Abstract: Graphic Processing Unit（GPU） uses a huge number of transistors to create a large number of processing units，applying to the handling of Single-Instruction Multiple-Data（SIMD） computing tasks.In this paper，the architecture of GPU and the programming model of CUDA are first studied，and then the traditional algorithm of Hill encryption and decryption based on CPU are redesigned into SIMD mode，so that time-consuming part of matrix multiplication during the whole process can be implemented by parallel threads.Finally，the effect of the number of threads per thread block on speed-up is analyzed.The result shows that the method of parallel matrix multiplication based on GPU is with high efficiency compared with the tradition Hill encryption and decryption algorithms implemented on CPU，can get the acceleration of more than 12 times，and shows great potential in data encryption and decryption.

Key words: Compute Unified Device Architecture（CUDA）, Graphic Processing Unit（GPU）, Hill encryption, Hill decryption

摘要： GPU有效地利用了数量巨大的晶体管制造大量的处理单元，适用于处理单任务多数据（SIMD）的计算任务。研究了GPU的体系结构及CUDA的编程模式，改进了基于CPU的希尔加解密方法，使用多个线程将计算中耗时的矩阵相乘部分改造成SIMD模式，并分析了线程块内线程数对加速比的影响。实验结果表明，基于GPU的并行矩阵相乘的希尔加解密方法成功实现了硬件加速，相对于CPU上运行的希尔加解密方法，其执行效率明显提高，可获取12倍以上的加速，并易于扩展，对大规模数据加密和解密处理呈现出高效的处理能力。

关键词: 统一计算设备架构, 图形处理单元, 希尔加密, 希尔解密