GPGPU optimized parallel implementation of AES using C++ AMP

Nowadays, the characterization of a computing system using attributes like "single core" is, for most applications, deprecated. Multiprocessor or multi-core platforms are, now, widespread and serve for solving more and more complex problems in shorter execution times. Video cards make no exception to this rule since, for the past years, they are based on powerful GPUs with high parallelism architectures and extremely fast memories. In addition, new development languages and platforms became available for the programmers. This way, the processing power of the GPU (Graphics Processing Unit) can now be used even for non-video or non-graphics applications that imply a serious amount of parallel processing. This paper presents a comparative study of AES algorithm implementation on CPU and two different GPGPU platforms. Similar studies involving GPGPUs are based on Nvidia's CUDA platform but this approach imposes a severe limitation over the application portability. In our approach a platform independent application was designed and implemented using C-H- AMP, the latest C++ extension oriented to parallel programming. Tests were conducted over two GPGPU platforms, one from NVidia and one from AMD and a multi-core CPU from Intel. Results show that cross-platform portability was achieved while the performances are similar or better as compared to similar studies.