Parallel and High-speed Implementation of SM4 Encryption Algorithm on OpenCL

The current SM4 encryption algorithm is almost implemented on the CPU using a serial algorithm, which cannot meet the requirements of real-time data receiving and transmitting under the current high-speed network environment. In order to improve the speed of encryption and decryption to handle large-scale data such as 5G and Internet of Things, the paper designed and implemented a parallel SM4 algorithm of GPU based on OpenCL. And the performance of the original serial SM4 algorithm and the parallel SM4 algorithm based on OpenCL are compared and verified. The experimental results demonstrate that when the plaintext data exceeded 256KB, the speed-up ratio began to increase significantly. When the plaintext data reached about 8MB, the speed-up ratio of single GPU basically stabilized and tended to be flat. There is also a small increase in encryption speed by storing data in local variables. In the optimal case realized in this paper, the single GPU can bring about 180 times of acceleration efficiency even in the experimental environment of the same CPU, and the multi-GPU device can achieve 760 times of acceleration efficiency improvement. The SM4 computing using GPU can meet the real-time encryption and decryption requirements under 10 Gigabit Ethernet.