基于CUDA的测井数据并行处理方法初探

On Log Data Parallel Processing Method Based on CUDA

测井数据处理程序大多基于CPU的串行程序,在处理信息量较大或算法较为复杂的数据时,处理速度难以满足实际需求。探索了GPU并行处理技术在测井数据处理中的应用;在CUDA(Compute Unified Device Architecture)平台上实现了基于GPU的核磁共振T_2谱反演算法程序;对T_2谱反演程序进行了详细分析,在移植到CUDA平台时对程序的访存模式进行了优化。对基于CPU和GPU的测井数据处理程序进行了对比实验,测试的T_2谱反演程序获得了约18倍的加速比,表明基于GPU的处理程序具有明显的效率优势。

Log data processing software is mostly based on CPU serial program.When the data is big or the algorithm is complicated,the processing speed is difficult to meet the actual demand.The GPU parallel processing technology was applied in well logging data process in this paper.And the nuclear magnetic resonance T_2 spectrum inversion algorithm is realized in CUDA(Compute Unified Device Architecture)platform.What's more,the fetch mode is optimized when transplanting the T_2 spectrum inversion procedure to CUDA platform.Finally,logging data processing program based on CPU and GPU is carried on the contrast experiment,the results show that the later has obvious efficiency advantage with about 18 times speedup.