[committed] libstdc++: Fix std::atomic<std::shared_ptr<T>> for AIX [PR104101]
Commit Message
Tested powerpc64le-linux, powerpc-aix, pushed to trunk.
This fixes a testsuite failure on AIX.
The lock function currently just spins, which should be changed to use
back-off, and maybe then _M_val.wait(__current) when supported.
libstdc++-v3/ChangeLog:
PR libstdc++/104101
* include/bits/shared_ptr_atomic.h (_Sp_atomic::_Atomic_count::lock):
Only use __thread_relax if __cpp_lib_atomic_wait is defined.
---
libstdc++-v3/include/bits/shared_ptr_atomic.h | 4 ++++
1 file changed, 4 insertions(+)
Comments
> From: Jonathan Wakely via Gcc-patches <gcc-patches@gcc.gnu.org>
> Date: Wed, 19 Jan 2022 02:03:58 +0100
> Tested powerpc64le-linux, powerpc-aix, pushed to trunk.
>
>
> This fixes a testsuite failure
I'm guessing the actual number is in the tens.
> on AIX.
...and by the looks of it, some 60 regressions (as counted
by contrib/regression/btest-gcc.sh) on newlib targets too
for the same reason.
brgds, H-P
@@ -392,7 +392,9 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
auto __current = _M_val.load(memory_order_relaxed);
while (__current & _S_lock_bit)
{
+#if __cpp_lib_atomic_wait
__detail::__thread_relax();
+#endif
__current = _M_val.load(memory_order_relaxed);
}
@@ -401,7 +403,9 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
__o,
memory_order_relaxed))
{
+#if __cpp_lib_atomic_wait
__detail::__thread_relax();
+#endif
__current = __current & ~_S_lock_bit;
}
return reinterpret_cast<pointer>(__current);