Re: [committed] libstdc++: Optimize std::to_array for trivial types [PR110167]

[Jonathan Wakely <jwakely@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 6512 bytes --]

On Tue, 20 Jun 2023 at 01:54, Patrick Palka <ppalka@redhat.com> wrote:

> On Fri, 9 Jun 2023, Jonathan Wakely via Libstdc++ wrote:
>
> > Tested powerpc64le-linux. Pushed to trunk.
> >
> > This makes sense to backport after some soak time on trunk.
> >
> > -- >8 --
> >
> > As reported in PR libstdc++/110167, std::to_array compiles extremely
> > slowly for very large arrays. It needs to instantiate a very large
> > specialization of std::index_sequence<N...> and then create a very large
> > aggregate initializer from the pack expansion. For trivial types we can
> > simply default-initialize the std::array and then use memcpy to copy the
> > values. For non-trivial types we need to use the existing
> > implementation, despite the compilation cost.
> >
> > As also noted in the PR, using a generic lambda instead of the
> > __to_array helper compiles faster since gcc-13. It also produces
> > slightly smaller code at -O1, due to additional inlining. The code at
> > -Os, -O2 and -O3 seems to be the same. This new implementation requires
> > __cpp_generic_lambdas >= 201707L (i.e. P0428R2) but that is supported
> > since Clang 10 and since Intel icc 2021.5.0 (and since GCC 10.1).
> >
> > libstdc++-v3/ChangeLog:
> >
> >       PR libstdc++/110167
> >       * include/std/array (to_array): Initialize arrays of trivial
> >       types using memcpy. For non-trivial types, use lambda
> >       expressions instead of a separate helper function.
> >       (__to_array): Remove.
> >       * testsuite/23_containers/array/creation/110167.cc: New test.
> > ---
> >  libstdc++-v3/include/std/array                | 53 +++++++++++++------
> >  .../23_containers/array/creation/110167.cc    | 14 +++++
> >  2 files changed, 51 insertions(+), 16 deletions(-)
> >  create mode 100644
> libstdc++-v3/testsuite/23_containers/array/creation/110167.cc
> >
> > diff --git a/libstdc++-v3/include/std/array
> b/libstdc++-v3/include/std/array
> > index 70280c1beeb..b791d86ddb2 100644
> > --- a/libstdc++-v3/include/std/array
> > +++ b/libstdc++-v3/include/std/array
> > @@ -414,19 +414,8 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
> >        return std::move(std::get<_Int>(__arr));
> >      }
> >
> > -#if __cplusplus > 201703L
> > +#if __cplusplus >= 202002L && __cpp_generic_lambdas >= 201707L
> >  #define __cpp_lib_to_array 201907L
> > -
> > -  template<bool _Move = false, typename _Tp, size_t... _Idx>
> > -    constexpr array<remove_cv_t<_Tp>, sizeof...(_Idx)>
> > -    __to_array(_Tp (&__a)[sizeof...(_Idx)], index_sequence<_Idx...>)
> > -    {
> > -      if constexpr (_Move)
> > -     return {{std::move(__a[_Idx])...}};
> > -      else
> > -     return {{__a[_Idx]...}};
> > -    }
> > -
> >    template<typename _Tp, size_t _Nm>
> >      [[nodiscard]]
> >      constexpr array<remove_cv_t<_Tp>, _Nm>
> > @@ -436,8 +425,24 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
> >        static_assert(!is_array_v<_Tp>);
> >        static_assert(is_constructible_v<_Tp, _Tp&>);
> >        if constexpr (is_constructible_v<_Tp, _Tp&>)
> > -     return __to_array(__a, make_index_sequence<_Nm>{});
> > -      __builtin_unreachable(); // FIXME: see PR c++/91388
> > +     {
> > +       if constexpr (is_trivial_v<_Tp> && _Nm != 0)
>
> redundant _Nm != 0 test?
>

Ah yes, I added it below to ensure we don't use memcpy with a null
__arr.data() and forgot to remove it here.


>
> > +         {
> > +           array<remove_cv_t<_Tp>, _Nm> __arr;
> > +           if (!__is_constant_evaluated() && _Nm != 0)
> > +             __builtin_memcpy(__arr.data(), __a, sizeof(__a));
> > +           else
> > +             for (size_t __i = 0; __i < _Nm; ++__i)
> > +               __arr._M_elems[__i] = __a[__i];
> > +           return __arr;
> > +         }
> > +       else
> > +         return [&__a]<size_t... _Idx>(index_sequence<_Idx...>) {
> > +           return array<remove_cv_t<_Tp>, _Nm>{{ __a[_Idx]... }};
> > +         }(make_index_sequence<_Nm>{});
> > +     }
> > +      else
> > +     __builtin_unreachable(); // FIXME: see PR c++/91388
> >      }
> >
> >    template<typename _Tp, size_t _Nm>
> > @@ -449,8 +454,24 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
> >        static_assert(!is_array_v<_Tp>);
> >        static_assert(is_move_constructible_v<_Tp>);
> >        if constexpr (is_move_constructible_v<_Tp>)
> > -     return __to_array<1>(__a, make_index_sequence<_Nm>{});
> > -      __builtin_unreachable(); // FIXME: see PR c++/91388
> > +     {
> > +       if constexpr (is_trivial_v<_Tp>)
> > +         {
> > +           array<remove_cv_t<_Tp>, _Nm> __arr;
> > +           if (!__is_constant_evaluated() && _Nm != 0)
> > +             __builtin_memcpy(__arr.data(), __a, sizeof(__a));
> > +           else
> > +             for (size_t __i = 0; __i < _Nm; ++__i)
> > +               __arr._M_elems[__i] = std::move(__a[__i]);
>
> IIUC this std::move is unnecessary for trivial arrays?
>

Good point, thanks.

That makes the lvalue and rvalue overloads identical for trivial types. It
seems a shame to duplicate the code, so the rvalue one could do:

  if constexpr (is_trivial_v<_Tp>)
   return std::to_array<_Tp, _Num>(__a);
 else

But that would imply an extra function call at -O0, and repeating overload
resolution. Since the duplicated code is just a single function call to
memcpy, this probably isn't an improvement.

I'll test and push this:

    libstdc++: Remove redundant code in std::to_array

    libstdc++-v3/ChangeLog:

            * include/std/array (to_array(T(&)[N])): Remove redundant
            condition.
            (to_array(T(&&)[N])): Remove redundant std::move.

diff --git a/libstdc++-v3/include/std/array b/libstdc++-v3/include/std/array
index b791d86ddb2..ad36cdad6d2 100644
--- a/libstdc++-v3/include/std/array
+++ b/libstdc++-v3/include/std/array
@@ -426,7 +426,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       static_assert(is_constructible_v<_Tp, _Tp&>);
       if constexpr (is_constructible_v<_Tp, _Tp&>)
        {
-         if constexpr (is_trivial_v<_Tp> && _Nm != 0)
+         if constexpr (is_trivial_v<_Tp>)
            {
              array<remove_cv_t<_Tp>, _Nm> __arr;
              if (!__is_constant_evaluated() && _Nm != 0)
@@ -462,7 +462,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
                __builtin_memcpy(__arr.data(), __a, sizeof(__a));
              else
                for (size_t __i = 0; __i < _Nm; ++__i)
-                 __arr._M_elems[__i] = std::move(__a[__i]);
+                 __arr._M_elems[__i] = __a[__i];
              return __arr;
            }
          else