一种基于SIMD技术的快速并行代数重建算法  

A Rapid Parellel ART Based on SIMD Technology

作　　者：刘远[1] 张定华[1] 赵歆波[1] 毛海鹏[1] 刘晓鹏[1] 

机构地区：[1]西北工业大学现代设计与集成制造技术教育部重点实验室,西安710072

出　　处：《中国图象图形学报》2007年第1期73-77,共5页Journal of Image and Graphics

基　　金：国家自然科学基金项目(50375126)

摘　　要：代数重建算法是解决非完全投影数据重建的有效方法,尤其在对于超出探测器尺寸范围的大型零件的无损检测中已成为最有力的关键技术,但以往算法计算量较大、耗时较长。为了快速地进行代数重建,提出了一种基于Intel处理器单指令多数据(single instruction multiple data,SIMD)技术[2]的快速并行算法,并在充分分析代数重建公式特点的基础上,设计了一套便于并行化运算的数据结构及计算流程,其在运算中可一次性加载多个打包数据,利用MMX(multimedia extension)、SSE(streaming SIMD extension)和SSE2指令完成SIMD方式计算。通过仿真实验证明,该算法在达到同样精度的前提下,不仅提高了重建速度(加速比4倍),解决了传统代数重建算法运算速度慢的瓶颈问题,并且能够较好地重建部分数据缺失的投影图像,该算法对于航空航天大型零部件的无损检测具有重要的理论意义及工程应用价值。Algebraic reconstruction method（ART） is an effective approach to reconstruction of incomplete projection data and a most powerful key technology especially in nondestructive detection of parts of aircraft and spacecraft which are larger than the size of CT detector. However, former procedures have huge amount of computation and are extremely time- consuming. In order to improve these disadvantages, this paper brings forward a rapid parellel algebraic reconstruction procedure based on SIMD technology of Intel central process units. Having maturely comprehended the feature of ART formula, this novel procedure designs a set of data structures convenient for parallel computation and a procedure pipeline to load a number of packed data in one time and to complete reconstruction computation in SIMD method by MMX, SSE and SSE2 instructions. Proved by the simulating experiment, this method promotes the speed about 4 times with the same precision of ordinary procedures, and solves the bottle-neck problem of traditional ART procedures, which possesses important engineering applicational value of nondestructive detection for large parts of aircrafts and spacecrafts.

关 键 词：CT 代数重建 单指令多数据并行运算 SSE和SSE2指令 

分 类 号：TP391.72[自动化与计算机技术—计算机应用技术]