A lot of what is done in a computer is branching logic code, and that's not so easy to deploy on opencl or cuda which are to a first approximation arithmetic engines.