面向CPU+GPU异构计算的多目标测试用例优先排序

测试用例优先排序是一种基于整个测试用例集以寻找最优测试用例执行序列的软件回归测试技术.由于其能够尽早地发现错误,同时应用灵活度高、不会漏掉重要测试用例等,在实际软件测试过程中可以有效提高测试效率.多目标测试用例优化排序是寻找同时覆盖多个测试准则的用例执行序列,通常采用演化算法优化求解,但执行时间较长,严重影响了在实际软件测试中的应用.采用先进的GPU图形卡通用并行计算技术,提出了面向CPU+GPU异构计算下的多目标测试用例优先排序技术,在NSGA-II算法中,实现了基于序列编码的适应度函数计算和交叉操作的GPU并行计算,在近6万行有效代码的工业界开源程序上实现了30倍的计算效率提升.同时,实验验证了不同并行策略的计算加速比,提出了切实可行的CPU+GPU异构计算模式,并提供了相应的原形工具.

CPU+GPU Heterogeneous Computing Orientated Multi-Objective Test Case Prioritization

Test case prioritization is a type of technique that aims at searching for the test case execution sequence to find faults early based on the whole test case suite. This technique is flexible and barely can miss important test cases, which contributes much benefit to regression testing. Multi-objective test case prioritization aims to find a test case execution sequence that suits multiple test criteria, in which evolutionary algorithms have been widely used. However, the drawback of large computation cost of the algorithms can greatly reduce the value of industrial application. In this paper, we propose a CPU+GPU heterogeneous Computing orientation based multi-objective test case prioritization technique, which is using advanced General Purpose Graphic Process Unit (GPGPU) technique to accelerate the process of test case prioritization. In experiment based on parallel structure, we design the sequential based parallel fitness and crossover operation computation on NSGA-II and at last achieve 30 times speed-up rate on an well-known industrial open source project. Based on the systematically study on the benefit of different types of parallel strategies, we propose a CPU+GPU heterogeneous Computing framework and the prototype of tools is developed and available online.