AES finalists implementation for GPU and multi-core CPU based on OpenCL

Benefit from the OpenCL (Open Computing Language), applications can be easily transplanted among different GPUs, multi-core CPUs, and other processors. In this paper, we present implementation of AES finalists (Rijndael, Serpent, Twofish) in XTS mode, based on OpenCL. Benchmark testing is performed on 4 mainstream GPUs and multi-core CPUs. The results are also compared with implementations based on traditional serial programming model and CUDA. The resulting data shows that throughputs based on OpenCL are higher than serial programming model, while a little lower than CUDA. Which demonstrates that OpenCL promises a portable language for GPU programming, while entail a performance penalty.