This report builds up on the previous one by presenting two new Python scripts that facilitates the process of displaying the executed QEMU helpers and function callees without the need of setting up KCachegrind. The PowerPC performance degradation is then re-analysed using the new scripts. The report also introduces the analysis of three other targets, hppa and sh4, explaining why they were not affected the same way as ppc, and mips, explaining why it showed an increase in performance.

As we get into the soft freeze period starting tomorrow, it is necessary to have a reference of whether there were some performance improvements or degradations across targets between QEMU 5.0 and the current version of QEMU. The report dissects the execution of 17 different QEMU targets and compares the results obtained from running with QEMU 5.0 and QEMU 5.1-pre-soft-freeze.

The previous report presented an overview of measuring basic performance metrics of QEMU, and one of these metrics, naturally, was the total number of executed host instructions. This report further utilizes Callgrind to break down that total number into numbers that correspond to three main parts of QEMU operation; code generation, JIT-ed code execution, and helpers execution.

Welcome to the TCG Continuous Benchmarking project! This report presents two Linux profiling tools - Perf and Callgrind. It gives you an overview of their setup, usage, pros and cons. The motivation is to provide you with a better understanding of them. This way, you will be able to make decisions regarding their usage in a better and easier way. You will also learn two ways of finding top N most executed QEMU functions of your scenario, without even having to know absolutely any detail of the used profiling tools!