On Tue, Feb 10, 2026 at 10:55 PM Patrick Palka <[email protected]> wrote:
> Tested on x86_64-pc-linux-gnu with GLIBCXX_TESTSUITE_STDS=11,14,20,
> does this look OK for trunk and backports (after r16-7199 gets
> backported)? Also tested with Clang 21.
>
> -- >8 --
>
> After the front end change r16-7199 both GCC and Clang allow
> non-empty constexpr constructor bodies in C++11 as an extension,
> so we can unconditionally enable the __builtin_clear_padding logic
> in std::atomic's constructor.
>
> PR libstdc++/114865
>
> libstdc++-v3/ChangeLog:
>
> * include/std/atomic (atomic<_Tp>::atomic(_Tp))
> [C++11]: Enable __builtin_clear_padding logic.
> * testsuite/29_atomics/atomic/cons/static_zero_padding.cc
> [C++11]: Enable test verifying cleared padding bits for a
> std::atomic local variable.
> ---
> libstdc++-v3/include/std/atomic | 5 ++++-
> .../testsuite/29_atomics/atomic/cons/static_zero_padding.cc | 2 --
> 2 files changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/libstdc++-v3/include/std/atomic
> b/libstdc++-v3/include/std/atomic
> index 56dbe7bf5b95..306667e3c633 100644
> --- a/libstdc++-v3/include/std/atomic
> +++ b/libstdc++-v3/include/std/atomic
> @@ -247,14 +247,17 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
> atomic& operator=(const atomic&) = delete;
> atomic& operator=(const atomic&) volatile = delete;
>
> +#pragma GCC diagnostic push
> +#pragma GCC diagnostic ignored "-Wc++14-extensions" // constexpr ctor body
> constexpr atomic(_Tp __i) noexcept : _M_i(__i)
> {
> -#if __cplusplus >= 201402L && __has_builtin(__builtin_clear_padding)
> +#if __has_builtin(__builtin_clear_padding)
> if _GLIBCXX17_CONSTEXPR
> (__atomic_impl::__maybe_has_padding<_Tp>())
if (!std::__is_constant_evaluated())
> __builtin_clear_padding(std::__addressof(_M_i));
> #endif
> }
> +#pragma GCC diagnostic pop
>
> operator _Tp() const noexcept
> { return load(); }
> diff --git
> a/libstdc++-v3/testsuite/29_atomics/atomic/cons/static_zero_padding.cc
> b/libstdc++-v3/testsuite/29_atomics/atomic/cons/static_zero_padding.cc
> index 506513647c6c..e3594fb1c8ea 100644
> --- a/libstdc++-v3/testsuite/29_atomics/atomic/cons/static_zero_padding.cc
> +++ b/libstdc++-v3/testsuite/29_atomics/atomic/cons/static_zero_padding.cc
> @@ -57,9 +57,7 @@ void test_struct(std::atomic<T>& g, const T& zp)
>
> std::atomic<T> l(T{1, 2});
> std::memcpy(&t, &zp, sizeof(T));
> -#if __cplusplus >= 201402L // Remove once PR114865 is fixed
> VERIFY( l.compare_exchange_strong(t, d) );
> -#endif
>
Could you add here, and in functions below a test where atomic is allocated
on heap. Recently stack variable got erroneous value, and sure how that
impacts
padding (test was failing for me), but we sure it will not be zeroed on
heap.
I.e. having something like:
std::unique_ptr<std::atomic<T>> h(new atomic<T>(T{1, 2})); // make_unique
is C++14
std::memcpy(&t, &zp, sizeof(T));
VERIFY( h->compare_exchange_strong(t, d) );
> constexpr std::atomic<T> cl(T{1, 2});
> }
> --
> 2.53.0.68.g864f55e190
>
>
