OpenPerf Project

OpenPerf is an open source benchmark for evaluating test the performance of the microarchitecture in processor core, developed and be maintained by One Student One Chip Project.

How to use

This project plans to integrate with YSYX's workbench and NJU ProjectN.

To use openperf in baseline practice, clone it into your project's AbstractMachine Kernels.

cd am-kernels
git clone --recursive-submodules git@github.com:OSCPU/openperf.git
cd openperf

The make recipe, like other benchmark programs in am-kernel, allows you to choose the architectures and the test scale.

Building

You can use script/build.sh to build this project. build.sh take 2 flags: -a specify architecture of abstract-machine and -s specify size of test programs you want to build (test, ref, huge).

# Make sure you have set AM_HOME correctly
scripts/build.sh build -s test,ref -a native

# Find binary images under build/images
ls build/images

When cross compiling, you need to explicitly provide the variable CROSS_COMPILE to make, so it knows where to find the toolchain. For example, when compiling to riscv32:

CROSS_COMPILE=riscv32-unknown-elf- scripts/build.sh build -s test,ref -a native

Under the hood, this script first build abstract-machine. Different from NJU-ProjectN/abstract-machine, abstract-machine in this project is built with command make -C "$AM_HOME" ARCH=$ARCH install. Header files, archives and pkgconfig file will be installed under abstract-machine/build/install by default. We then use pkg-config to determine what flags is needed to build and link against abstract-machine.

# What script/build.sh does (Assuming AM_HOME and ARCH is set):
make -C "$AM_HOME" ARCH=$ARCH install 
make ARCH=$ARCH mainargs=<size> all

Run benchmark

After building the images, you can run this benchmark with scripts/run.py. Note that tqdm and pyyaml is needed, install them with

pip3 install pyyaml tqdm

scripts/run.py takes a YAML configuration file, checkout test_config.example.yaml for reference. Run the benchmark with:

python3 scripts/run.py test_config.example.yaml

You can filter tests with tags, a test is selected if it has any of the given tag. For example, if you only want to run tests with the tag native, you can run run.py with

python3 scripts/run.py test_config.example.yaml --tags native

Benchmark Programs

• Stream Memory Access: STREAM
• Irregular Memory Access: mcf
• Float Memory: GEMM
• Footprint: Gsim and essent which simulate various RISC-V processor cores such as riscv-mini, Nutshell, Rocket Core, BOOM and XiangShan
• Branch Prediction: TCC
• Floating-point Arithmetic: Linpack, Whetstone
• Utils: soft-fp(cyl), abstract-machine, openlibm

Community

One Student One Chip Project

"One Student One Chip" is a public welfare learning program to supply talents & trained students to chip industry, which means registering and studying "One Student One Chip" is free. In the mainline, even if he or she has zero experience or skills, every student could learn deeply about how computer system works and create their own processor cores from scratch.

Visit the main page to know more information.

Ranking (Coming Soon)

The One Student One Chip Project is planning to build a ranking system for students! You can run OpenPerf with your NEMU or NPC to find out your performance and compare with others.

Contributing

Roadmap

• Transplant benchmark programs
• Test the sensitivity of various programs
• Collect performance data from various open source processor cores
• Get early access

Acknowledgement

Thanks to the Open Source Community for developing related programs & tools included in OpenPerf. To know more copyright info you can check the Licence headers in programs' code. The modification and additional features & code written by us is under Mulan PSL v2 Licence.

Copyright (c) 2024 One Student One Chip Project OpenPerf is licensed under Mulan PSL v2. You can use this software according to the terms and conditions of the Mulan PSL v2.

You may obtain a copy of Mulan PSL v2 at: http://license.coscl.org.cn/MulanPSL2

THIS SOFTWARE IS PROVIDED ON AN “AS IS” BASIS, WITHOUT warranties of any kind, either express or implied, including but not limited to non-infringement, merchantability or fit for a particular purpose. See the Mulan PSL v2 for more details.