BenchmarkV3

Benchmark how fast your system compiles C++ projects. Created to answer the question of whether it is worth investing a new dev machine.

This benchmark measures the time to build LLVM's llc using clang, cmake and ninja. It uses prebuilt toolchains and build for the same target sysroot so hopefully it benchmarks the same thing on different systems.

Run prebuilt

Go to the release page to download the latest build and run directly.

V=linux-amd64
curl -o BenchmarkV3 -L https://github.com/afq984/BenchmarkV3/releases/download/latest/BenchmarkV3-$V
chmod +x BenchmarkV3
./BenchmarkV3

Run from source

git clone https://github.com/afq984/BenchmarkV3
cd BenchmarkV3
go build .
./BenchmarkV3

BenchmarkV3 command line flags

• --quick: Do a quick build instead. This measures the build time of llvm-cxxfilt instead of llc which has only about 1/8 build targets.

• --detect: Detect the system only. Does not actually run the benchmark.

• -c <config>: Use a specific config:

  • auto - Auto detect the config to use (default)
  • linux-amd64 - Works on ubuntu 20.04 or up-to-date linux systems
  • linux-amd64-ubuntu1604 - Try this one on older linux systems
  • linux-arm64 - For ARM64 linux systems. To use this config, you need to have ninja on your system.
  • macos-amd64 - Use this on macOS. Should also work on M1 through Rosetta.

FAQ

• Why no macos-arm64 config?

  LLVM 13.0.0 does not have a arm64 clang+llvm binary for macOS.

• Missing libtinfo.so.6 or libtinfo.so.5 on linux.

  You need to install libncurses6 or libncurses5, respectively. The naming of the package depends on your distribution.

  Currently:

  • requires libtinfo.so.6: linux-amd64
  • requires libtinfo.so.5: linux-amd64-ubuntu1604, linux-arm64