International Association
for Cryptologic Research

We introduce a CUDA GPU library to accelerate evaluations with homomorphic schemes defined over polynomial rings enabled with a number of optimizations including algebraic techniques for efficient evaluation, memory minimization techniques, memory and thread scheduling and low level CUDA hand-tuned assembly optimizations to take full advantage of the mass parallelism and high memory bandwidth GPUs offer. The arithmetic functions constructed to handle very large polynomial operands using number-theoretic transform (NTT) and Chinese remainder theorem (CRT) based methods are then extended to implement the primitives of the leveled homomorphic encryption scheme proposed by L{\\\'o}pez-Alt, Tromer and Vaikuntanathan. To compare the performance of the proposed CUDA library we implemented two applications: the Prince block cipher and homomorphic sorting algorithms on two GPU platforms in single GPU and multiple GPU configurations. We observed a speedup of 25 times and 51 times over the best previous GPU implementation for Prince with single and triple GPUs, respectively. Similarly for homomorphic sorting we obtained 12-41 times speedup depending on the number and size of the sorted elements.