High-performance parallel acceleration method for domestic public key cryptographic algorithm SM2 based on domestic GPU

To meet the strategic demand for independently controllable national information security and to ensure algorithm transparency and security, a high-performance parallel acceleration method for the domestic public key cryptographic SM2 digital signature algorithm based on domestic GPU was proposed. Firstly, low-level functions for field operations were designed to optimize the efficiency of finite field arithmetic, with two-round carry resolution adopted in modular reduction to resist timing attacks. Then, point addition and point doubling were implemented using Jacobian coordinates, leveraging register and global memory features, and offline/online precomputation tables were introduced to enhance scalar multiplication efficiency. Finally, experimental design was carried out based on the characteristics of the Hygon deep compute unit (DCU) platforms, achieving high-performance SM2 signature and verification with throughput of 6 816 kops/s and 1 385 kops/s, respectively. Results demonstrate the feasibility and effectiveness of implementing clemestic public key cryptographic SM2 based on domestic GPU, providing crucial support for national information security autonomy.