High quality commit messages are important for program understanding and maintenance, which describes the content of code changes. Neural-based methods are most popular ways to generate commit messages, but they neglect retrieval results including retrieval diffs and retrieval messages. The combination models of neural-based methods and retrieval-based methods have two major limitations: a) Ignore retrieval messages b) Seldom consider similarity and disparity between retrieval results and given diff. To address the above two issues, we propose a retrieve-guided method named ReGenSD to generate commit messages. Specifically, we decompose the commit message generation task into three steps. Firstly, we apply an similarity-based IR technique to get retrieval diff and retrieval messages. Secondly, we introduce a selective mechanism to decide following generation models based on lexical similarity between retrieval diff and given diff. Lastly, we design two kinds of generation models. For simple generation model, we use seq2seq network that only takes given diff as input. For retrieve-guided model, we design a novel seq2seq network with Bi-LSTM that takes given diff, retrieval diff and retrieval message as input. To capture semantic relations between retrieval results and given diff, we introduce a relation gate in encoder to leverage retrieval message adaptively based on semantic similarity, and a difference vector in decoder to refine the utilization of retrieval message based on semantic disparity. Experimental results on an open source dataset demonstrate that retrieval messages guidance can facilitate commit message generation task. Besides, ablation experiments prove the effectiveness of our proposed mechanisms on adjusting the use of retrieval results.