Crate statslicer

statslicer

Statslicer is a platform for running experiments and collecting their results.

use guacamole::combinators::*;
use guacamole::Guacamole;

use statslicer::{benchmark, black_box, statslicer_main, Bencher, Parameter, Parameters};

#[derive(Debug, Default, Eq, PartialEq)]
struct MyParameters {
    elements: usize,
}

impl Parameters for MyParameters {
    fn params(&self) -> Vec<(&'static str, Parameter)> {
        vec![
            ("elements", Parameter::Integer(self.elements as u64)),
        ]
    }
}

fn bench_sort(params: &MyParameters, b: &mut Bencher) {
    let mut guac = Guacamole::new(b.seed());
    let array = to_vec(constant(params.elements), any::<u64>)(&mut guac);
    let bin_searches = to_vec(constant(b.size()), any::<u64>)(&mut guac);
    b.run(|| {
        for needle in bin_searches.iter() {
            let _ = black_box(array.binary_search(needle));
        }
    });
}

benchmark! {
    name = my_sort_benchmark;
    MyParameters {
        elements in [0, 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024],
    }
    bench_sort
}

statslicer_main! { my_sort_benchmark }

Status

Experimental. Likely to change in the near future.

Scope

This library provides the statslicer benchmark tools and a binary to derive data and colate histograms.

Warts

Documentation

The latest documentation is always available at docs.rs.

Macros

benchmark

A macro for defining a benchmark sweep.

statslicer_main

The macro for creating main functions.

Structs

Bencher

BenchmarkOptions

Options for the benchmark.

Cycle

Moments

Moments are the statistical moments of mean (m1), standard deviation (m2), skewness (m3) and kurtosis (m4). When a distribution goes long tailed, skewness and kurtosis blow up, so the hope is that this type will be good for monitoring general “no bad tail” insights.

UnboundParameters

Unbound parameters convers from a string like foo,bar,baz,quux.

UntypedParameters

Untyped parameters converts from a string like foo=1,bar=3.14,baz=true,quux. The input is untyped, and the result is free-form, as opposed to a type like what implements Parameters.

Enums

Parameter

Parameter binds a typed value to an experiment.

Type

Type captures the type of a Parameter.

Traits

Parameters

Parameters provides the parameters for an experiment/benchmark.

Functions

benchmark

A stub used to run the benchmarks, so that profilers can zero counters on entry and count them on exit.

benchmark_main

black_box

Try to prevent the compiler from eliminating code.

compute_difference

experiment_and_parameters

increment_indices

summarize