KJ16609/gcc-reflection
1
0
Fork 0
mirror of https://forge.sourceware.org/marek/gcc.git synced 2026-02-22 12:00:11 -05:00
Code Issues Packages Projects Releases Wiki Activity
Files
ea42e28cbc3d4d579666582e8893e7629bf6a390
gcc-reflection/libstdc++-v3/include/bits/valarray_array.h
Jonathan Wakely 38a441689b libstdc++: Make std::valarray support overaligned types [PR108951] 
Switch std::valarray<T> memory allocation to use std::__new_allocator<T>
to allocate and deallocate memory. This adds support for extended
alignment types without needing to duplicate all the logic from
__new_allocator::allocate and __new_allocator::dellocate.

std::__new_allocator is used instead of std::allocator because we want
to ensure that the memory is still allocated with operator new, so we
don't want to use any possible program-defined specialization of
std::allocator<T> which the user might have provided.

To make using an allocator possible, __valarray_release_memory needs to
become a function template so that it knows the type T. It also needs an
additional parameter specifying the size of the allocation.

This change doesn't cause an ABI change for types with fundamental
alignment, because __new_allocator still uses the same operator delete
function (or the sized version, which is ABI compatible) to free the
memory. So if memory for a valarray is allocated in one translation unit
and deallocated in another, and those TUs are compiled with different
versions of GCC, we still get memory from the same operator new and
release it with the same operator delete (or the compatibled sized
version). For types with extended alignment this does potentially cause
an ABI change on targets where the aligned version of operator delete
doesn't just call free(p), but support for extended alignment types was
previously just broken and had undefined behaviour.

libstdc++-v3/ChangeLog:

	PR libstdc++/108951
	* include/bits/valarray_array.h( __valarray_get_storage): Use
	std::__new_allocator.
	(__valarray_release_memory): Likewise.
	* include/std/valarray: Pass _M_size to
	__valarray_release_memory.
	* testsuite/26_numerics/valarray/108951.cc: New test.

Reviewed-by: Tomasz Kamiński <tkaminsk@redhat.com>
2025-10-27 09:54:55 +00:00

626 lines
20 KiB
C++
Raw Blame History

// The template and inlines for the -*- C++ -*- internal _Array helper class.
// Copyright (C) 1997-2025 Free Software Foundation, Inc.
//
// This file is part of the GNU ISO C++ Library. This library is free
// software; you can redistribute it and/or modify it under the
// terms of the GNU General Public License as published by the
// Free Software Foundation; either version 3, or (at your option)
// any later version.
// This library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// Under Section 7 of GPL version 3, you are granted additional
// permissions described in the GCC Runtime Library Exception, version
// 3.1, as published by the Free Software Foundation.
// You should have received a copy of the GNU General Public License and
// a copy of the GCC Runtime Library Exception along with this program;
// see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
// <http://www.gnu.org/licenses/>.
/** @file bits/valarray_array.h
* This is an internal header file, included by other library headers.
* Do not attempt to use it directly. @headername{valarray}
*/
// Written by Gabriel Dos Reis <Gabriel.Dos-Reis@DPTMaths.ENS-Cachan.Fr>
#ifndef _VALARRAY_ARRAY_H
#define _VALARRAY_ARRAY_H 1
#ifdef _GLIBCXX_SYSHDR
#pragma GCC system_header
#endif
#include <bits/c++config.h>
#include <bits/cpp_type_traits.h>
#include <bits/new_allocator.h>
#include <cstdlib>
#include <new>
namespace std _GLIBCXX_VISIBILITY(default)
{
_GLIBCXX_BEGIN_NAMESPACE_VERSION
//
// Helper functions on raw pointers
//
// We get memory the old fashioned way
template<typename _Tp>
_Tp*
__valarray_get_storage(size_t) __attribute__((__malloc__));
template<typename _Tp>
inline _Tp*
__valarray_get_storage(size_t __n)
{ return std::__new_allocator<_Tp>().allocate(__n); }
// Return memory to the system
template<typename _Tp>
inline void
__valarray_release_memory(_Tp* __p, size_t __n)
{ std::__new_allocator<_Tp>().deallocate(__p, __n); }
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wc++17-extensions" // if constexpr
// Turn raw-memory into an array of _Tp filled with _Tp().
// This is used in `valarray<T> v(n);` and in `valarray<T>::shift(n)`.
template<typename _Tp>
inline void
__valarray_default_construct(_Tp* __b, _Tp* __e)
{
if _GLIBCXX_CONSTEXPR (__is_trivial(_Tp))
__builtin_memset(__b, 0, (__e - __b) * sizeof(_Tp));
else
while (__b != __e)
::new(static_cast<void*>(__b++)) _Tp();
}
// Turn a raw-memory into an array of _Tp filled with __t
// This is the required in valarray<T> v(n, t). Also
// used in valarray<>::resize().
template<typename _Tp>
inline void
__valarray_fill_construct(_Tp* __b, _Tp* __e, const _Tp __t)
{
while (__b != __e)
::new(static_cast<void*>(__b++)) _Tp(__t);
}
// copy-construct raw array [__o, *) from plain array [__b, __e)
template<typename _Tp>
inline void
__valarray_copy_construct(const _Tp* __b, const _Tp* __e,
_Tp* __restrict__ __o)
{
if _GLIBCXX_CONSTEXPR (__is_trivial(_Tp))
{
if (__b)
__builtin_memcpy(__o, __b, (__e - __b) * sizeof(_Tp));
}
else
while (__b != __e)
::new(static_cast<void*>(__o++)) _Tp(*__b++);
}
// copy-construct raw array [__o, *) from strided array __a[<__n : __s>]
template<typename _Tp>
inline void
__valarray_copy_construct (const _Tp* __restrict__ __a, size_t __n,
size_t __s, _Tp* __restrict__ __o)
{
if _GLIBCXX_CONSTEXPR (__is_trivial(_Tp))
while (__n--)
{
*__o++ = *__a;
__a += __s;
}
else
while (__n--)
{
new(__o++) _Tp(*__a);
__a += __s;
}
}
// copy-construct raw array [__o, *) from indexed array __a[__i[<__n>]]
template<typename _Tp>
inline void
__valarray_copy_construct (const _Tp* __restrict__ __a,
const size_t* __restrict__ __i,
_Tp* __restrict__ __o, size_t __n)
{
if _GLIBCXX_CONSTEXPR (__is_trivial(_Tp))
while (__n--)
*__o++ = __a[*__i++];
else
while (__n--)
new (__o++) _Tp(__a[*__i++]);
}
// Do the necessary cleanup when we're done with arrays.
template<typename _Tp>
inline void
__valarray_destroy_elements(_Tp* __b, _Tp* __e)
{
if _GLIBCXX_CONSTEXPR (!__is_trivial(_Tp))
while (__b != __e)
{
__b->~_Tp();
++__b;
}
}
#pragma GCC diagnostic pop
// Fill a plain array __a[<__n>] with __t
template<typename _Tp>
inline void
__valarray_fill(_Tp* __restrict__ __a, size_t __n, const _Tp& __t)
{
while (__n--)
*__a++ = __t;
}
// fill strided array __a[<__n-1 : __s>] with __t
template<typename _Tp>
inline void
__valarray_fill(_Tp* __restrict__ __a, size_t __n,
size_t __s, const _Tp& __t)
{
for (size_t __i = 0; __i < __n; ++__i, __a += __s)
*__a = __t;
}
// fill indirect array __a[__i[<__n>]] with __i
template<typename _Tp>
inline void
__valarray_fill(_Tp* __restrict__ __a, const size_t* __restrict__ __i,
size_t __n, const _Tp& __t)
{
for (size_t __j = 0; __j < __n; ++__j, ++__i)
__a[*__i] = __t;
}
// copy plain array __a[<__n>] in __b[<__n>]
// For non-fundamental types, it is wrong to say 'memcpy()'
template<typename _Tp, bool>
struct _Array_copier
{
inline static void
_S_do_it(const _Tp* __restrict__ __a, size_t __n, _Tp* __restrict__ __b)
{
while(__n--)
*__b++ = *__a++;
}
};
template<typename _Tp>
struct _Array_copier<_Tp, true>
{
inline static void
_S_do_it(const _Tp* __restrict__ __a, size_t __n, _Tp* __restrict__ __b)
{
if (__n != 0)
__builtin_memcpy(__b, __a, __n * sizeof (_Tp));
}
};
// Copy a plain array __a[<__n>] into a play array __b[<>]
template<typename _Tp>
inline void
__valarray_copy(const _Tp* __restrict__ __a, size_t __n,
_Tp* __restrict__ __b)
{
_Array_copier<_Tp, __is_trivial(_Tp)>::_S_do_it(__a, __n, __b);
}
// Copy strided array __a[<__n : __s>] in plain __b[<__n>]
template<typename _Tp>
inline void
__valarray_copy(const _Tp* __restrict__ __a, size_t __n, size_t __s,
_Tp* __restrict__ __b)
{
for (size_t __i = 0; __i < __n; ++__i, ++__b, __a += __s)
*__b = *__a;
}
// Copy a plain array __a[<__n>] into a strided array __b[<__n : __s>]
template<typename _Tp>
inline void
__valarray_copy(const _Tp* __restrict__ __a, _Tp* __restrict__ __b,
size_t __n, size_t __s)
{
for (size_t __i = 0; __i < __n; ++__i, ++__a, __b += __s)
*__b = *__a;
}
// Copy strided array __src[<__n : __s1>] into another
// strided array __dst[< : __s2>]. Their sizes must match.
template<typename _Tp>
inline void
__valarray_copy(const _Tp* __restrict__ __src, size_t __n, size_t __s1,
_Tp* __restrict__ __dst, size_t __s2)
{
for (size_t __i = 0; __i < __n; ++__i)
__dst[__i * __s2] = __src[__i * __s1];
}
// Copy an indexed array __a[__i[<__n>]] in plain array __b[<__n>]
template<typename _Tp>
inline void
__valarray_copy(const _Tp* __restrict__ __a,
const size_t* __restrict__ __i,
_Tp* __restrict__ __b, size_t __n)
{
for (size_t __j = 0; __j < __n; ++__j, ++__b, ++__i)
*__b = __a[*__i];
}
// Copy a plain array __a[<__n>] in an indexed array __b[__i[<__n>]]
template<typename _Tp>
inline void
__valarray_copy(const _Tp* __restrict__ __a, size_t __n,
_Tp* __restrict__ __b, const size_t* __restrict__ __i)
{
for (size_t __j = 0; __j < __n; ++__j, ++__a, ++__i)
__b[*__i] = *__a;
}
// Copy the __n first elements of an indexed array __src[<__i>] into
// another indexed array __dst[<__j>].
template<typename _Tp>
inline void
__valarray_copy(const _Tp* __restrict__ __src, size_t __n,
const size_t* __restrict__ __i,
_Tp* __restrict__ __dst, const size_t* __restrict__ __j)
{
for (size_t __k = 0; __k < __n; ++__k)
__dst[*__j++] = __src[*__i++];
}
//
// Compute the sum of elements in range [__f, __l) which must not be empty.
// This is a naive algorithm. It suffers from cancelling.
// In the future try to specialize for _Tp = float, double, long double
// using a more accurate algorithm.
//
template<typename _Tp>
inline _Tp
__valarray_sum(const _Tp* __f, const _Tp* __l)
{
_Tp __r = *__f++;
while (__f != __l)
__r += *__f++;
return __r;
}
// Compute the min/max of an array-expression
template<typename _Ta>
inline typename _Ta::value_type
__valarray_min(const _Ta& __a)
{
size_t __s = __a.size();
typedef typename _Ta::value_type _Value_type;
_Value_type __r = __s == 0 ? _Value_type() : __a[0];
for (size_t __i = 1; __i < __s; ++__i)
{
_Value_type __t = __a[__i];
if (__t < __r)
__r = __t;
}
return __r;
}
template<typename _Ta>
inline typename _Ta::value_type
__valarray_max(const _Ta& __a)
{
size_t __s = __a.size();
typedef typename _Ta::value_type _Value_type;
_Value_type __r = __s == 0 ? _Value_type() : __a[0];
for (size_t __i = 1; __i < __s; ++__i)
{
_Value_type __t = __a[__i];
if (__t > __r)
__r = __t;
}
return __r;
}
//
// Helper class _Array, first layer of valarray abstraction.
// All operations on valarray should be forwarded to this class
// whenever possible. -- gdr
//
template<typename _Tp>
struct _Array
{
explicit _Array(_Tp* const __restrict__);
explicit _Array(const valarray<_Tp>&);
_Array(const _Tp* __restrict__, size_t);
_Tp* begin() const;
_Tp* const __restrict__ _M_data;
};
// Copy-construct plain array __b[<__n>] from indexed array __a[__i[<__n>]]
template<typename _Tp>
inline void
__valarray_copy_construct(_Array<_Tp> __a, _Array<size_t> __i,
_Array<_Tp> __b, size_t __n)
{ std::__valarray_copy_construct(__a._M_data, __i._M_data,
__b._M_data, __n); }
// Copy-construct plain array __b[<__n>] from strided array __a[<__n : __s>]
template<typename _Tp>
inline void
__valarray_copy_construct(_Array<_Tp> __a, size_t __n, size_t __s,
_Array<_Tp> __b)
{ std::__valarray_copy_construct(__a._M_data, __n, __s, __b._M_data); }
template<typename _Tp>
inline void
__valarray_fill (_Array<_Tp> __a, size_t __n, const _Tp& __t)
{ std::__valarray_fill(__a._M_data, __n, __t); }
template<typename _Tp>
inline void
__valarray_fill(_Array<_Tp> __a, size_t __n, size_t __s, const _Tp& __t)
{ std::__valarray_fill(__a._M_data, __n, __s, __t); }
template<typename _Tp>
inline void
__valarray_fill(_Array<_Tp> __a, _Array<size_t> __i,
size_t __n, const _Tp& __t)
{ std::__valarray_fill(__a._M_data, __i._M_data, __n, __t); }
// Copy a plain array __a[<__n>] into a play array __b[<>]
template<typename _Tp>
inline void
__valarray_copy(_Array<_Tp> __a, size_t __n, _Array<_Tp> __b)
{ std::__valarray_copy(__a._M_data, __n, __b._M_data); }
// Copy strided array __a[<__n : __s>] in plain __b[<__n>]
template<typename _Tp>
inline void
__valarray_copy(_Array<_Tp> __a, size_t __n, size_t __s, _Array<_Tp> __b)
{ std::__valarray_copy(__a._M_data, __n, __s, __b._M_data); }
// Copy a plain array __a[<__n>] into a strided array __b[<__n : __s>]
template<typename _Tp>
inline void
__valarray_copy(_Array<_Tp> __a, _Array<_Tp> __b, size_t __n, size_t __s)
{ __valarray_copy(__a._M_data, __b._M_data, __n, __s); }
// Copy strided array __src[<__n : __s1>] into another
// strided array __dst[< : __s2>]. Their sizes must match.
template<typename _Tp>
inline void
__valarray_copy(_Array<_Tp> __a, size_t __n, size_t __s1,
_Array<_Tp> __b, size_t __s2)
{ std::__valarray_copy(__a._M_data, __n, __s1, __b._M_data, __s2); }
// Copy an indexed array __a[__i[<__n>]] in plain array __b[<__n>]
template<typename _Tp>
inline void
__valarray_copy(_Array<_Tp> __a, _Array<size_t> __i,
_Array<_Tp> __b, size_t __n)
{ std::__valarray_copy(__a._M_data, __i._M_data, __b._M_data, __n); }
// Copy a plain array __a[<__n>] in an indexed array __b[__i[<__n>]]
template<typename _Tp>
inline void
__valarray_copy(_Array<_Tp> __a, size_t __n, _Array<_Tp> __b,
_Array<size_t> __i)
{ std::__valarray_copy(__a._M_data, __n, __b._M_data, __i._M_data); }
// Copy the __n first elements of an indexed array __src[<__i>] into
// another indexed array __dst[<__j>].
template<typename _Tp>
inline void
__valarray_copy(_Array<_Tp> __src, size_t __n, _Array<size_t> __i,
_Array<_Tp> __dst, _Array<size_t> __j)
{
std::__valarray_copy(__src._M_data, __n, __i._M_data,
__dst._M_data, __j._M_data);
}
template<typename _Tp>
inline
_Array<_Tp>::_Array(_Tp* const __restrict__ __p)
: _M_data (__p) {}
template<typename _Tp>
inline
_Array<_Tp>::_Array(const valarray<_Tp>& __v)
: _M_data (__v._M_data) {}
template<typename _Tp>
inline
_Array<_Tp>::_Array(const _Tp* __restrict__ __b, size_t __s)
: _M_data(__valarray_get_storage<_Tp>(__s))
{ std::__valarray_copy_construct(__b, __s, _M_data); }
template<typename _Tp>
inline _Tp*
_Array<_Tp>::begin () const
{ return _M_data; }
#define _DEFINE_ARRAY_FUNCTION(_Op, _Name) \
template<typename _Tp> \
inline void \
_Array_augmented_##_Name(_Array<_Tp> __a, size_t __n, const _Tp& __t) \
{ \
for (_Tp* __p = __a._M_data; __p < __a._M_data + __n; ++__p) \
*__p _Op##= __t; \
} \
\
template<typename _Tp> \
inline void \
_Array_augmented_##_Name(_Array<_Tp> __a, size_t __n, _Array<_Tp> __b) \
{ \
_Tp* __p = __a._M_data; \
for (_Tp* __q = __b._M_data; __q < __b._M_data + __n; ++__p, ++__q) \
*__p _Op##= *__q; \
} \
\
template<typename _Tp, class _Dom> \
void \
_Array_augmented_##_Name(_Array<_Tp> __a, \
const _Expr<_Dom, _Tp>& __e, size_t __n) \
{ \
_Tp* __p(__a._M_data); \
for (size_t __i = 0; __i < __n; ++__i, ++__p) \
*__p _Op##= __e[__i]; \
} \
\
template<typename _Tp> \
inline void \
_Array_augmented_##_Name(_Array<_Tp> __a, size_t __n, size_t __s, \
_Array<_Tp> __b) \
{ \
_Tp* __q(__b._M_data); \
for (_Tp* __p = __a._M_data; __p < __a._M_data + __s * __n; \
__p += __s, ++__q) \
*__p _Op##= *__q; \
} \
\
template<typename _Tp> \
inline void \
_Array_augmented_##_Name(_Array<_Tp> __a, _Array<_Tp> __b, \
size_t __n, size_t __s) \
{ \
_Tp* __q(__b._M_data); \
for (_Tp* __p = __a._M_data; __p < __a._M_data + __n; \
++__p, __q += __s) \
*__p _Op##= *__q; \
} \
\
template<typename _Tp, class _Dom> \
void \
_Array_augmented_##_Name(_Array<_Tp> __a, size_t __s, \
const _Expr<_Dom, _Tp>& __e, size_t __n) \
{ \
_Tp* __p(__a._M_data); \
for (size_t __i = 0; __i < __n; ++__i, __p += __s) \
*__p _Op##= __e[__i]; \
} \
\
template<typename _Tp> \
inline void \
_Array_augmented_##_Name(_Array<_Tp> __a, _Array<size_t> __i, \
_Array<_Tp> __b, size_t __n) \
{ \
_Tp* __q(__b._M_data); \
for (size_t* __j = __i._M_data; __j < __i._M_data + __n; \
++__j, ++__q) \
__a._M_data[*__j] _Op##= *__q; \
} \
\
template<typename _Tp> \
inline void \
_Array_augmented_##_Name(_Array<_Tp> __a, size_t __n, \
_Array<_Tp> __b, _Array<size_t> __i) \
{ \
_Tp* __p(__a._M_data); \
for (size_t* __j = __i._M_data; __j<__i._M_data + __n; \
++__j, ++__p) \
*__p _Op##= __b._M_data[*__j]; \
} \
\
template<typename _Tp, class _Dom> \
void \
_Array_augmented_##_Name(_Array<_Tp> __a, _Array<size_t> __i, \
const _Expr<_Dom, _Tp>& __e, size_t __n) \
{ \
size_t* __j(__i._M_data); \
for (size_t __k = 0; __k<__n; ++__k, ++__j) \
__a._M_data[*__j] _Op##= __e[__k]; \
} \
\
template<typename _Tp> \
void \
_Array_augmented_##_Name(_Array<_Tp> __a, _Array<bool> __m, \
_Array<_Tp> __b, size_t __n) \
{ \
bool* __ok(__m._M_data); \
_Tp* __p(__a._M_data); \
for (_Tp* __q = __b._M_data; __q < __b._M_data + __n; \
++__q, ++__ok, ++__p) \
{ \
while (! *__ok) \
{ \
++__ok; \
++__p; \
} \
*__p _Op##= *__q; \
} \
} \
\
template<typename _Tp> \
void \
_Array_augmented_##_Name(_Array<_Tp> __a, size_t __n, \
_Array<_Tp> __b, _Array<bool> __m) \
{ \
bool* __ok(__m._M_data); \
_Tp* __q(__b._M_data); \
for (_Tp* __p = __a._M_data; __p < __a._M_data + __n; \
++__p, ++__ok, ++__q) \
{ \
while (! *__ok) \
{ \
++__ok; \
++__q; \
} \
*__p _Op##= *__q; \
} \
} \
\
template<typename _Tp, class _Dom> \
void \
_Array_augmented_##_Name(_Array<_Tp> __a, _Array<bool> __m, \
const _Expr<_Dom, _Tp>& __e, size_t __n) \
{ \
bool* __ok(__m._M_data); \
_Tp* __p(__a._M_data); \
for (size_t __i = 0; __i < __n; ++__i, ++__ok, ++__p) \
{ \
while (! *__ok) \
{ \
++__ok; \
++__p; \
} \
*__p _Op##= __e[__i]; \
} \
}
_DEFINE_ARRAY_FUNCTION(+, __plus)
_DEFINE_ARRAY_FUNCTION(-, __minus)
_DEFINE_ARRAY_FUNCTION(*, __multiplies)
_DEFINE_ARRAY_FUNCTION(/, __divides)
_DEFINE_ARRAY_FUNCTION(%, __modulus)
_DEFINE_ARRAY_FUNCTION(^, __bitwise_xor)
_DEFINE_ARRAY_FUNCTION(|, __bitwise_or)
_DEFINE_ARRAY_FUNCTION(&, __bitwise_and)
_DEFINE_ARRAY_FUNCTION(<<, __shift_left)
_DEFINE_ARRAY_FUNCTION(>>, __shift_right)
#undef _DEFINE_ARRAY_FUNCTION
_GLIBCXX_END_NAMESPACE_VERSION
} // namespace
# include <bits/valarray_array.tcc>
#endif /* _ARRAY_H */
Reference in New Issue View Git Blame Copy Permalink