基于仿真的多核处理器功能验证技术研究

发布时间：2018-06-08 06:00

本文选题：功能验证 + 遗传算法　；参考：《浙江大学》2017年博士论文

【摘要】：随着嵌入式领域应用需求的不断提高,多核处理器的体系结构变得愈加复杂,设计规模越来越庞大。功能验证已经成为多核处理器设计过程的制约瓶颈,高效的功能验证方法可以有效地缩短多核处理器的验证时间和面市时间。本文的研究工作针对多核处理器的功能验证,从基于仿真的验证方法出发,对其相关的领域进行了研究和探索,提出多种技术以提高多核处理器的功能验证效率,并成功应用于两款多核处理器的功能验证。本文的研究工作主要分成以下三个方面逐步展开:(一)基于遗传算法的覆盖率驱动测试产生器研究在多核处理器的功能验证过程中,验证人员利用覆盖率识别没有验证的功能,评估验证工作的进度。为了更好地建立覆盖率与测试产生器之间的联系,驱动测试产生器生成高质量的测试,本文提出一种基于遗传算法的覆盖率驱动测试产生器。利用改进的自适应遗传算法建立覆盖率与测试产生器之间的联系,根据覆盖率分析结果自动改变测试产生器的约束和限制,驱动测试产生器生成全新的测试。这些测试比原有的测试质量更高,可以覆盖到原有的测试无法覆盖的功能点。实验结果表明,和随机测试产生器相比,本文提出的测试产生器可以得到较高的覆盖率,同时节约大约70%的验证时间,显著提高了多核处理器的功能验证效率。(二)基于k均值算法的测试排序技术对于基于仿真的多核处理器功能验证,利用测试产生器在短时间内产生成百上千的测试是很有必要的,然而仿真大量的测试会耗费很长的时间,造成验证效率低下,在验证初期筛选出质量较高的测试可以极大程度提高验证效率。为此,本文提出一种基于k均值算法的测试排序技术,根据测试的预估覆盖率对测试进行重新排序,筛选出可以在较短时间内得到较高覆盖率的高质量测试,从而提高多核处理器的功能验证效率。并在此基础上提出三种改进k均值算法,进一步提高测试排序的效率和准确度。实验结果表明,本文提出的改进k均值算法比标准的k均值算法具有更高的准确度和效率。相对于随机仿真测试,本文提出的方法可以在节约90%仿真时间的情况下,得到相当的覆盖率。(三)对称多核处理器参考模型研究基于仿真的多核处理器功能验证效率很大程度决定于参考模型的仿真速度和准确度。本文提出一个高效的对称多核处理器参考模型,建立一个功能精准级模型作为CPU流水线的参考模型,在保证准确度的同时,显著缩短建立时间,提高仿真速度;利用一个时序精准级模型验证对称多核处理器的高速缓存一致性;并利用一个依赖时序序列作为以上两个模型之间的接口。实验结果表明,本文提出的参考模型拥有较低的建立难度,并且可以很好地兼顾仿真速度和准确度,能够帮助验证人员快速发现和定位对称多核处理器的功能错误,提高功能验证效率。
[Abstract]:With the increasing requirement of embedded applications, the architecture of multi-core processors becomes more and more complex and the design scale becomes larger and larger. Functional verification has become a bottleneck in the design process of multi-core processors. Efficient functional verification methods can effectively shorten the verification time and market time of multi-core processors. In this paper, the functional verification of multi-core processors is studied. Based on the simulation verification method, the related fields are studied and explored, and a variety of technologies are proposed to improve the efficiency of multi-core processors function verification. It has been successfully applied to the functional verification of two multi-core processors. The research work in this paper is divided into the following three aspects: (1) A coverage driven test generator based on genetic algorithm is developed step by step. In the process of functional verification of multi-core processors, validators use coverage to identify functions that are not verified. Evaluate the progress of validation work. In order to establish better relation between coverage and test generator and drive test generator to generate high quality test, this paper proposes a coverage driven test generator based on genetic algorithm. An improved adaptive genetic algorithm is used to establish the relation between the test generator and the test generator. According to the results of the coverage analysis, the constraints and limitations of the test generator are changed automatically, and the test generator is driven to generate a new test. These tests are of better quality than the original tests and can cover functional points that the original tests could not. The experimental results show that compared with the random test generator, the proposed test generator can obtain higher coverage, save about 70% of the verification time, and improve the efficiency of functional verification of the multi-core processor. (2) the test sequencing technology based on k-means algorithm is necessary for the functional verification of emulation based multi-core processors, and the use of test generators to generate hundreds of tests in a short time. However, simulation of a large number of tests will take a long time, resulting in low efficiency of verification, screening out high quality tests in the early stage of verification can greatly improve the efficiency of verification. For this reason, this paper proposes a test sorting technique based on k-means algorithm, which reorder the test according to the estimated coverage of the test, and select the high quality test which can get higher coverage in a short time. In order to improve the efficiency of functional verification of multi-core processors. On this basis, three improved k-means algorithms are proposed to further improve the efficiency and accuracy of test sorting. The experimental results show that the improved k-means algorithm is more accurate and efficient than the standard k-means algorithm. Compared with the random simulation test, the method proposed in this paper can save 90% of the simulation time and get considerable coverage. (3) Research on the reference Model of symmetric Multicore processor; the efficiency of multi-core processor function verification based on simulation depends largely on the simulation speed and accuracy of the reference model. In this paper, an efficient reference model of symmetric multi-core processor is proposed, and a functional precision level model is established as the reference model of CPU pipeline. It can shorten the time of establishment and improve the speed of simulation while ensuring accuracy. A timing precision level model is used to verify the cache consistency of a symmetric multicore processor, and a dependent sequence is used as the interface between the two models. The experimental results show that the reference model presented in this paper has lower difficulty in setting up, and it can well balance the simulation speed and accuracy, and can help the verifier to quickly detect and locate the functional errors of the symmetric multicore processor. Improve the efficiency of functional verification.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP332

【相似文献】