//===----------------------------------------------------------------------===//
//
// Part of libcu++, the C++ Standard Library for your entire system,
// under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
// SPDX-FileCopyrightText: Copyright (c) 2023 NVIDIA CORPORATION & AFFILIATES.
//
//===----------------------------------------------------------------------===//

#ifndef _CUDA_STD_SPAN
#define _CUDA_STD_SPAN

#include <cuda/std/detail/__config>

#if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
#  pragma GCC system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_CLANG)
#  pragma clang system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_MSVC)
#  pragma system_header
#endif // no system header

#include <cuda/std/__concepts/concept_macros.h>
#include <cuda/std/__concepts/convertible_to.h>
#include <cuda/std/__concepts/equality_comparable.h>
#include <cuda/std/__exception/exception_macros.h>
#include <cuda/std/__fwd/array.h>
#include <cuda/std/__fwd/span.h>
#include <cuda/std/__fwd/string.h>
#include <cuda/std/__host_stdlib/stdexcept>
#include <cuda/std/__iterator/access.h>
#include <cuda/std/__iterator/concepts.h>
#include <cuda/std/__iterator/data.h>
#include <cuda/std/__iterator/distance.h>
#include <cuda/std/__iterator/empty.h>
#include <cuda/std/__iterator/reverse_access.h>
#include <cuda/std/__iterator/reverse_iterator.h>
#include <cuda/std/__iterator/size.h>
#include <cuda/std/__iterator/wrap_iter.h>
#include <cuda/std/__ranges/concepts.h>
#include <cuda/std/__ranges/data.h>
#include <cuda/std/__ranges/enable_borrowed_range.h>
#include <cuda/std/__ranges/enable_view.h>
#include <cuda/std/__ranges/size.h>
#include <cuda/std/__type_traits/integral_constant.h>
#include <cuda/std/__type_traits/integral_constant_like.h>
#include <cuda/std/__type_traits/is_array.h>
#include <cuda/std/__type_traits/is_const.h>
#include <cuda/std/__type_traits/is_convertible.h>
#include <cuda/std/__type_traits/is_integral.h>
#include <cuda/std/__type_traits/is_same.h>
#include <cuda/std/__type_traits/remove_const.h>
#include <cuda/std/__type_traits/remove_cv.h>
#include <cuda/std/__type_traits/remove_cvref.h>
#include <cuda/std/__type_traits/remove_reference.h>
#include <cuda/std/__type_traits/type_identity.h>
#include <cuda/std/array>
#include <cuda/std/cstddef>
#include <cuda/std/initializer_list>
#include <cuda/std/version>

#include <cuda/std/__cccl/prologue.h>

_CCCL_BEGIN_NAMESPACE_CUDA_STD

template <class _From, class _To>
_CCCL_CONCEPT __span_array_convertible = is_convertible_v<_From (*)[], _To (*)[]>;

template <class _Range, class _ElementType>
_CCCL_CONCEPT __span_compatible_range = _CCCL_REQUIRES_EXPR((_Range, _ElementType))(
  requires(::cuda::std::ranges::contiguous_range<_Range>),
  requires(::cuda::std::ranges::sized_range<_Range>),
  requires((::cuda::std::ranges::borrowed_range<_Range> || is_const_v<_ElementType>) ),
  requires((!is_array_v<remove_cvref_t<_Range>>) ),
  requires((!__is_cuda_std_span_v<remove_cvref_t<_Range>> && !__is_cuda_std_array_v<remove_cvref_t<_Range>>) ),
  requires(
    is_convertible_v<remove_reference_t<::cuda::std::ranges::range_reference_t<_Range>> (*)[], _ElementType (*)[]>));

template <class _It, class _Tp>
_CCCL_CONCEPT __span_compatible_iterator = _CCCL_REQUIRES_EXPR((_It, _Tp)) //
  (requires(contiguous_iterator<_It>),
   requires(__span_array_convertible<remove_reference_t<iter_reference_t<_It>>, _Tp>));

template <class _Sentinel, class _It>
_CCCL_CONCEPT __span_compatible_sentinel_for = _CCCL_REQUIRES_EXPR((_Sentinel, _It)) //
  (requires(sized_sentinel_for<_Sentinel, _It>), requires(!is_convertible_v<_Sentinel, size_t>));

template <class _Tp, bool = __integral_constant_like<_Tp>>
inline constexpr size_t __maybe_static_ext = dynamic_extent;

template <class _Tp>
inline constexpr size_t __maybe_static_ext<_Tp, true> = {_Tp::value};

template <typename _Tp, size_t _Extent>
class _CCCL_TYPE_VISIBILITY_DEFAULT span
{
public:
  //  constants and types
  using element_type     = _Tp;
  using value_type       = remove_cv_t<_Tp>;
  using size_type        = size_t;
  using difference_type  = ptrdiff_t;
  using pointer          = _Tp*;
  using const_pointer    = const _Tp*;
  using reference        = _Tp&;
  using const_reference  = const _Tp&;
  using iterator         = __wrap_iter<pointer>;
  using reverse_iterator = ::cuda::std::reverse_iterator<iterator>;

  static constexpr size_type extent = _Extent;

  // [span.cons], span constructors, copy, assignment, and destructor
  _CCCL_TEMPLATE(size_t _Sz = _Extent)
  _CCCL_REQUIRES((_Sz == 0))
  _CCCL_API constexpr span() noexcept
      : __data_{nullptr}
  {}

  _CCCL_HIDE_FROM_ABI span(const span&) noexcept            = default;
  _CCCL_HIDE_FROM_ABI span& operator=(const span&) noexcept = default;

  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _It)
  _CCCL_REQUIRES(__span_compatible_iterator<_It, element_type>)
  _CCCL_API constexpr explicit span(_It __first, [[maybe_unused]] size_type __count)
      : __data_{::cuda::std::to_address(__first)}
  {
    _CCCL_ASSERT(_Extent == __count, "size mismatch in span's constructor (iterator, len)");
  }

  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _It, class _End)
  _CCCL_REQUIRES(__span_compatible_iterator<_It, element_type> _CCCL_AND __span_compatible_sentinel_for<_End, _It>)
  _CCCL_API constexpr explicit span(_It __first, [[maybe_unused]] _End __last)
      : __data_{::cuda::std::to_address(__first)}
  {
    _CCCL_ASSERT((__last - __first >= 0), "invalid range in span's constructor (iterator, sentinel)");
    _CCCL_ASSERT(__last - __first == _Extent,
                 "invalid range in span's constructor (iterator, sentinel): last - first != extent");
  }

  _CCCL_TEMPLATE(size_t _Sz = _Extent)
  _CCCL_REQUIRES((_Sz != 0))
  _CCCL_API constexpr span(type_identity_t<element_type> (&__arr)[_Sz]) noexcept
      : __data_{__arr}
  {}

  _CCCL_TEMPLATE(class _OtherElementType)
  _CCCL_REQUIRES(__span_array_convertible<_OtherElementType, element_type>)
  _CCCL_API constexpr span(array<_OtherElementType, _Extent>& __arr) noexcept
      : __data_{__arr.data()}
  {}

  _CCCL_TEMPLATE(class _OtherElementType)
  _CCCL_REQUIRES(__span_array_convertible<const _OtherElementType, element_type>)
  _CCCL_API constexpr span(const array<_OtherElementType, _Extent>& __arr) noexcept
      : __data_{__arr.data()}
  {}

#if _CCCL_HAS_HOST_STD_LIB()
  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _OtherElementType)
  _CCCL_REQUIRES(__span_array_convertible<_OtherElementType, element_type>)
  _CCCL_HOST_API constexpr span(::std::array<_OtherElementType, _Extent>& __arr) noexcept
      : __data_{__arr.data()}
  {}

  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _OtherElementType)
  _CCCL_REQUIRES(__span_array_convertible<const _OtherElementType, element_type>)
  _CCCL_HOST_API constexpr span(const ::std::array<_OtherElementType, _Extent>& __arr) noexcept
      : __data_{__arr.data()}
  {}
#endif // _CCCL_HAS_HOST_STD_LIB()

  // NOLINTBEGIN(bugprone-forwarding-reference-overload)
  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _Range)
  _CCCL_REQUIRES(__span_compatible_range<_Range, element_type>)
  _CCCL_API constexpr explicit span(_Range&& __r)
      : __data_{::cuda::std::ranges::__data_cpo{}(__r)}
  {
    _CCCL_ASSERT(::cuda::std::ranges::__size_cpo{}(__r) == _Extent, "size mismatch in span's constructor (range)");
  }
  // NOLINTEND(bugprone-forwarding-reference-overload)

  // Deviation from P4144R1 Hana Dusikova's proposed fix: uses initializer_list<_InitListValueType>
  // instead of initializer_list<const _InitListValueType>. The const is redundant because
  // initializer_list<T>::begin() already returns const T*, so initializer_list<const T>::begin()
  // returns const const T* which collapses to const T* — identical behavior. Omitting the const
  // also avoids complications with template argument deduction in the CCCL SFINAE fallback path.
  _CCCL_TEMPLATE(class _InitListValueType, class _Ep = element_type)
  _CCCL_REQUIRES(is_same_v<_InitListValueType, value_type> _CCCL_AND is_const_v<_Ep>)
  _CCCL_API constexpr span(initializer_list<_InitListValueType> __il)
      : __data_{__il.begin()}
  {
    _CCCL_ASSERT(_Extent == __il.size(), "size mismatch in span's constructor (initializer_list)");
  }

  _CCCL_TEMPLATE(class _OtherElementType, size_t _Extent2 = _Extent)
  _CCCL_REQUIRES((_Extent2 != dynamic_extent) _CCCL_AND __span_array_convertible<_OtherElementType, element_type>)
  _CCCL_API constexpr span(const span<_OtherElementType, _Extent2>& __other) noexcept
      : __data_{__other.data()}
  {}

  _CCCL_TEMPLATE(class _OtherElementType)
  _CCCL_REQUIRES(__span_array_convertible<_OtherElementType, element_type>)
  _CCCL_API constexpr explicit span(const span<_OtherElementType, dynamic_extent>& __other) noexcept
      : __data_{__other.data()}
  {
    _CCCL_ASSERT(_Extent == __other.size(), "size mismatch in span's constructor (other span)");
  }

  //  ~span() noexcept = default;

  template <size_t _Count>
  _CCCL_API constexpr span<element_type, _Count> first() const noexcept
  {
    // ternary avoids "pointless comparison of unsigned integer with zero" warning
    static_assert(_Count == 0 ? true : _Count <= _Extent, "span<T, N>::first<Count>(): Count out of range");
    return span<element_type, _Count>{data(), _Count};
  }

  template <size_t _Count>
  _CCCL_API constexpr span<element_type, _Count> last() const noexcept
  {
    // ternary avoids "pointless comparison of unsigned integer with zero" warning
    static_assert(_Count == 0 ? true : _Count <= _Extent, "span<T, N>::last<Count>(): Count out of range");
    return span<element_type, _Count>{data() + size() - _Count, _Count};
  }

  _CCCL_API constexpr span<element_type, dynamic_extent> first(size_type __count) const noexcept
  {
    _CCCL_ASSERT(__count <= size(), "span<T, N>::first(count): count out of range");
    return {data(), __count};
  }

  _CCCL_API constexpr span<element_type, dynamic_extent> last(size_type __count) const noexcept
  {
    _CCCL_ASSERT(__count <= size(), "span<T, N>::last(count): count out of range");
    return {data() + size() - __count, __count};
  }

  template <size_t _Offset, size_t _Count>
  using __subspan_t = span<element_type, _Count != dynamic_extent ? _Count : _Extent - _Offset>;

  template <size_t _Offset, size_t _Count = dynamic_extent>
  _CCCL_API constexpr __subspan_t<_Offset, _Count> subspan() const noexcept
  {
    // ternary avoids "pointless comparison of unsigned integer with zero" warning
    if constexpr (_Offset != 0)
    {
      static_assert(_Offset <= _Extent, "span<T, N>::subspan<Offset, Count>(): Offset out of range");
    }
    if constexpr (_Count != 0)
    {
      static_assert(_Count == dynamic_extent || _Count <= _Extent - _Offset,
                    "span<T, N>::subspan<Offset, Count>(): Offset + Count out of range");
    }
    return __subspan_t<_Offset, _Count>{data() + _Offset, _Count == dynamic_extent ? size() - _Offset : _Count};
  }

  _CCCL_API constexpr span<element_type, dynamic_extent>
  subspan(size_type __offset, size_type __count = dynamic_extent) const noexcept
  {
    _CCCL_ASSERT(__offset <= size(), "span<T, N>::subspan(offset, count): offset out of range");
    _CCCL_ASSERT(__count <= size() || __count == dynamic_extent,
                 "span<T, N>::subspan(offset, count): count out of range");
    if (__count == dynamic_extent)
    {
      return {data() + __offset, size() - __offset};
    }
    _CCCL_ASSERT(__count <= size() - __offset, "span<T, N>::subspan(offset, count): offset + count out of range");
    return {data() + __offset, __count};
  }

  _CCCL_API constexpr size_type size() const noexcept
  {
    return _Extent;
  }
  _CCCL_API constexpr size_type size_bytes() const noexcept
  {
    return _Extent * sizeof(element_type);
  }
  [[nodiscard]] _CCCL_API constexpr bool empty() const noexcept
  {
    return _Extent == 0;
  }

  _CCCL_API constexpr reference operator[](size_type __idx) const noexcept
  {
    _CCCL_ASSERT(__idx < size(), "span<T, N>::operator[](index): index out of range");
    return __data_[__idx];
  }

  _CCCL_API constexpr reference at(size_type __idx) const
  {
    if (__idx >= size())
    {
      _CCCL_THROW(::std::out_of_range, "span::at");
    }
    return __data_[__idx];
  }

  _CCCL_API constexpr reference front() const noexcept
  {
    _CCCL_ASSERT(!empty(), "span<T, N>::front() on empty span");
    return __data_[0];
  }

  _CCCL_API constexpr reference back() const noexcept
  {
    _CCCL_ASSERT(!empty(), "span<T, N>::back() on empty span");
    return __data_[size() - 1];
  }

  _CCCL_API constexpr pointer data() const noexcept
  {
    return __data_;
  }

  // [span.iter], span iterator support
  _CCCL_API constexpr iterator begin() const noexcept
  {
    return iterator(data());
  }
  _CCCL_API constexpr iterator end() const noexcept
  {
    return iterator(data() + size());
  }
  _CCCL_API constexpr reverse_iterator rbegin() const noexcept
  {
    return reverse_iterator(end());
  }
  _CCCL_API constexpr reverse_iterator rend() const noexcept
  {
    return reverse_iterator(begin());
  }

  _CCCL_API inline span<const byte, _Extent * sizeof(element_type)> __as_bytes() const noexcept
  {
    return span<const byte, _Extent * sizeof(element_type)>{reinterpret_cast<const byte*>(data()), size_bytes()};
  }

  _CCCL_API inline span<byte, _Extent * sizeof(element_type)> __as_writable_bytes() const noexcept
  {
    return span<byte, _Extent * sizeof(element_type)>{reinterpret_cast<byte*>(data()), size_bytes()};
  }

private:
  pointer __data_;
};

template <typename _Tp>
class _CCCL_TYPE_VISIBILITY_DEFAULT span<_Tp, dynamic_extent>
{
public:
  //  constants and types
  using element_type     = _Tp;
  using value_type       = remove_cv_t<_Tp>;
  using size_type        = size_t;
  using difference_type  = ptrdiff_t;
  using pointer          = _Tp*;
  using const_pointer    = const _Tp*;
  using reference        = _Tp&;
  using const_reference  = const _Tp&;
  using iterator         = __wrap_iter<pointer>;
  using reverse_iterator = ::cuda::std::reverse_iterator<iterator>;

  static constexpr size_type extent = dynamic_extent;

  // [span.cons], span constructors, copy, assignment, and destructor
  _CCCL_API constexpr span() noexcept
      : __data_{nullptr}
      , __size_{0}
  {}

  _CCCL_HIDE_FROM_ABI span(const span&) noexcept            = default;
  _CCCL_HIDE_FROM_ABI span& operator=(const span&) noexcept = default;

  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _It)
  _CCCL_REQUIRES(__span_compatible_iterator<_It, element_type>)
  _CCCL_API constexpr span(_It __first, size_type __count)
      : __data_{::cuda::std::to_address(__first)}
      , __size_{__count}
  {}

  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _It, class _End)
  _CCCL_REQUIRES(__span_compatible_iterator<_It, element_type> _CCCL_AND __span_compatible_sentinel_for<_End, _It>)
  _CCCL_API constexpr span(_It __first, _End __last)
      : __data_(::cuda::std::to_address(__first))
      , __size_(__last - __first)
  {
    _CCCL_ASSERT(__last - __first >= 0, "invalid range in span's constructor (iterator, sentinel)");
  }

  template <size_t _Sz>
  _CCCL_API constexpr span(type_identity_t<element_type> (&__arr)[_Sz]) noexcept
      : __data_{__arr}
      , __size_{_Sz}
  {}

  _CCCL_TEMPLATE(class _OtherElementType, size_t _Sz)
  _CCCL_REQUIRES(__span_array_convertible<_OtherElementType, element_type>)
  _CCCL_API constexpr span(array<_OtherElementType, _Sz>& __arr) noexcept
      : __data_{__arr.data()}
      , __size_{_Sz}
  {}

  _CCCL_TEMPLATE(class _OtherElementType, size_t _Sz)
  _CCCL_REQUIRES(__span_array_convertible<const _OtherElementType, element_type>)
  _CCCL_API constexpr span(const array<_OtherElementType, _Sz>& __arr) noexcept
      : __data_{__arr.data()}
      , __size_{_Sz}
  {}

#if _CCCL_HAS_HOST_STD_LIB()
  _CCCL_TEMPLATE(class _OtherElementType, size_t _Sz)
  _CCCL_REQUIRES(__span_array_convertible<_OtherElementType, element_type>)
  _CCCL_HOST_API constexpr span(::std::array<_OtherElementType, _Sz>& __arr) noexcept
      : __data_{__arr.data()}
      , __size_{_Sz}
  {}

  _CCCL_TEMPLATE(class _OtherElementType, size_t _Sz)
  _CCCL_REQUIRES(__span_array_convertible<const _OtherElementType, element_type>)
  _CCCL_HOST_API constexpr span(const ::std::array<_OtherElementType, _Sz>& __arr) noexcept
      : __data_{__arr.data()}
      , __size_{_Sz}
  {}
#endif // _CCCL_HAS_HOST_STD_LIB()

  // NOLINTBEGIN(bugprone-forwarding-reference-overload)
  _CCCL_EXEC_CHECK_DISABLE
  _CCCL_TEMPLATE(class _Range)
  _CCCL_REQUIRES(__span_compatible_range<_Range, element_type>)
  _CCCL_API constexpr span(_Range&& __r)
      : __data_(::cuda::std::ranges::__data_cpo{}(__r))
      , __size_{::cuda::std::ranges::__size_cpo{}(__r)}
  {}
  // NOLINTEND(bugprone-forwarding-reference-overload)

  // Deviation from P4144R1 Hana Dusikova's proposed fix: uses initializer_list<_InitListValueType>
  // instead of initializer_list<const _InitListValueType>. The const is redundant because
  // initializer_list<T>::begin() already returns const T*, so initializer_list<const T>::begin()
  // returns const const T* which collapses to const T* — identical behavior. Omitting the const
  // also avoids complications with template argument deduction in the CCCL SFINAE fallback path.
  _CCCL_TEMPLATE(class _InitListValueType, class _Ep = element_type)
  _CCCL_REQUIRES(is_same_v<_InitListValueType, value_type> _CCCL_AND is_const_v<_Ep>)
  _CCCL_API constexpr span(initializer_list<_InitListValueType> __il)
      : __data_{__il.begin()}
      , __size_{__il.size()}
  {}

  _CCCL_TEMPLATE(class _OtherElementType, size_t _OtherExtent)
  _CCCL_REQUIRES(__span_array_convertible<_OtherElementType, element_type>)
  _CCCL_API constexpr span(const span<_OtherElementType, _OtherExtent>& __other) noexcept
      : __data_{__other.data()}
      , __size_{__other.size()}
  {}

  //    ~span() noexcept = default;

  template <size_t _Count>
  _CCCL_API constexpr span<element_type, _Count> first() const noexcept
  {
    // ternary avoids "pointless comparison of unsigned integer with zero" warning
    _CCCL_ASSERT(_Count == 0 ? true : _Count <= size(), "span<T>::first<Count>(): Count out of range");
    return span<element_type, _Count>{data(), _Count};
  }

  template <size_t _Count>
  _CCCL_API constexpr span<element_type, _Count> last() const noexcept
  {
    // ternary avoids "pointless comparison of unsigned integer with zero" warning
    _CCCL_ASSERT(_Count == 0 ? true : _Count <= size(), "span<T>::last<Count>(): Count out of range");
    return span<element_type, _Count>{data() + size() - _Count, _Count};
  }

  _CCCL_API constexpr span<element_type, dynamic_extent> first(size_type __count) const noexcept
  {
    _CCCL_ASSERT(__count <= size(), "span<T>::first(count): count out of range");
    return {data(), __count};
  }

  _CCCL_API constexpr span<element_type, dynamic_extent> last(size_type __count) const noexcept
  {
    _CCCL_ASSERT(__count <= size(), "span<T>::last(count): count out of range");
    return {data() + size() - __count, __count};
  }

  template <size_t _Offset, size_t _Count>
  using __subspan_t = span<element_type, _Count>;

  template <size_t _Offset, size_t _Count = dynamic_extent>
  _CCCL_API constexpr __subspan_t<_Offset, _Count> subspan() const noexcept
  {
    // ternary avoids "pointless comparison of unsigned integer with zero" warning
    _CCCL_ASSERT(_Offset == 0 ? true : _Offset <= size(), "span<T>::subspan<Offset, Count>(): Offset out of range");
    _CCCL_ASSERT(_Count == dynamic_extent || (_Count == 0 ? true : _Count <= size() - _Offset),
                 "span<T>::subspan<Offset, Count>(): Offset + Count out of range");
    return __subspan_t<_Offset, _Count>{data() + _Offset, _Count == dynamic_extent ? size() - _Offset : _Count};
  }

  constexpr span<element_type, dynamic_extent> _CCCL_API inline subspan(
    size_type __offset, size_type __count = dynamic_extent) const noexcept
  {
    _CCCL_ASSERT(__offset <= size(), "span<T>::subspan(offset, count): offset out of range");
    _CCCL_ASSERT(__count <= size() || __count == dynamic_extent, "span<T>::subspan(offset, count): count out of range");
    if (__count == dynamic_extent)
    {
      return {data() + __offset, size() - __offset};
    }
    _CCCL_ASSERT(__count <= size() - __offset, "span<T>::subspan(offset, count): offset + count out of range");
    return {data() + __offset, __count};
  }

  _CCCL_API constexpr size_type size() const noexcept
  {
    return __size_;
  }
  _CCCL_API constexpr size_type size_bytes() const noexcept
  {
    return __size_ * sizeof(element_type);
  }
  [[nodiscard]] _CCCL_API constexpr bool empty() const noexcept
  {
    return __size_ == 0;
  }

  _CCCL_API constexpr reference operator[](size_type __idx) const noexcept
  {
    _CCCL_ASSERT(__idx < size(), "span<T>::operator[](index): index out of range");
    return __data_[__idx];
  }

  _CCCL_API constexpr reference at(size_type __idx) const
  {
    if (__idx >= size())
    {
      _CCCL_THROW(::std::out_of_range, "span::at");
    }
    return __data_[__idx];
  }

  _CCCL_API constexpr reference front() const noexcept
  {
    _CCCL_ASSERT(!empty(), "span<T>::front() on empty span");
    return __data_[0];
  }

  _CCCL_API constexpr reference back() const noexcept
  {
    _CCCL_ASSERT(!empty(), "span<T>::back() on empty span");
    return __data_[size() - 1];
  }

  _CCCL_API constexpr pointer data() const noexcept
  {
    return __data_;
  }

  // [span.iter], span iterator support
  _CCCL_API constexpr iterator begin() const noexcept
  {
    return iterator(data());
  }
  _CCCL_API constexpr iterator end() const noexcept
  {
    return iterator(data() + size());
  }
  _CCCL_API constexpr reverse_iterator rbegin() const noexcept
  {
    return reverse_iterator(end());
  }
  _CCCL_API constexpr reverse_iterator rend() const noexcept
  {
    return reverse_iterator(begin());
  }

  _CCCL_API inline span<const byte, dynamic_extent> __as_bytes() const noexcept
  {
    return span<const byte, dynamic_extent>{reinterpret_cast<const byte*>(data()), size_bytes()};
  }

  _CCCL_API inline span<byte, dynamic_extent> __as_writable_bytes() const noexcept
  {
    return span<byte, dynamic_extent>{reinterpret_cast<byte*>(data()), size_bytes()};
  }

private:
  pointer __data_;
  size_type __size_;
};

//  as_bytes & as_writable_bytes
template <class _Tp, size_t _Extent>
_CCCL_API inline auto as_bytes(span<_Tp, _Extent> __s) noexcept
{
  return __s.__as_bytes();
}

_CCCL_TEMPLATE(class _Tp, size_t _Extent)
_CCCL_REQUIRES((!is_const<_Tp>::value))
_CCCL_API inline auto as_writable_bytes(span<_Tp, _Extent> __s) noexcept
{
  return __s.__as_writable_bytes();
}

//  Deduction guides
template <class _Tp, size_t _Sz>
_CCCL_DEDUCTION_GUIDE_ATTRIBUTES span(_Tp (&)[_Sz]) -> span<_Tp, _Sz>;

template <class _Tp, size_t _Sz>
_CCCL_DEDUCTION_GUIDE_ATTRIBUTES span(array<_Tp, _Sz>&) -> span<_Tp, _Sz>;

template <class _Tp, size_t _Sz>
_CCCL_DEDUCTION_GUIDE_ATTRIBUTES span(const array<_Tp, _Sz>&) -> span<const _Tp, _Sz>;

#if _CCCL_HAS_HOST_STD_LIB()
template <class _Tp, size_t _Sz>
_CCCL_DEDUCTION_GUIDE_ATTRIBUTES span(::std::array<_Tp, _Sz>&) -> span<_Tp, _Sz>;

template <class _Tp, size_t _Sz>
_CCCL_DEDUCTION_GUIDE_ATTRIBUTES span(const ::std::array<_Tp, _Sz>&) -> span<const _Tp, _Sz>;
#endif // _CCCL_HAS_HOST_STD_LIB()

_CCCL_TEMPLATE(class _It, class _EndOrSize)
_CCCL_REQUIRES(contiguous_iterator<_It>)
_CCCL_DEDUCTION_GUIDE_ATTRIBUTES span(_It, _EndOrSize)
  -> span<remove_reference_t<iter_reference_t<_It>>, __maybe_static_ext<_EndOrSize>>;

_CCCL_TEMPLATE(class _Range)
_CCCL_REQUIRES(::cuda::std::ranges::contiguous_range<_Range>)
_CCCL_DEDUCTION_GUIDE_ATTRIBUTES span(_Range&&)
  -> span<remove_reference_t<::cuda::std::ranges::range_reference_t<_Range>>>;

_CCCL_END_NAMESPACE_CUDA_STD

_CCCL_BEGIN_NAMESPACE_CUDA_STD_RANGES
template <class _Tp, size_t _Extent>
inline constexpr bool enable_borrowed_range<span<_Tp, _Extent>> = true;

template <class _Tp, size_t _Extent>
inline constexpr bool enable_view<span<_Tp, _Extent>> = true;
_CCCL_END_NAMESPACE_CUDA_STD_RANGES

#include <cuda/std/__cccl/epilogue.h>

#endif // _CUDA_STD_SPAN