nanobench

Hint

View and download nanobench on Github. Get the latest release.

ankerl::nanobench is a platform independent microbenchmarking library for C++11/14/17/20.

#define ANKERL_NANOBENCH_IMPLEMENT
#include <nanobench.h>

int main() {
    double d = 1.0;
    ankerl::nanobench::Bench().run("some double ops", [&] {
        d += 1.0 / d;
        if (d > 5.0) {
            d -= 5.0;
        }
        ankerl::nanobench::doNotOptimizeAway(d);
    });
}

The whole executable runs for ~60ms and prints

|               ns/op |                op/s |    err% |          ins/op |          cyc/op |    IPC |         bra/op |   miss% |     total | benchmark
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
|                7.52 |      132,948,239.79 |    1.1% |            6.65 |           24.07 |  0.276 |           1.00 |    8.9% |      0.00 | `some double ops`

Which github renders like

ns/op	op/s	err%	ins/op	cyc/op	IPC	bra/op	miss%	total	benchmark
7.52	132,948,239.79	1.1%	6.65	24.07	0.276	1.00	8.9%	0.00	`some double ops`

The benchmarked code takes 7.52 nanoseconds to run, so ~133 million times per seconds. Measurements fluctuate by 1.1%. On average 6.65 instructions are executed in 24.07 CPU cycles, resulting in 0.276 instructions per cycle. A single branch is in the code, which branch prediction missed in 8.9% of the cases. Total runtime of the benchmark with the name some double ops is 0.00, so just a few milliseconds.

Design Goals

Ease of use: Simple but powerful API, fast compile times, easy to integrate anywhere.
Fast: Get accurate results as fast as possible
Accurate: Get deterministic, repeatable, and accurate results that you can make sound decisions on.
Robust: Be robust against outliers, warn if results are not reliable.