A programming model for GPU-based parallel computing with scalability and abstraction