General-purpose GPU computing: practice and experience