Crates.io | criterion-perf-events |
lib.rs | criterion-perf-events |
version | 0.4.0 |
source | src |
created_at | 2020-05-02 14:31:22.532902 |
updated_at | 2023-05-26 21:04:42.871931 |
description | Measure perf events for criterion |
homepage | |
repository | https://github.com/jbreitbart/criterion-perf-events |
max_upload_size | |
id | 236582 |
size | 48,836 |
This is a measurement plugin for Criterion.rs to measure events of the Linux perf interface.
Criterion-perf-events uses the perfcnt
crate and supports events provided by this crate. If you are interested in more details, please take a look at the events listed here:
If you get a "Permission denied" error, update perf_event_paranoid
:
sudo sh -c 'echo 1 >/proc/sys/kernel/perf_event_paranoid'
For further details please take a look at the following link.
The following code shows how to count retired instructions.
use criterion::{criterion_group, criterion_main, BenchmarkId, black_box, Criterion};
use criterion_perf_events::Perf;
use perfcnt::linux::HardwareEventType as Hardware;
use perfcnt::linux::PerfCounterBuilderLinux as Builder;
fn fibonacci_slow(n: usize) -> usize {
match n {
0 => 1,
1 => 1,
n => fibonacci_slow(n - 1) + fibonacci_slow(n - 2),
}
}
fn bench(c: &mut Criterion<Perf>) {
let mut group = c.benchmark_group("fibonacci");
let fibo_arg = 30;
group.bench_function(BenchmarkId::new("slow", fibo_arg), |b| {
b.iter(|| fibonacci_slow(black_box(fibo_arg)))
});
group.finish()
}
criterion_group!(
name = instructions_bench;
config = Criterion::default().with_measurement(Perf::new(Builder::from_hardware_event(Hardware::Instructions)));
targets = bench
);
criterion_main!(instructions_bench);
run with:
cargo criterion
Open target/criterion/reports/index.html
to view detailed results with plots.
For all event types (Hardware::Instructions
, Hardware::CacheMisses
...) criterion will always report cycles as the unit.
Note that your event type is what is being shown, not CPU cycles.