[mdspan.accessor]

A range of indices

[0, N) is an accessible range of a given data handle and an accessor if, for each i in the range, the accessor policy's access function produces a valid reference to an object .

A type A meets the accessor policy requirements if

(1.1)
A models copyable,
(1.2)
is_nothrow_move_constructible_v<A> is true,
(1.3)
is_nothrow_move_assignable_v<A> is true,
(1.4)
is_nothrow_swappable_v<A> is true, and
(1.5)
the following types and expressions are well-formed and have the specified semantics.

🔗

typename A::element_type

Result: A complete object type that is not an abstract class type.

🔗

typename A::data_handle_type

Result: A type that models copyable, and for which is_nothrow_move_constructible_v<A::data_handle_type> is true, is_nothrow_move_assignable_v<A::data_handle_type> is true, and is_nothrow_swappable_v<A::data_handle_type> is true.

[Note 1:

The type of data_handle_type need not be element_type*.

— end note]

🔗

typename A::reference

Result: A type that models common_reference_with<A::reference&&, A::element_type&>.

[Note 2:

The type of reference need not be element_type&.

— end note]

🔗

typename A::offset_policy

Result: A type OP such that:

(5.1)
OP meets the accessor policy requirements,
(5.2)
constructible_from<OP, const A&> is modeled, and
(5.3)
is_same_v<typename OP::element_type, typename A::element_type> is true.

🔗

a.access(p, i)

[Note 3:

Concrete accessor policies can impose preconditions for their access function.

However, they might not.

For example, an accessor where p is span<A::element_type, dynamic_extent> and access(p, i) returns p[i % p.size()] does not need to impose a precondition on i.

— end note]

🔗

a.offset(p, i)

Returns: q such that for b being A::offset_policy(a), and any integer n for which [0, n) is an accessible range of p and a:

(10.1)
$[0, n - i) is an accessible range of q and b; and$
(10.2)
b.access(q, j) provides access to the same element as a.access(p, i + j), for every j in the range $[0, n - i) .$

Remarks: The expression is equality-preserving.

23.7.3.5.3 Class template default_accessor [mdspan.accessor.default]

23.7.3.5.3.1 Overview [mdspan.accessor.default.overview]

namespace std { template<class ElementType> struct default_accessor { using offset_policy = default_accessor; using element_type = ElementType; using reference = ElementType&; using data_handle_type = ElementType*; constexpr default_accessor() noexcept = default; template<class OtherElementType> constexpr default_accessor(default_accessor<OtherElementType>) noexcept; constexpr reference access(data_handle_type p, size_t i) const noexcept; constexpr data_handle_type offset(data_handle_type p, size_t i) const noexcept; }; }

ElementType is required to be a complete object type that is neither an abstract class type nor an array type.

Each specialization of default_accessor is a trivially copyable type that models semiregular .

[0, n) is an accessible range for an object p of type data_handle_type and an object of type default_accessor if and only if [p, p + n) is a valid range .

23.7.3.5.3.2 Members [mdspan.accessor.default.members]

🔗

template<class OtherElementType>
  constexpr default_accessor(default_accessor<OtherElementType>) noexcept {}

Constraints: is_convertible_v<OtherElementType(*)[], element_type(*)[]> is true.

🔗

constexpr reference access(data_handle_type p, size_t i) const noexcept;

Effects: Equivalent to: return p[i];

🔗

constexpr data_handle_type offset(data_handle_type p, size_t i) const noexcept;

Effects: Equivalent to: return p + i;

23.7.3.5.4 Class template aligned_accessor [mdspan.accessor.aligned]

23.7.3.5.4.1 Overview [mdspan.accessor.aligned.overview]

namespace std { template<class ElementType, size_t ByteAlignment> struct aligned_accessor { using offset_policy = default_accessor<ElementType>; using element_type = ElementType; using reference = ElementType&; using data_handle_type = ElementType*; static constexpr size_t byte_alignment = ByteAlignment; constexpr aligned_accessor() noexcept = default; template<class OtherElementType, size_t OtherByteAlignment> constexpr aligned_accessor( aligned_accessor<OtherElementType, OtherByteAlignment>) noexcept; template<class OtherElementType> constexpr explicit aligned_accessor(default_accessor<OtherElementType>) noexcept; template<class OtherElementType> constexpr operator default_accessor<OtherElementType>() const noexcept; constexpr reference access(data_handle_type p, size_t i) const noexcept; constexpr typename offset_policy::data_handle_type offset( data_handle_type p, size_t i) const noexcept; }; }

ElementType is required to be a complete object type that is neither an abstract class type nor an array type.

Each specialization of aligned_accessor is a trivially copyable type that models semiregular .

[0, n) is an accessible range for an object p of type data_handle_type and an object of type aligned_accessor if and only if

(5.1)
[p, p + n) is a valid range, and,
(5.2)
if n is greater than zero, then is_sufficiently_aligned<byte_alignment>(p) is true.

[Example 1:

The following function compute uses is_sufficiently_aligned to check whether a given mdspan with default_accessor has a data handle with sufficient alignment to be used with aligned_accessor<float, 4 * sizeof(float)>.

If so, the function dispatches to a function compute_using_fourfold_overalignment that requires fourfold over-alignment of arrays, but can therefore use hardware-specific instructions, such as four-wide SIMD (Single Instruction Multiple Data) instructions.

Otherwise, compute dispatches to a possibly less optimized function compute_without_requiring_overalignment that has no over-alignment requirement.

void compute_using_fourfold_overalignment( mdspan<float, dims<1>, layout_right, aligned_accessor<float, 4 * alignof(float)> x); void compute_without_requiring_overalignment( mdspan<float, dims<1>, layout_right> x); void compute(mdspan<float, dims<1> x) { constexpr auto byte_alignment = 4 * sizeof(float); auto accessor = aligned_accessor<float, byte_alignment>{}; auto x_handle = x.data_handle(); if (is_sufficiently_aligned<byte_alignment>(x_handle)) { compute_using_fourfold_overalignment(mdspan{x_handle, x.mapping(), accessor}); } else { compute_without_requiring_overalignment(x); } } — end example]

23.7.3.5.4.2 Members [mdspan.accessor.aligned.members]

🔗

template<class OtherElementType, size_t OtherByteAlignment>
  constexpr aligned_accessor(aligned_accessor<OtherElementType, OtherByteAlignment>) noexcept;

Constraints:

(1.1)
is_convertible_v<OtherElementType(*)[], element_type(*)[]> is true.
(1.2)
OtherByteAlignment >= byte_alignment is true.

Effects: None.

🔗

template<class OtherElementType>
  constexpr explicit aligned_accessor(default_accessor<OtherElementType>) noexcept;

Constraints: is_convertible_v<OtherElementType(*)[], element_type(*)[]> is true.

Effects: None.

🔗

constexpr reference access(data_handle_type p, size_t i) const noexcept;

Effects: Equivalent to: return assume_aligned<byte_alignment>(p)[i];

🔗

template<class OtherElementType>
  constexpr operator default_accessor<OtherElementType>() const noexcept;

Constraints: is_convertible_v<element_type(*)[], OtherElementType(*)[]> is true.

Effects: Equivalent to: return {};

🔗

constexpr typename offset_policy::data_handle_type
  offset(data_handle_type p, size_t i) const noexcept;

Effects: Equivalent to: return assume_aligned<byte_alignment>(p) + i;