* [PATCH] libstdc++: Fix std::fill and std::fill_n optimizations [PR109150]
@ 2024-06-20 15:27 Jonathan Wakely
2024-06-21 16:08 ` Jonathan Wakely
0 siblings, 1 reply; 2+ messages in thread
From: Jonathan Wakely @ 2024-06-20 15:27 UTC (permalink / raw)
To: libstdc++, gcc-patches
I think the new conditions are correct. They're certainly an improvment
on just checking __is_scalar without considering what we're assigning it
to.
Tested x86_64-linux.
-- >8 --
As noted in the PR, the optimization used for scalar types in std::fill
and std::fill_n is non-conforming, because it doesn't consider that
assigning a scalar type might have non-trivial side effects which are
affected by the optimization.
By changing the condition under which the optimization is done we ensure
it's only performed when safe to do so, and we also enable it for
additional types, which was the original subject of the PR.
Instead of two overloads using __enable_if<__is_scalar<T>::__value, R>
we can combine them into one and create a local variable which is either
a local copy of __value or another reference to it, depending on whether
the optimization is allowed.
This removes a use of std::__is_scalar, which is a step towards fixing
PR 115497 by removing std::__is_pointer from <bits/cpp_type_traits.h>
libstdc++-v3/ChangeLog:
PR libstdc++/109150
* include/bits/stl_algobase.h (__fill_a1): Combine the
!__is_scalar and __is_scalar overloads into one and rewrite the
condition used to decide whether to perform the load outside the
loop.
* testsuite/25_algorithms/fill/109150.cc: New test.
* testsuite/25_algorithms/fill_n/109150.cc: New test.
---
libstdc++-v3/include/bits/stl_algobase.h | 75 ++++++++++++-------
.../testsuite/25_algorithms/fill/109150.cc | 62 +++++++++++++++
.../testsuite/25_algorithms/fill_n/109150.cc | 62 +++++++++++++++
3 files changed, 171 insertions(+), 28 deletions(-)
create mode 100644 libstdc++-v3/testsuite/25_algorithms/fill/109150.cc
create mode 100644 libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc
diff --git a/libstdc++-v3/include/bits/stl_algobase.h b/libstdc++-v3/include/bits/stl_algobase.h
index d831e0e9883..1a0f8c14073 100644
--- a/libstdc++-v3/include/bits/stl_algobase.h
+++ b/libstdc++-v3/include/bits/stl_algobase.h
@@ -929,28 +929,39 @@ _GLIBCXX_END_NAMESPACE_CONTAINER
#define _GLIBCXX_MOVE_BACKWARD3(_Tp, _Up, _Vp) std::copy_backward(_Tp, _Up, _Vp)
#endif
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wc++17-extensions"
template<typename _ForwardIterator, typename _Tp>
_GLIBCXX20_CONSTEXPR
- inline typename
- __gnu_cxx::__enable_if<!__is_scalar<_Tp>::__value, void>::__type
+ inline void
__fill_a1(_ForwardIterator __first, _ForwardIterator __last,
const _Tp& __value)
{
- for (; __first != __last; ++__first)
- *__first = __value;
- }
+ // We can optimize this loop by moving the load from __value outside
+ // the loop, but only if we know that making that copy is trivial,
+ // and the assignment in the loop is also trivial (so that the identity
+ // of the operand doesn't matter).
+ const bool __load_outside_loop =
+#if __has_builtin(__is_trivially_constructible) \
+ && __has_builtin(__is_trivially_assignable)
+ __is_trivially_constructible(_Tp, const _Tp&)
+ && __is_trivially_assignable(__decltype(*__first), const _Tp&)
+#else
+ __is_trivially_copyable(_Tp)
+ && __is_same(_Tp, __typeof__(*__first))
+#endif
+ && sizeof(_Tp) <= sizeof(long long);
- template<typename _ForwardIterator, typename _Tp>
- _GLIBCXX20_CONSTEXPR
- inline typename
- __gnu_cxx::__enable_if<__is_scalar<_Tp>::__value, void>::__type
- __fill_a1(_ForwardIterator __first, _ForwardIterator __last,
- const _Tp& __value)
- {
- const _Tp __tmp = __value;
+ // When the condition is true, we use a copy of __value,
+ // otherwise we just use another reference.
+ typedef typename __gnu_cxx::__conditional_type<__load_outside_loop,
+ const _Tp,
+ const _Tp&>::__type _Up;
+ _Up __val(__value);
for (; __first != __last; ++__first)
- *__first = __tmp;
+ *__first = __val;
}
+#pragma GCC diagnostic pop
// Specialization: for char types we can use memset.
template<typename _Tp>
@@ -1079,28 +1090,36 @@ _GLIBCXX_END_NAMESPACE_CONTAINER
__size_to_integer(__float128 __n) { return (long long)__n; }
#endif
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wc++17-extensions"
template<typename _OutputIterator, typename _Size, typename _Tp>
_GLIBCXX20_CONSTEXPR
- inline typename
- __gnu_cxx::__enable_if<!__is_scalar<_Tp>::__value, _OutputIterator>::__type
+ inline _OutputIterator
__fill_n_a1(_OutputIterator __first, _Size __n, const _Tp& __value)
{
- for (; __n > 0; --__n, (void) ++__first)
- *__first = __value;
- return __first;
- }
+ // See std::__fill_a1 for explanation of this condition.
+ const bool __load_outside_loop =
+#if __has_builtin(__is_trivially_constructible) \
+ && __has_builtin(__is_trivially_assignable)
+ __is_trivially_constructible(_Tp, const _Tp&)
+ && __is_trivially_assignable(__decltype(*__first), const _Tp&)
+#else
+ __is_trivially_copyable(_Tp)
+ && __is_same(_Tp, __typeof__(*__first))
+#endif
+ && sizeof(_Tp) <= sizeof(long long);
- template<typename _OutputIterator, typename _Size, typename _Tp>
- _GLIBCXX20_CONSTEXPR
- inline typename
- __gnu_cxx::__enable_if<__is_scalar<_Tp>::__value, _OutputIterator>::__type
- __fill_n_a1(_OutputIterator __first, _Size __n, const _Tp& __value)
- {
- const _Tp __tmp = __value;
+ // When the condition is true, we use a copy of __value,
+ // otherwise we just use another reference.
+ typedef typename __gnu_cxx::__conditional_type<__load_outside_loop,
+ const _Tp,
+ const _Tp&>::__type _Up;
+ _Up __val(__value);
for (; __n > 0; --__n, (void) ++__first)
- *__first = __tmp;
+ *__first = __val;
return __first;
}
+#pragma GCC diagnostic pop
template<typename _Ite, typename _Seq, typename _Cat, typename _Size,
typename _Tp>
diff --git a/libstdc++-v3/testsuite/25_algorithms/fill/109150.cc b/libstdc++-v3/testsuite/25_algorithms/fill/109150.cc
new file mode 100644
index 00000000000..9491a998ff0
--- /dev/null
+++ b/libstdc++-v3/testsuite/25_algorithms/fill/109150.cc
@@ -0,0 +1,62 @@
+// { dg-do run }
+
+// Test the problematic cases identified in PR libstdc++/109150
+// where the previous std::fill was non-conforming.
+
+#include <algorithm>
+#include <testsuite_hooks.h>
+
+const int global = 0;
+
+struct X {
+ void operator=(const int& i) { VERIFY(&i == &global); }
+};
+
+void
+test_identity_matters()
+{
+ X x;
+ // Assigning int to X has non-trivial side effects, so we cannot
+ // hoist the load outside the loop, we have to do exactly what the
+ // standard says to do.
+ std::fill(&x, &x+1, global);
+}
+
+struct Y {
+ int i;
+ void operator=(int ii) { i = ii + 1; }
+};
+
+void
+test_self_aliasing()
+{
+ Y y[2] = { };
+ // Assigning int to X has non-trivial side effects, altering the value
+ // used to fill the later elements. Must not load it outside the loop.
+ std::fill(y, y+2, y[0].i);
+ VERIFY(y[1].i == 2);
+}
+
+struct Z
+{
+ Z() { }
+#if __cplusplus >= 201103L
+ explicit Z(const Z&) = default;
+#endif
+};
+
+void
+test_explicit_copy_ctor()
+{
+ Z z;
+ // The optimization that copies the fill value must use direct-initialization
+ // otherwise this case would be ill-formed due to the explicit constructor.
+ std::fill(&z, &z, z);
+}
+
+int main()
+{
+ test_identity_matters();
+ test_self_aliasing();
+ test_explicit_copy_ctor();
+}
diff --git a/libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc b/libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc
new file mode 100644
index 00000000000..13bb31c9344
--- /dev/null
+++ b/libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc
@@ -0,0 +1,62 @@
+// { dg-do run }
+
+// Test the problematic cases identified in PR libstdc++/109150
+// where the previous std::fill_n was non-conforming.
+
+#include <algorithm>
+#include <testsuite_hooks.h>
+
+const int global = 0;
+
+struct X {
+ void operator=(const int& i) { VERIFY(&i == &global); }
+};
+
+void
+test_identity_matters()
+{
+ X x;
+ // Assigning int to X has non-trivial side effects, so we cannot
+ // hoist the load outside the loop, we have to do exactly what the
+ // standard says to do.
+ std::fill_n(&x, 1, global);
+}
+
+struct Y {
+ int i;
+ void operator=(int ii) { i = ii + 1; }
+};
+
+void
+test_self_aliasing()
+{
+ Y y[2] = { };
+ // Assigning int to X has non-trivial side effects, altering the value
+ // used to fill the later elements. Must not load it outside the loop.
+ std::fill_n(y, 2, y[0].i);
+ VERIFY(y[1].i == 2);
+}
+
+struct Z
+{
+ Z() { }
+#if __cplusplus >= 201103L
+ explicit Z(const Z&) = default;
+#endif
+};
+
+void
+test_explicit_copy_ctor()
+{
+ Z z;
+ // The optimization that copies the fill value must use direct-initialization
+ // otherwise this case would be ill-formed due to the explicit constructor.
+ std::fill_n(&z, 1, z);
+}
+
+int main()
+{
+ test_identity_matters();
+ test_self_aliasing();
+ test_explicit_copy_ctor();
+}
--
2.45.2
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] libstdc++: Fix std::fill and std::fill_n optimizations [PR109150]
2024-06-20 15:27 [PATCH] libstdc++: Fix std::fill and std::fill_n optimizations [PR109150] Jonathan Wakely
@ 2024-06-21 16:08 ` Jonathan Wakely
0 siblings, 0 replies; 2+ messages in thread
From: Jonathan Wakely @ 2024-06-21 16:08 UTC (permalink / raw)
To: libstdc++, gcc-patches
Pushed to trunk now.
On Thu, 20 Jun 2024 at 16:28, Jonathan Wakely <jwakely@redhat.com> wrote:
>
> I think the new conditions are correct. They're certainly an improvment
> on just checking __is_scalar without considering what we're assigning it
> to.
>
> Tested x86_64-linux.
>
> -- >8 --
>
> As noted in the PR, the optimization used for scalar types in std::fill
> and std::fill_n is non-conforming, because it doesn't consider that
> assigning a scalar type might have non-trivial side effects which are
> affected by the optimization.
>
> By changing the condition under which the optimization is done we ensure
> it's only performed when safe to do so, and we also enable it for
> additional types, which was the original subject of the PR.
>
> Instead of two overloads using __enable_if<__is_scalar<T>::__value, R>
> we can combine them into one and create a local variable which is either
> a local copy of __value or another reference to it, depending on whether
> the optimization is allowed.
>
> This removes a use of std::__is_scalar, which is a step towards fixing
> PR 115497 by removing std::__is_pointer from <bits/cpp_type_traits.h>
>
> libstdc++-v3/ChangeLog:
>
> PR libstdc++/109150
> * include/bits/stl_algobase.h (__fill_a1): Combine the
> !__is_scalar and __is_scalar overloads into one and rewrite the
> condition used to decide whether to perform the load outside the
> loop.
> * testsuite/25_algorithms/fill/109150.cc: New test.
> * testsuite/25_algorithms/fill_n/109150.cc: New test.
> ---
> libstdc++-v3/include/bits/stl_algobase.h | 75 ++++++++++++-------
> .../testsuite/25_algorithms/fill/109150.cc | 62 +++++++++++++++
> .../testsuite/25_algorithms/fill_n/109150.cc | 62 +++++++++++++++
> 3 files changed, 171 insertions(+), 28 deletions(-)
> create mode 100644 libstdc++-v3/testsuite/25_algorithms/fill/109150.cc
> create mode 100644 libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc
>
> diff --git a/libstdc++-v3/include/bits/stl_algobase.h b/libstdc++-v3/include/bits/stl_algobase.h
> index d831e0e9883..1a0f8c14073 100644
> --- a/libstdc++-v3/include/bits/stl_algobase.h
> +++ b/libstdc++-v3/include/bits/stl_algobase.h
> @@ -929,28 +929,39 @@ _GLIBCXX_END_NAMESPACE_CONTAINER
> #define _GLIBCXX_MOVE_BACKWARD3(_Tp, _Up, _Vp) std::copy_backward(_Tp, _Up, _Vp)
> #endif
>
> +#pragma GCC diagnostic push
> +#pragma GCC diagnostic ignored "-Wc++17-extensions"
> template<typename _ForwardIterator, typename _Tp>
> _GLIBCXX20_CONSTEXPR
> - inline typename
> - __gnu_cxx::__enable_if<!__is_scalar<_Tp>::__value, void>::__type
> + inline void
> __fill_a1(_ForwardIterator __first, _ForwardIterator __last,
> const _Tp& __value)
> {
> - for (; __first != __last; ++__first)
> - *__first = __value;
> - }
> + // We can optimize this loop by moving the load from __value outside
> + // the loop, but only if we know that making that copy is trivial,
> + // and the assignment in the loop is also trivial (so that the identity
> + // of the operand doesn't matter).
> + const bool __load_outside_loop =
> +#if __has_builtin(__is_trivially_constructible) \
> + && __has_builtin(__is_trivially_assignable)
> + __is_trivially_constructible(_Tp, const _Tp&)
> + && __is_trivially_assignable(__decltype(*__first), const _Tp&)
> +#else
> + __is_trivially_copyable(_Tp)
> + && __is_same(_Tp, __typeof__(*__first))
> +#endif
> + && sizeof(_Tp) <= sizeof(long long);
>
> - template<typename _ForwardIterator, typename _Tp>
> - _GLIBCXX20_CONSTEXPR
> - inline typename
> - __gnu_cxx::__enable_if<__is_scalar<_Tp>::__value, void>::__type
> - __fill_a1(_ForwardIterator __first, _ForwardIterator __last,
> - const _Tp& __value)
> - {
> - const _Tp __tmp = __value;
> + // When the condition is true, we use a copy of __value,
> + // otherwise we just use another reference.
> + typedef typename __gnu_cxx::__conditional_type<__load_outside_loop,
> + const _Tp,
> + const _Tp&>::__type _Up;
> + _Up __val(__value);
> for (; __first != __last; ++__first)
> - *__first = __tmp;
> + *__first = __val;
> }
> +#pragma GCC diagnostic pop
>
> // Specialization: for char types we can use memset.
> template<typename _Tp>
> @@ -1079,28 +1090,36 @@ _GLIBCXX_END_NAMESPACE_CONTAINER
> __size_to_integer(__float128 __n) { return (long long)__n; }
> #endif
>
> +#pragma GCC diagnostic push
> +#pragma GCC diagnostic ignored "-Wc++17-extensions"
> template<typename _OutputIterator, typename _Size, typename _Tp>
> _GLIBCXX20_CONSTEXPR
> - inline typename
> - __gnu_cxx::__enable_if<!__is_scalar<_Tp>::__value, _OutputIterator>::__type
> + inline _OutputIterator
> __fill_n_a1(_OutputIterator __first, _Size __n, const _Tp& __value)
> {
> - for (; __n > 0; --__n, (void) ++__first)
> - *__first = __value;
> - return __first;
> - }
> + // See std::__fill_a1 for explanation of this condition.
> + const bool __load_outside_loop =
> +#if __has_builtin(__is_trivially_constructible) \
> + && __has_builtin(__is_trivially_assignable)
> + __is_trivially_constructible(_Tp, const _Tp&)
> + && __is_trivially_assignable(__decltype(*__first), const _Tp&)
> +#else
> + __is_trivially_copyable(_Tp)
> + && __is_same(_Tp, __typeof__(*__first))
> +#endif
> + && sizeof(_Tp) <= sizeof(long long);
>
> - template<typename _OutputIterator, typename _Size, typename _Tp>
> - _GLIBCXX20_CONSTEXPR
> - inline typename
> - __gnu_cxx::__enable_if<__is_scalar<_Tp>::__value, _OutputIterator>::__type
> - __fill_n_a1(_OutputIterator __first, _Size __n, const _Tp& __value)
> - {
> - const _Tp __tmp = __value;
> + // When the condition is true, we use a copy of __value,
> + // otherwise we just use another reference.
> + typedef typename __gnu_cxx::__conditional_type<__load_outside_loop,
> + const _Tp,
> + const _Tp&>::__type _Up;
> + _Up __val(__value);
> for (; __n > 0; --__n, (void) ++__first)
> - *__first = __tmp;
> + *__first = __val;
> return __first;
> }
> +#pragma GCC diagnostic pop
>
> template<typename _Ite, typename _Seq, typename _Cat, typename _Size,
> typename _Tp>
> diff --git a/libstdc++-v3/testsuite/25_algorithms/fill/109150.cc b/libstdc++-v3/testsuite/25_algorithms/fill/109150.cc
> new file mode 100644
> index 00000000000..9491a998ff0
> --- /dev/null
> +++ b/libstdc++-v3/testsuite/25_algorithms/fill/109150.cc
> @@ -0,0 +1,62 @@
> +// { dg-do run }
> +
> +// Test the problematic cases identified in PR libstdc++/109150
> +// where the previous std::fill was non-conforming.
> +
> +#include <algorithm>
> +#include <testsuite_hooks.h>
> +
> +const int global = 0;
> +
> +struct X {
> + void operator=(const int& i) { VERIFY(&i == &global); }
> +};
> +
> +void
> +test_identity_matters()
> +{
> + X x;
> + // Assigning int to X has non-trivial side effects, so we cannot
> + // hoist the load outside the loop, we have to do exactly what the
> + // standard says to do.
> + std::fill(&x, &x+1, global);
> +}
> +
> +struct Y {
> + int i;
> + void operator=(int ii) { i = ii + 1; }
> +};
> +
> +void
> +test_self_aliasing()
> +{
> + Y y[2] = { };
> + // Assigning int to X has non-trivial side effects, altering the value
> + // used to fill the later elements. Must not load it outside the loop.
> + std::fill(y, y+2, y[0].i);
> + VERIFY(y[1].i == 2);
> +}
> +
> +struct Z
> +{
> + Z() { }
> +#if __cplusplus >= 201103L
> + explicit Z(const Z&) = default;
> +#endif
> +};
> +
> +void
> +test_explicit_copy_ctor()
> +{
> + Z z;
> + // The optimization that copies the fill value must use direct-initialization
> + // otherwise this case would be ill-formed due to the explicit constructor.
> + std::fill(&z, &z, z);
> +}
> +
> +int main()
> +{
> + test_identity_matters();
> + test_self_aliasing();
> + test_explicit_copy_ctor();
> +}
> diff --git a/libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc b/libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc
> new file mode 100644
> index 00000000000..13bb31c9344
> --- /dev/null
> +++ b/libstdc++-v3/testsuite/25_algorithms/fill_n/109150.cc
> @@ -0,0 +1,62 @@
> +// { dg-do run }
> +
> +// Test the problematic cases identified in PR libstdc++/109150
> +// where the previous std::fill_n was non-conforming.
> +
> +#include <algorithm>
> +#include <testsuite_hooks.h>
> +
> +const int global = 0;
> +
> +struct X {
> + void operator=(const int& i) { VERIFY(&i == &global); }
> +};
> +
> +void
> +test_identity_matters()
> +{
> + X x;
> + // Assigning int to X has non-trivial side effects, so we cannot
> + // hoist the load outside the loop, we have to do exactly what the
> + // standard says to do.
> + std::fill_n(&x, 1, global);
> +}
> +
> +struct Y {
> + int i;
> + void operator=(int ii) { i = ii + 1; }
> +};
> +
> +void
> +test_self_aliasing()
> +{
> + Y y[2] = { };
> + // Assigning int to X has non-trivial side effects, altering the value
> + // used to fill the later elements. Must not load it outside the loop.
> + std::fill_n(y, 2, y[0].i);
> + VERIFY(y[1].i == 2);
> +}
> +
> +struct Z
> +{
> + Z() { }
> +#if __cplusplus >= 201103L
> + explicit Z(const Z&) = default;
> +#endif
> +};
> +
> +void
> +test_explicit_copy_ctor()
> +{
> + Z z;
> + // The optimization that copies the fill value must use direct-initialization
> + // otherwise this case would be ill-formed due to the explicit constructor.
> + std::fill_n(&z, 1, z);
> +}
> +
> +int main()
> +{
> + test_identity_matters();
> + test_self_aliasing();
> + test_explicit_copy_ctor();
> +}
> --
> 2.45.2
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-06-21 16:08 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-06-20 15:27 [PATCH] libstdc++: Fix std::fill and std::fill_n optimizations [PR109150] Jonathan Wakely
2024-06-21 16:08 ` Jonathan Wakely
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).