Benchmarking with Criterion.rs

[RobertJacobsonCDC]

• Added Criterion.rs and criterion.toml for cargo-criterion.
• Created benches directory, example_births_deaths, example_basic_infection benchmarks.
• Wrote a README.md explaining how to run benchmarks and profile with samply.

[RobertJacobsonCDC]

Could someone verify whether the pre-commit failure to compile issue is spurious? It works on my machine, so I wonder if it just doesn't know what to do with the Criterion benchmark code.

[k88hudson-cfa]

@RobertJacobsonCDC I can reproduce locally running clippy with --all-targets

[ekr-cfa]

I'm kind of suspicious of this overall structure where we have to incorporate all the files from the programs we're benchmarking. Is there instead some way to make them a library and then pull in the library separately from the criterion driver and the example main? Open to other alternatives, but I don't think linking to all the individual files is good practice.

[RobertJacobsonCDC]

The libraries for the examples can now have their own target directory.

[RobertJacobsonCDC]

The examples are largely unchanged. The paths to input and output files were adjusted for them.

Running benchmarks:

cargo bench --bench example_births_deaths
cargo bench --bench example_basic_infection
cargo bench -- --bench example_benches
cargo bench

Running examples:

cargo run --example births-deaths
cargo run --example basic-infection

[ekr-cfa]

This structure still doesn't feel right. Added a comment inline, but happy to discuss live.

[ekr-cfa]

This code appears to largely duplicate basic-infection/lib.rs, which isn't great.

The way I expected this to work is that each example lib exposes a single function that encapsulates the entire simulation and is called something like main_inner() and is effectively main. And then you would call main_inner() from either the true main or from the benchmarking framework.

Is there a reason that can't work?