture Parallel Computing[J]. Acta Optica Sinica, 2015, 35(2): 0209001.