摘要
为提高植物模拟生长算法求解性能,基于GPU SIMT执行模型对算法步骤进行详细并行化设计并绘制算法流程图。算法代码利用C++和CUDA实现,能够在CPU+GPU混合架构平台上运行;在此基础上,利用7个不同类型的基准测试函数对算法进行测试,与串行算法代码进行对比。分析结果表明,在保证求解质量的前提下,相比串行算法,该并行算法将目标函数复杂的大规模问题的求解效率提升了十几倍到几百倍,其加速比较好,不同变量维数下的求解时间增加不明显,验证了当前工作的可行性和有效性。
With the purpose of improving the solving performance of plant growth simulation algorithm（PGSA）,the steps of PGSA were parallelized in detail using the SIMT（single instruction multiple threads）execute model of GPU（graphic processing unit）,and the algorithm flow chart was also presented.The code of the proposed parallel PGSA was written in C＋＋ with CUDA（compute unified device architecture）,which could be run on the CPU＋GPU hybrid platform.Based on such works,the proposed parallel PGSA was tested by seven benchmarks with different types,and it was also compared with the serial PGSA code.The analysis results show that,on the premise of maintaining the normal calculating precision,the solving efficiency of the proposed parallel PGSA is improved more than tens to hundreds of times comparing with the serial one for large-scale problems with complex objective functions.Its speedup factor is also excellent,the variation of solving time is not obvious with the increase of functions dimension.The analysis results validate both the viability and the effectiveness of the proposed algorithm.
出处
《计算机工程与设计》
北大核心
2016年第12期3271-3279,共9页
Computer Engineering and Design
基金
国家自然科学基金项目(11375085)
国家磁约束聚变能发展专项基金项目(2014GB108002)
关键词
植物模拟生长算法
并行化
图形处理器
统一计算设备架构
混合架构平台
plant growth simulation algorithm
parallelization
graphic processing unit（GPU）
compute unified device architecture（CUDA）
hybrid architecture platform
