Optimal implementation of PIPO block cipher CTR mode on GPU

In this paper, we propose a technique to optimize the CTR mode of the PIPO block cipher in a GPU environment. Unlike the existing individual block encryption method, we improve the computation speed by applying an optimization technique that packs four plaintexts and processes them in parallel. In addition, we optimize the packing and unpacking of plaintexts and master keys by utilizing the built-in functions of CUDA, and minimize the number of memory accesses and improve the encryption performance by performing Sbox, Pbox, and KeyAddition operations on four plaintexts simultaneously. The encryption method applying this optimization technique showed the effect of reducing the number of global memory accesses and improving the computational performance compared to the existing method.