On Exploring the Reasoning Capability of Large Language Models with Knowledge Graphs