A Highly Efficient GPU-CPU Hybrid Parallel Implementation of Sparse LU Factorization