Implementation of Multiple-precision Modular Multiplication on GPU
Implementation of Multiple-precision Modular Multiplication on GPU
复制标题
多精度模乘法在GPU上的实现
DOI：
--
复制
发表时间：
2009
期刊：
影响因子：
--
通讯作者：
Kaiyong Zhao
中科院分区：
文献类型：
--
作者：
Kaiyong Zhao
多精确模块化乘法是安全应用程序中的关键组件，例如用于加密和签名数字数据的公共密钥密码学。但是不幸的是，对于当代CPU而言，它们在计算上很昂贵。通过利用多核GPU的计算能力，我们使用CUDA实现了一个多精确的整数库。在本文中，我们将研究GPU上多精确模块化乘法的两种方法的实现。我们将分析GPU上多精度模块化乘法指令的详细信息，并找到命中问题，然后我们建议使用内联ASM来改善此功能的实现。我们的实验结果表明，多精确模块化乘法的性能已提高了20％。
Multiple-precision modular multiplications are the key components in security applications, like public-key cryptography for encrypting and signing digital data. But unfortunately they are computationally expensive for contemporary CPUs. By exploiting the computing power of the many-core GPUs, we implemented a multiple-precision integer library with CUDA. In this paper, we will investigate the implementation of two approaches of multiple-precision modular multiplications on GPU. We will analyze the detail of the instructions of multiple-precision modular multiplication on the GPU and find the hit issues, and then we propose to use the inline ASM to improve the implementation of this function. Our experimental results show that the performance of multiple-precision modular multiplication has been improved by 20%.