An Optimized Isomorphic Design for the SM4 Block Cipher Over the Tower Field.

The SM4 block cipher is Chinese domestic cryptography widely used to secure data confidentiality. Its performance is a key indicator in measuring the efficiency of encryption and decryption in large-scale data scenarios. In traditional tower field optimization, the paired forward and backward linear filed transformations are sandwiched between S-boxes and L-boxes for every round of the encryption operation, which introduces heavily burdensome computation times and complexity. In this paper, we propose a novel isomorphic design for the SM4 round function, where all operations are remaining in the tower filed through multiple rounds, and the paired forward and backward field transformations in each round can be omitted. Based on the isomorphic design, we introduce a more flexible fine-grained bitsliced scheme for the SM4 block cipher with the SIMD instructions, requiring only 32 independent data blocks to be processed in parallel. The experiments show that the proposed isomorphic design for the round function is superior to the traditional design, and the fine-grained bitsliced SM4 implementation on the server device and terminal device achieved 6.4 and 17.2 cycles per byte respectively, showing a performance increase of 284.3% and 54.4% compared to OpenSSL implementation(24.7 and 26.6 cycles per byte respectively).