A High-productivity Framework for Multi-GPU computation of Mesh-based applications