结合预处理BiCGStab和CUDA的BLT快速并行前向方法  

Fast Parallel Forward Method of BLT Combining Preconditioned Bi CGStab and CUDA

作　　者：冯金超 李祎楠 李哲 贾克斌 

机构地区：[1]北京工业大学信息学部,北京100124

出　　处：《北京工业大学学报》2017年第11期1658-1665,共8页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(81370038);北京市自然科学基金资助项目(7142012);北京市科技新星计划资助项目(Z141101001814107)

摘　　要：针对基于简化球谐波(simplified spherical harmonics,SPN)方程开展生物发光断层成像(bioluminescence tomography,BLT)前向问题研究时计算量大、求解速度偏慢的问题,提出了一种基于稳定双共轭梯度下降(biconjugate gradient stabilized,Bi CGStab)的快速并行求解算法.该算法结合不完全Cholesky分解的预处理方式与压缩行格式存储法(compressed row storage scheme,CSR)的稀疏矩阵存储方式,并采用统一计算设备架构(compute unified device architecture,CUDA)实现了并行加速.数值仿真结果表明,该算法在保证前向问题求解准确度的同时可以极大地缩短求解时间.A fast parallel algorithm based on Bi-conjugate gradient stabilized（ BiCGStab） was proposed to solve the simplified spherical harmonics（ SPn） equations to reducethe computational burden and improve computational efficiency for forward problem of bioluminescence tomography （ BLT）. In the algorithm, preconditioned method of incomplete-Cholesky factorization and sparse matrix’s compressed row storage scheme （CSR） representing method were adopted. Furthermore, compute unified device architecture （CUDA） parallel programming model was used to implement parallel accelerating. The results of numerical simulation show that the proposed algorithm not only can ensure the solution accuracy of the forward problem but also greatly shorten the equation-solving time.

关 键 词：生物发光断层成像 简化球谐波方程 预处理稳定双共轭梯度下降算法 统一计算设备架构(CUDA) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]