Efficient Transformation for Bottom-up Computation of Stable Models