Skip to content

Latest commit

 

History

History
43 lines (34 loc) · 2.05 KB

how_to_profile.md

File metadata and controls

43 lines (34 loc) · 2.05 KB

How to profile these benchmarks

Windows

We'll be profiling these benchmarks with Visual Studio. Make sure you have basic C/C++ payloads installed for it.

  1. Add the following beneath the [[bench]] line in Cargo.toml: 
    [profile.bench]
debug = true
  2. Build the benchmark without running it with cargo bench --no-run
  3. Locate the benchmark binary at target/release/deps/bench-*.exe. If there are multiple bench executables in deps, you can determine which one is the most recent by deleting all of them and rebuilding.
  4. Start Visual Studio and open a new project. Browse to the .exe location and open it.
  5. Right click on the bench-* project in the Solution Explorer and select Properties.
  6. Set the Arguments parameter to --bench <FILTER> where <FILTER> is the filter you would pass to cargo bench -- <FILTER>.
  7. Save the project properties and select Debug > Performance Profiler....
  8. Check CPU Usage and click Start. The benchmarks will run and upon finishing the data will be analyzed.
  9. Select the areas of the diagnostics session when your benchmark is running to analyze.

Linux

With the "pprof" feature enabled, the Criterion benchmarking library can be used to automatically output profiles with flamegraphs for each benchmark. Simply run the desired benchmarks while passing --profile-time=10 or similar to the benchmark binary. (Passing this argument probably requires specifying the "bench" benchmark specifically, as other test binaries that cargo thinks might have benchmarks in them do not understand the flag.)

Example command: cargo bench --bench bench --no-default-features --features fooencoder,pprof -- 'minecraft_savedata/fooencoder/serialize' --profile-time=10

This will run the "minecraft_savedata/fooencoder/serialize" benchmark and profile it, then save a flamegraph of the profile results as a .svg file to target/criterion/minecraft_savedata_fooencoder/serialize/profile/flamegraph.svg.

Other platforms

Know how to profile these benchmarks on other platforms? Open a pull request!