Rust中如何用Divan对同一函数在不同Feature下做基准测试？

当然可以实现同一函数在Feature启用/禁用状态下的基准测试，不管用Divan还是其他工具（比如Criterion.rs），核心思路都是编译不同Feature配置的二进制文件，分别运行基准测试。以下是具体实现方法：

用Divan实现的步骤 #

1. 配置Cargo.toml #

先把依赖和Feature定义好：

[package]
name = "bench-feature-demo"
version = "0.1.0"
edition = "2021"

[dependencies]
rayon = { version = "1.7", optional = true }

[dev-dependencies]
divan = "0.1.10"

[features]
parallel = ["dep:rayon"]

2. 编写基准测试代码 #

基于你提供的代码调整，确保逻辑在不同Feature下正常工作：

use divan::counter::ItemsCount;
use divan::{black_box, Bencher};

fn main() {
    divan::main()
}

#[cfg(feature = "parallel")]
use rayon::iter::ParallelIterator;

fn compute_foo(size: usize) -> usize {
    #[cfg(feature = "parallel")]
    {
        (0..size).into_par_iter()
    }
    #[cfg(not(feature = "parallel"))]
    { (0..size).into_iter() }.map(|i| i * 123 / 4).sum()
}

#[divan::bench(args = [16, 32, 64, 128, 256, 512, 1024, 2048, 4096])]
fn bench_compute_foo(bencher: Bencher, size: usize) {
    bencher
        .input_counter(|| ItemsCount::new(size))
        .bench_refs(|| black_box(compute_foo(size)));
}

3. 分别运行基准测试 #

• 禁用parallel Feature时：

cargo divan --no-default-features

• 启用parallel Feature时：

cargo divan --features parallel

可以把两次的输出结果保存下来，直接对比性能差异。

用Criterion.rs实现的备选方案 #

如果你想换用更主流的基准测试工具，Criterion.rs的实现逻辑类似：

1. 配置Cargo.toml #

[package]
name = "bench-feature-demo"
version = "0.1.0"
edition = "2021"

[dependencies]
rayon = { version = "1.7", optional = true }

[dev-dependencies]
criterion = { version = "0.5", features = ["html_reports"] }

[features]
parallel = ["dep:rayon"]

2. 编写基准测试代码 #

use criterion::{criterion_group, criterion_main, Criterion};
use rayon::iter::ParallelIterator;

fn compute_foo(size: usize) -> usize {
    #[cfg(feature = "parallel")]
    {
        (0..size).into_par_iter()
    }
    #[cfg(not(feature = "parallel"))]
    { (0..size).into_iter() }.map(|i| i * 123 / 4).sum()
}

fn bench_compute_foo(c: &mut Criterion) {
    let test_sizes = [16, 32, 64, 128, 256, 512, 1024, 2048, 4096];
    for &size in &test_sizes {
        c.bench_function(&format!("compute_foo/{}", size), |b| {
            b.iter(|| compute_foo(size))
        });
    }
}

criterion_group!(benches, bench_compute_foo);
criterion_main!(benches);

3. 运行测试 #

• 禁用Feature：

cargo bench --no-default-features

• 启用Feature：

cargo bench --features parallel

Criterion会生成HTML报告，方便直观对比两次测试的性能数据。

内容的提问来源于stack exchange，提问作者Netwave