In part two of the final TCG Continuous Benchmarking report, the same procedures introduced in part one are used for inspecting the performance of QEMU system mode emulation. The only difference is instead of emulating the same OS for all targets, different images where selected from a Qemu-devel thread and the official QEMU documentation.

The final report of the TCG Continuos Benchmarking project introduces basic performance measurements for QEMU system mode emulation. Boot-up time and number of executed instructions are compared for the emulation of five different targets. The report also presents a new tool for finding the topN most executed functions in the emulation process.

QEMU currently lacks a system for measuring the performance of targets automatically. The previous reports introduced different tools and methods for locating performance regressions, but all of them had to be manually executed by the user when needed. This report presents a new nightly tests system that runs automatically each night. After the execution is completed, it sends a report to the QEMU mailing list with the performance measurements of seventeen different QEMU targets, and how these measurements compare to previously obtained ones.

In this report, a method for measuring the TCG emulation efficiency of QEMU is presented. This is achieved for seventeen different targets by comparing the number of guest instructions (running the program natively on the target) and the number of QEMU instructions (running the program through QEMU). For each target, the ratio between these two numbers presents a rough estimation of the emulation efficiency for that target.

This report presents a performance comparison between two different QEMU builds, GCC and Clang. To provide a variety of test workloads, five new benchmarks are also introduced in the report. For each benchmark, the performance of QEMU is compared across the two builds for seventeen different targets.

This report concludes a mini series of three reports that presented a comparison between QEMU 5.0 and QEMU 5.1-pre-soft-freeze. The first report in the series presented the comparison results as well as an analysis of the performance degradation introduced in the PowerPC targets using KCachegrind. The second report presented two new scripts for inspecting performance changes as well as the analysis of three other targets. Now it's time to find the commits responsible for the performance changes discussed in the two previous reports. This report provides a new script for performing the process of locating those commits automatically.