Skip to content

fix(divan): support threaded walltime benchmarks #189

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
not-matthias wants to merge 2 commits into
base: main
Choose a base branch
from
Open

fix(divan): support threaded walltime benchmarks #189

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cbde9bb
test(divan): add native threaded benchmark fixture
not-matthias Sep 1, 2026
224cf47
fix(divan): support threaded walltime benchmarks
not-matthias Sep 1, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
86 changes: 53 additions & 33 deletions crates/codspeed/src/instrument_hooks/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,18 +6,30 @@ mod linux_impl {

use super::ffi;
use std::ffi::CString;
use std::sync::OnceLock;
use std::sync::{Mutex, OnceLock};

static INSTRUMENT_HOOKS_LOCK: Mutex<()> = Mutex::new(());

fn with_transport_lock<T>(operation: impl FnOnce() -> T) -> T {
let _guard = INSTRUMENT_HOOKS_LOCK
.lock()
.expect("InstrumentHooks mutex poisoned");
operation()
}

#[derive(PartialEq)]
pub struct InstrumentHooks(*mut ffi::InstrumentHooks);

// SAFETY: The pointer is immutable after construction, and every access to
// any instance handle or process-global instrument-hooks state is serialized
// by `INSTRUMENT_HOOKS_LOCK`.
unsafe impl Send for InstrumentHooks {}
unsafe impl Sync for InstrumentHooks {}

impl InstrumentHooks {
#[inline(always)]
pub fn new() -> Option<Self> {
let ptr = unsafe { ffi::instrument_hooks_init() };
let ptr = with_transport_lock(|| unsafe { ffi::instrument_hooks_init() });
if ptr.is_null() {
None
} else {
Expand Down Expand Up @@ -76,12 +88,13 @@ mod linux_impl {

#[inline(always)]
pub fn is_instrumented(&self) -> bool {
unsafe { ffi::instrument_hooks_is_instrumented(self.0) }
with_transport_lock(|| unsafe { ffi::instrument_hooks_is_instrumented(self.0) })
}

#[inline(always)]
pub fn start_benchmark(&self) -> Result<(), u8> {
let result = unsafe { ffi::instrument_hooks_start_benchmark(self.0) };
let result =
with_transport_lock(|| unsafe { ffi::instrument_hooks_start_benchmark(self.0) });
if result == 0 {
Ok(())
} else {
Expand All @@ -91,7 +104,8 @@ mod linux_impl {

#[inline(always)]
pub fn stop_benchmark(&self) -> Result<(), u8> {
let result = unsafe { ffi::instrument_hooks_stop_benchmark(self.0) };
let result =
with_transport_lock(|| unsafe { ffi::instrument_hooks_stop_benchmark(self.0) });
if result == 0 {
Ok(())
} else {
Expand All @@ -103,9 +117,9 @@ mod linux_impl {
pub fn set_executed_benchmark(&self, uri: &str) -> Result<(), u8> {
let pid = std::process::id() as i32;
let c_uri = CString::new(uri).map_err(|_| 1u8)?;
let result = unsafe {
let result = with_transport_lock(|| unsafe {
ffi::instrument_hooks_set_executed_benchmark(self.0, pid, c_uri.as_ptr())
};
});
if result == 0 {
Ok(())
} else {
Expand All @@ -117,9 +131,9 @@ mod linux_impl {
pub fn set_integration(&self, name: &str, version: &str) -> Result<(), u8> {
let c_name = CString::new(name).map_err(|_| 1u8)?;
let c_version = CString::new(version).map_err(|_| 1u8)?;
let result = unsafe {
let result = with_transport_lock(|| unsafe {
ffi::instrument_hooks_set_integration(self.0, c_name.as_ptr(), c_version.as_ptr())
};
});
if result == 0 {
Ok(())
} else {
Expand All @@ -131,22 +145,24 @@ mod linux_impl {
pub fn add_benchmark_timestamps(&self, start: u64, end: u64) {
let pid = std::process::id() as i32;

unsafe {
ffi::instrument_hooks_add_marker(
self.0,
pid,
ffi::MARKER_TYPE_BENCHMARK_START as u8,
start,
)
};
unsafe {
ffi::instrument_hooks_add_marker(
self.0,
pid,
ffi::MARKER_TYPE_BENCHMARK_END as u8,
end,
)
};
with_transport_lock(|| {
unsafe {
ffi::instrument_hooks_add_marker(
self.0,
pid,
ffi::MARKER_TYPE_BENCHMARK_START as u8,
start,
)
};
unsafe {
ffi::instrument_hooks_add_marker(
self.0,
pid,
ffi::MARKER_TYPE_BENCHMARK_END as u8,
end,
)
};
});
}

#[inline(always)]
Expand Down Expand Up @@ -175,14 +191,14 @@ mod linux_impl {
let c_section = CString::new(section_name).map_err(|_| 1u8)?;
let c_key = CString::new(key).map_err(|_| 1u8)?;
let c_value = CString::new(value).map_err(|_| 1u8)?;
let result = unsafe {
let result = with_transport_lock(|| unsafe {
ffi::instrument_hooks_set_environment(
self.0,
c_section.as_ptr(),
c_key.as_ptr(),
c_value.as_ptr(),
)
};
});
if result == 0 {
Ok(())
} else {
Expand All @@ -192,7 +208,9 @@ mod linux_impl {

pub fn write_environment(&self) -> Result<(), u8> {
let pid = std::process::id() as i32;
let result = unsafe { ffi::instrument_hooks_write_environment(self.0, pid) };
let result = with_transport_lock(|| unsafe {
ffi::instrument_hooks_write_environment(self.0, pid)
});
if result == 0 {
Ok(())
} else {
Expand All @@ -201,20 +219,22 @@ mod linux_impl {
}

pub fn disable_callgrind_markers() {
unsafe {
with_transport_lock(|| unsafe {
ffi::instrument_hooks_set_feature(
ffi::instrument_hooks_feature_t_FEATURE_DISABLE_CALLGRIND_MARKERS.into(),
true,
)
};
});
}
}

impl Drop for InstrumentHooks {
fn drop(&mut self) {
if !self.0.is_null() {
unsafe { ffi::instrument_hooks_deinit(self.0) };
}
with_transport_lock(|| {
if !self.0.is_null() {
unsafe { ffi::instrument_hooks_deinit(self.0) };
}
});
}
}
}
Expand Down
9 changes: 9 additions & 0 deletions crates/divan_compat/benches/thread_example.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,15 @@ fn fib_in_thread_bench_local(bencher: codspeed_divan_compat::Bencher, n: usize)
})
}

#[cfg_attr(
not(codspeed),
codspeed_divan_compat::bench(threads = 4, sample_count = 1)
)]
#[cfg_attr(codspeed, codspeed_divan_compat::bench(sample_count = 1))]
fn divan_threads() {
codspeed_divan_compat::black_box(fibo(20));
}

fn main() {
codspeed_divan_compat::main();
}
44 changes: 29 additions & 15 deletions crates/divan_compat/divan_fork/src/bench/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -698,10 +698,18 @@ impl<'a> BenchContext<'a> {
};

// Sample loop:
let ([start, end], alloc_info) =
let ([start, end], alloc_info, [benchmark_start, benchmark_end]) =
record_sample(sample_size as usize, barrier.as_ref(), &mut count_input);

RawSample { start, end, timer, alloc_info, counter_totals }
RawSample {
start,
end,
benchmark_start,
benchmark_end,
timer,
alloc_info,
counter_totals,
}
};

// Sample loop:
Expand All @@ -723,6 +731,11 @@ impl<'a> BenchContext<'a> {
std::slice::from_raw_parts(raw_samples.as_ptr().cast(), raw_samples.len())
}
};
let benchmark_start =
raw_samples.iter().map(|sample| sample.benchmark_start).min().unwrap();
let benchmark_end =
raw_samples.iter().map(|sample| sample.benchmark_end).max().unwrap();
InstrumentHooks::instance().add_benchmark_timestamps(benchmark_start, benchmark_end);

// If testing, exit the benchmarking loop immediately after timing a
// single run.
Expand Down Expand Up @@ -825,8 +838,11 @@ impl<'a> BenchContext<'a> {
gen_input: impl Fn() -> I,
benched: impl Fn(&UnsafeCell<MaybeUninit<I>>) -> O,
drop_input: impl Fn(&UnsafeCell<MaybeUninit<I>>),
) -> impl Fn(usize, Option<&Barrier>, &mut dyn FnMut(&I)) -> ([Timestamp; 2], ThreadAllocInfo)
{
) -> impl Fn(
usize,
Option<&Barrier>,
&mut dyn FnMut(&I),
) -> ([Timestamp; 2], ThreadAllocInfo, [u64; 2]) {
// We defer:
// - Usage of `gen_input` values.
// - Drop destructor for `O`, preventing it from affecting sample
Expand Down Expand Up @@ -897,8 +913,9 @@ impl<'a> BenchContext<'a> {
// benchmarking.
let sample_start: UntaggedTimestamp;
let sample_end: UntaggedTimestamp;
let benchmark_start: u64;
let benchmark_end: u64;

let instrument_hooks = InstrumentHooks::instance();
if size_of::<I>() == 0 && (size_of::<O>() == 0 || !mem::needs_drop::<O>()) {
// Use a range instead of `defer_store` to make the benchmarking
// loop cheaper.
Expand All @@ -915,7 +932,7 @@ impl<'a> BenchContext<'a> {

sync_threads(true);

let start_time = InstrumentHooks::current_timestamp();
benchmark_start = InstrumentHooks::current_timestamp();
sample_start = UntaggedTimestamp::start(timer_kind);

// Sample loop:
Expand All @@ -928,8 +945,7 @@ impl<'a> BenchContext<'a> {
}

sample_end = UntaggedTimestamp::end(timer_kind);
let end_time = InstrumentHooks::current_timestamp();
instrument_hooks.add_benchmark_timestamps(start_time, end_time);
benchmark_end = InstrumentHooks::current_timestamp();

sync_threads(false);
save_alloc_info();
Expand Down Expand Up @@ -972,7 +988,7 @@ impl<'a> BenchContext<'a> {
let defer_slots_iter = defer_slots_slice.iter();

sync_threads(true);
let start_time = InstrumentHooks::current_timestamp();
benchmark_start = InstrumentHooks::current_timestamp();
sample_start = UntaggedTimestamp::start(timer_kind);

// Sample loop:
Expand All @@ -987,8 +1003,7 @@ impl<'a> BenchContext<'a> {
}

sample_end = UntaggedTimestamp::end(timer_kind);
let end_time = InstrumentHooks::current_timestamp();
instrument_hooks.add_benchmark_timestamps(start_time, end_time);
benchmark_end = InstrumentHooks::current_timestamp();
sync_threads(false);
save_alloc_info();

Expand Down Expand Up @@ -1028,7 +1043,7 @@ impl<'a> BenchContext<'a> {
let defer_inputs_iter = defer_inputs_slice.iter();

sync_threads(true);
let start_time = InstrumentHooks::current_timestamp();
benchmark_start = InstrumentHooks::current_timestamp();
sample_start = UntaggedTimestamp::start(timer_kind);

// Sample loop:
Expand All @@ -1039,8 +1054,7 @@ impl<'a> BenchContext<'a> {
}

sample_end = UntaggedTimestamp::end(timer_kind);
let end_time = InstrumentHooks::current_timestamp();
instrument_hooks.add_benchmark_timestamps(start_time, end_time);
benchmark_end = InstrumentHooks::current_timestamp();
sync_threads(false);
save_alloc_info();

Expand All @@ -1065,7 +1079,7 @@ impl<'a> BenchContext<'a> {
[sample_start.into_timestamp(timer_kind), sample_end.into_timestamp(timer_kind)]
};

(interval, saved_alloc_info)
(interval, saved_alloc_info, [benchmark_start, benchmark_end])
}
}

Expand Down
2 changes: 2 additions & 0 deletions crates/divan_compat/divan_fork/src/stats/sample.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,8 @@ pub(crate) struct TimeSample {
pub(crate) struct RawSample {
pub start: Timestamp,
pub end: Timestamp,
pub benchmark_start: u64,
pub benchmark_end: u64,
pub timer: Timer,
pub alloc_info: ThreadAllocInfo,
pub counter_totals: [u128; KnownCounterKind::COUNT],
Expand Down
Loading