sadads1337 / cache-benchmark

Tiny CPU caches benchmark

Main purpose of this application is to show, sizes, and speedup of using cpu caches. Special thanks for google-benchmark library developers, which makes it easy to measure. Also thanks for google_benchmark_plot library for an easy implementation of results plotting in python.

My own machine configuration and results

Laptop Intel i9 2,3 GHz, 8 cores, 16 threads(x2 hyper-threading)

CPU Caches:
  L1 Data 32K (x8)
  L1 Instruction 32K (x8)
  L2 Unified 262K (x8)
  L3 Unified 16777K (x1)

Build

The recommended way is to use conan package manager.

• Firstly, make a build directory and call conan install <path-to-conanfile.txt> to install all dependencies(There is only one - google benchmark library). Some users might want to check it's a truth, it's really easy with call conan info <path-to-conanfile.txt>.

• Then configure cmake. For example, in your build directory call cmake -G Ninja <path-to-root-CMakeLists.txt> -DCMAKE_BUILD_TYPE=Release.

• Final step is to call ninja or yours own generator in build directory to build executable and run it with ./cache-benchmark command.

Run and generate plot

• Execute ./cache-benchmark and watch console output.

• If you want to see plots, execute you benchmark with ./cache-benchmark --benchmark_format=csv > benchmark.csv to save results into csv. Then run patched plot.py from google_benchmark_plot directory. Example call is plot.py -f benchmark.csv --logx.