https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122878
Jonathan Wakely <redi at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |redi at gcc dot gnu.org
Ever confirmed|0 |1
Status|UNCONFIRMED |ASSIGNED
Target Milestone|--- |16.0
Last reconfirmed| |2025-11-26
--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> ---
This patch fixes it:
--- a/libstdc++-v3/src/c++20/atomic.cc
+++ b/libstdc++-v3/src/c++20/atomic.cc
@@ -669,11 +669,13 @@ __wait_until_impl([[maybe_unused]] const void* __addr,
__wait_args_base& __args,
if (__args & __wait_flags::__do_spin)
{
- auto __res = __detail::__spin_until_impl(__wait_addr, __args, __atime);
+ auto __res = __detail::__spin_impl(__wait_addr, __args);
if (!__res._M_timeout)
return __res;
if (__args & __wait_flags::__spin_only)
return __res;
+ if (__wait_clock_t::now() >= __atime)
+ return __res;
}
#ifdef _GLIBCXX_HAVE_PLATFORM_WAIT
The __spin_until_impl function should stop spinning after some finite duration,
but instead it just spins (and sleeps) until the deadline is passed. That means
there's no use for the futex wait, because we always just spin until the
timeout.
Additionally, there was no check after __spin_until_loop to see whether the
deadline is passed, so we were proceeding to the futex wait anyway, when we
should have returned with a timeout.
With the patch above the benchmark gives:
BM_counting_sema 2770 ns 2767 ns 250728
