snail's Introduction

TL;DR

A minimal benchmark library.

TODO:

The current implementation uses std.time.Timer for the measurement. It's not 100% clear what happens if the current thread gets suspended and resumes later at some point. Will the Timer's monotonic clock skip the ideal time or include it with the measurement?
For some reason, explicit measurement with ready() took longer than the run(). Might be additional overhead from start() and stop(). Nevertheless figure out what's causing this delay!
- It's the overhead for sure! and the ctx itself taking majority of the times for simple add(10, 20). Figure out how to eliminate this time.
Suspend and resume start() and stop() n amount of times in real code to benchmark a non-contagious code block.

Examples

const snail = @import("snail.zig");

pub fn main() void {
    // With default code iteration
    snail.benchmark("addition", measure_add, null);
    snail.benchmark("addition", measure_add_explicit, null);

    // With manual code iteration
    snail.benchmark("addition", measure_add, 1000);
    snail.benchmark("addition", measure_add_explicit, 1000);
}


fn add(x: i32, y: i32) i32 {
    return x + y;
}

fn measure_add(ctx: *snail.Context) void {
    while (ctx.run()) {
        _ = add(56, 44);
    }
}

fn measure_add_explicit(ctx: *snail.Context) void {
    while (ctx.ready()) {
        ctx.start();
        defer ctx.stop()
        _ = add(56, 44);
    }
}

Recommend Projects