Software evolution is an important activity during a software development lifecycle. Understanding semantic differences between two versions of a software system is a crucial yet challenging task, especially in many safety critical sectors. Consequently, various techniques have been proposed to check semantic differences between a program and its evolution. But, many current techniques are still far from being satisfactory in terms of the accuracy and efficiency. In this article, we propose a novel framework, called US 2 E , which can efficiently and effectively generate the minimal number of test cases that reveal as many semantic differences across two versions as possible. Specifically, given a unified control flow graph that denotes two versions of a program, US 2 E executes as many common nodes as possible and leaves execution of non-common nodes separately in a single concolic execution instance. We evaluate US 2 E on 86 pairs of C programs from 4 benchmarks, and experimental results show that US 2 E can efficiently and effectively generate test cases demonstrating the semantic differences across 2 versions, with better performance than 6 baseline tools.