Fast AES Implementation Using ARMv8 ASIMD Without Cryptography Extension

Abstract

While the ARMv8-A ISA allows for hardware accelerated cryptographic instructions, such extension is not available for every device, being added at the discretion of the CPU manufacturer. Prime examples of ARMv8 devices without this support are the low cost Raspberry Pi 3B/3B+/4 single board computers. This work presents an optimized AES implementation targeting CPUs without Cryptography Extension instructions, relying only on ASIMD operations. We show a new implementation that processes four blocks at the same time, which requires block permutations and modified versions of the main layers. In particular, we provide a new efficient formula for computing the MixColumns layer. The time performance our AES implementation outperforms the current ASIMD implementation found in the Linux Kernel by about 5%.