Acceleration of Homomorphic Unrolled Trace-Type Function using AVX512 instructions

Abstract

More and more data analysis is being outsourced due to the spread of cloud computing. Therefore, protection of the data from privacy violations and information leaks is required. In particular, homomorphic encryption, which allows computation to be performed with encrypted data, is being actively studied as one of the protection method. Ring learning with errors based homomorphic encryption schemes support packing which allows to pack several elements into slots of a plaintext and ciphertext. A trace-type function, which combines shifting slots (rotation) and homomorphic addition to obtain summation of slots, is often used in homomorphic encryption applications and acceleration of the trace-type function is important. In this paper, we further accelerate the trace-type function using Intel AVX512 compared to existing optimized trace-type function with loop unrolling. The results show that our AVX512 version was 1.05-2.30 times speedup compared to the non-AVX512 version.