CPU/GPU系统上存储高效的RNA二级结构预测算法  

Cache-efficient RNA Secondary Structure Prediction Algorithm on CPU / GPU Systems

作　　者：郑明[1] 钟诚[1] 

机构地区：[1]广西大学计算机与电子信息学院,南宁530004

出　　处：《小型微型计算机系统》2014年第5期1080-1084,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60963001)资助

摘　　要：通过建立映射变换函数来改进基于最小自由能的RNA二级结构预测计算模型,分析证明了改进后的计算模型与原计算模型的等价性,利用改进后的计算模型使得GPU每个warp线程束内的线程并行计算矩阵元素时其所需的数据处于全局存储器同一行中,以支持直接并行读取矩阵元素,显著地减少多线程并行访问全局存储器的次数;充分利用GPU纹理存储器、共享存储器及常量存储器,以减少查找表的时间;设计实现多核CPU/单GPU系统、多核CPU/多GPU系统上存储高效的RNA二级结构预测并行算法.实验结果表明,与已有的RNA二级结构预测算法相比,本文提出的算法效率更高.The RNA secondary structure prediction computation model with the minimum free energy is improved by establishing the mapping function, the equipollence of improved computation model and original computation model is analyzed and proved. Based on the improved RNA secondary structure prediction computation model, the two cache-efficient RNA secondary structure prediction al- gorithms are presented on hybrid multi-core CPU/uni-GPU system and multi-core CPU/multi-GPU system respectively, which the threads for each warp in GPU can directly read the dependent data that are located in the same line in the global memory when the data elements in the matrix are computed in parallel, and the required times to access the global memory can be remarkably reduced, the time to search free energy tables can be reduced by texture memory, shared memory and constant memory. The experimental results show that, the proposed parallel algorithms for RNA secondary structure prediction computation are superior to the existing RNA sec- ondary structure prediction algorithm.

关 键 词：RNA二级结构预测 GPU计算 并行算法 动态规划 最小自由能 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]