CPU Microbenchmarks

Copyright (c) 2024-2025 Antmicro

This repository allows for microbenchmarking Renode CPU implementation.

Dependencies

You'll need Google Benchmark to build these microbenchmarks. To install it, follow these steps:

git clone https://github.com/google/benchmark.git

# Build google benchmark
cd benchmark
cmake -E make_directory "build"
cmake -E chdir "build" cmake -DBENCHMARK_DOWNLOAD_DEPENDENCIES=on -DCMAKE_BUILD_TYPE=Release ../
cmake --build "build" --config Release -j$(nproc)

# Install globally
sudo cmake --build "build" --config Release --target install

You'll also need build-essential, cmake and libelf-dev, installed e.g. by apt install build-essential cmake libelf-dev.

Usage

To get the benchmark binary, call setup-benchmark.sh to clone and build the relevant branches.

Then, use run.sh to execute the benchmarks. Output will be generated in the repository directory as .csv files.

If you make changes to the benchmarking code and want to recompile, use make.sh.

The scripts accept a positional parameter, which is a regex string used to match which branch to use from config.sh. If none is provided, all are used.

For example, if you'd like to benchmark all configured branches then run:

./run.sh

and if you'd like to see that all benchmarks work correctly, dry-run them:

./run.sh ".*" --benchmark_dry_run

or if you want to benchmark a specific branch:

./run.sh "public-hst-weak-ir@.*"

The run.sh script passes additional positional parameters to the benchmark binary. Use run.sh ".*" --help for a list of options (the standard Google Benchmark options).

Once done, artifacts can be cleaned up using clean.sh.

Figures can be generated using python3 benchmark-plot.py "path/to/result-csvs/" "baseline@short_sha"