海光处理器上后量子签名算法的 AVX2 优化实现

With the rapid development of quantum computing technology, traditional cryptographic systems are facing huge security threats, and the research of post-quantum cryptography (PQC) is becoming increasingly imminent. In this context, lattice-based cryptography has become an important foundation tool for post-quantum digital signature algorithms with its excellent ability to resist quantum attacks effectively. The HuFu algorithm is a post-quantum digital signature algorithm independently developed by our country. It is based on the universal lattice problem in lattice cryptography and has good security and application prospects. It has been submitted to the National Institute of Standards and Technology (NIST) of the United States for standardization evaluation. However, the performance of the HuFu algorithm still needs to be improved, especially in terms of algorithm complexity, memory efficiency, and parallel computing capabilities. It also needs to enhance compatibility with different hardware and instruction sets. To this end. this paper is based on the domestic Hygon Processor, giving full play to its advantages of high parallelism, low energy consumption and high throughput, and providing strong hardware support for the efficient implementation of the HuFu algorithm. At the same time, combined with the 256-bit Advanced Vector Extension (AVX2) instruction set, this widely used single instruction multiple data (SIMD) technology, further enhances the parallel computing capability of the algorithm, thereby effectively improving the overall performance. This paper comprehensively considers multiple aspects such as matrix multiplication optimization, instruction set acceleration, encoding processing simplification and memory access efficiency, and adopts a series of algorithms and technical optimizations to significantly improve the calculation speed, reduce resource consumption, and improve the overall performance of signature generation and verification. Specifically, the optimization scheme of this paper includes several key technical points; First, the matrix multiplication is optimized in combination with the Strassen algorithm, which is obvious. The calculation speed is greatly improved and resource consumption is reduced. Secondly, the range variant encoding of asymmetric numbers (rANS) is optimized by using the AVX2 instruction set. which speeds up the speed of signature generation and verification. In addition, in view of the complex and time-consuming problem of sign bit processing in rANS encoding, unsigned parameters are used to achieve efficient signature and verification processing, which simplifies the calculation process and reduces the calculation overhead. Finally, by designing a reasonable function interface and memory access optimization technology, the memory usage efficiency in the signature and verification stages is improved, and the frequent writing of registers is reduced. Compared with the original AVX2 implementation of the HuFu algorithm, the optimization scheme proposed in this paper reduces the clock cycle consumption in key generation, online signature stage, offline signature stage, and the total signature and verification stage by about 46%, 54%, 45%, 30% and 46% respectively. Efficient signature algorithms significantly enhance processing capabilities in high-concurrency computing environments, improve overall system stability and security, better protect sensitive post-quantum cryptographic data from various quantum computing threats, and promote the continued development of domestic post-quantum cryptographic technology advancements.