Extraction Of Micro Architecture Independent Workload Characteristics Parameters For The Mobile Application

In view of the constant growth of the test program, it is necessary to select the representative test program subset by studying the inherent characteristics of the test program. The researchers have developed the MICA simulator for acquiring micro architecture independent workload characteristics of the benchmark which runs on the x86 CPU. The basic information of the whole test set can be covered by using the micro architecture independent workload characteristic analysis method to extract the representative subset which only needs to measure the limited micro structure. It has a lot of questions to obtain the micro architecture independent workload characteristics of the benchmark test program which runs on the ARM CPU.Based on the above discussion, the thesis first select the gem5 simulator with the ARM instruction set, and design the algorithm to extract the micro architecture independent workload characteristics of the embedded benchmark test program. Then the embedded test set Mibench (benchmark test program group) is selected. The statistics of the instructions will be split into a number of micro operation, which will affect the counter count and bring some errors. Therefore, in order to verify the correctness of statistical results,it is necessary to observe the results of the trace information. Some characteristic parameters in gem5 lack the definition, such as the working set size, spatial locality, temporal locality, etc. It is required to construct the micro architecture independent workload feature plug-in in the simulator to carry out the effective statistics. Compared with the MICA simulator which is studied by Ghent University, the embedded benchmark's micro architecture independent workload characteristics are extracted and analyzed on the gem5 simulator. Finally, it is proved that the micro architecture independent workload characteristics which is extracted on the gem5 simulator is nothing to do with the specific micro architecture. It can be verified that the extracted micro architecture independent workload characteristic parameters are independent of specific micro architecture by changing the Cache's different configurations on the gem5.Experimental results show that the trend similarity of the micro architecture independent workload characteristics parameters based on two different platforms reached more than 90%. By changing the different gem5 cache configuration, the extracted micro architecture independent workload characteristics parameters variation was less than 5%.This thesis realized micro architecture independent workload characteristic parameters extraction in the gem5 simulator. The micro architecture independent workload characteristic parameters were analyzed, which can truly represent the internal behavior and essential feature of embedded benchmark programs, and can be used to extract representative subset from the test set.