基于CUDA对RNA二级结构预测的并行研究

CUDA based parallel research of RNA secondary structure prediction

为解决长链RNA二级结构预测面临的计算压力,分析了当前基于最小自由能预测RNA二级结构的相关串行、并行算法,提出了一种基于CUDA(compute unified device architecture)编程模型的并行加速算法。在实现时充分利用了支持CUDA编程模型的GPU(graphic processing unit)设备中的共享存储器、常量存储器等硬件,对RNA二级结构预测算法中的动态规划进行了有效的并行。实验结果表明,在支持CUDA编程模型的GPU上实现的并行程序,获得了与已有的串行、并行算法相同的准确度,同时运行速度更快。

To overcome the computational pressure in the prediction of secondary structure of long-chain RNA, related serial and parallel algorithms according to MFE （minimum free energy） are analyzed, then a parallel implementation based on CUDA is proposed. The implementation takes fully advantage of shared memory and texture memory in CUDA based GPU for the effec tive parallel. The experimental results show that the parallel program implemented on the GPU which supports CUDA achieves the same accuracy compared with existing serial, parallel algorithms while yields reasonable speedups.