google/autofdo

What is it?

Key features

• 01Collects sampled profiles from Linux perf without program instrumentation.
• 02Converts sampled profiles into a format the compiler can use for optimization.
• 03Provides separate LLVM and GCOV tool builds selected through CMake.
• 04Builds with clang or g++ on Ubuntu 20.04, 22.04 and CentOS 9.

Use cases

• Improving the runtime performance of production C++ binaries using real workload profiles.
• Building AutoFDO tooling for LLVM-based compiler pipelines on Linux servers.
• Experimenting with profile-guided optimization without changing source code.

Quick start

1. 1

   Install prerequisites on Ubuntu 20.04 or 22.04

   sudo apt install libunwind-dev libgflags-dev libssl-dev libelf-dev protobuf-compiler cmake libzstd-dev clang g++
2. 2

   Clone the repository with submodules

   git clone --recursive --depth 1 https://github.com/google/autofdo.git
3. 3

   Create a build directory and build the LLVM tools

   cd autofdo && mkdir build && cd build && cmake -DENABLE_TOOL=LLVM -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED=On ../
4. 4

   Build the autofdo tools

   cmake -DENABLE_TOOL=GCOV -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED=On ../ && make -j 4

Things to know

• !Build instructions target Ubuntu 20.04, Ubuntu 22.04 and CentOS 9 only.
• !Ubuntu 20.04 needs a newer cmake than the one it ships, because the bundled LLVM build fails with the default version.
• !The Propeller codebase has moved to google/llvm-propeller.

In collections

Topics

FAQ

What is autofdo?

Toolchain for AutoFDO, a profile-guided optimization workflow that uses sampled hardware profiles from Linux perf to improve compiled binaries. AutoFDO is a set of tools for automatic feedback-directed optimization. Instead of instrumenting a program, it collects sampled execution profiles from hardware performance counters using Linux perf, converts them into a compiler-readable profile, and feeds that profile back into the compiler to guide optimization decisions.

Is autofdo free to use? What license does it have?

autofdo is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.

Is autofdo still maintained?

Updated within 6 months. The last commit to autofdo was on Sep 30, 2026.

What can I use autofdo for?

Improving the runtime performance of production C++ binaries using real workload profiles. Building AutoFDO tooling for LLVM-based compiler pipelines on Linux servers. Experimenting with profile-guided optimization without changing source code.

How do I install autofdo?

Install prerequisites on Ubuntu 20.04 or 22.04: sudo apt install libunwind-dev libgflags-dev libssl-dev libelf-dev protobuf-compiler cmake libzstd-dev clang g++ Clone the repository with submodules: git clone --recursive --depth 1 https://github.com/google/autofdo.git Create a build directory and build the LLVM tools: cd autofdo && mkdir build && cd build && cmake -DENABLE_TOOL=LLVM -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED=On ../ Build the autofdo tools: cmake -DENABLE_TOOL=GCOV -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED=On ../ && make -j 4

Where can I get autofdo?

The source code and installation instructions are on GitHub: https://github.com/google/autofdo