Project: | ISO JTC1/SC22/WG21: Programming Language C++ |
---|---|
Number: | P0009r4 |
Date: | 2017-10-11 |
Reply-to: | hcedwar@sandia.gov, balelbach@lbl.gov |
Author: | H. Carter Edwards |
Contact: | hcedwar@sandia.gov |
Author: | Bryce Lelbach |
Contact: | balelbach@lbl.gov |
Author: | Christian Trott |
Contact: | crtrott@sandia.gov |
Author: | Mauro Bianco |
Contact: | mbianco@cscs.ch |
Author: | Ben Sander |
Contact: | ben.sander@amd.com |
Author: | Athanasios Iliopoulos |
Contact: | athanasios.iliopoulos@nrl.navy.mil |
Author: | John Michopoulos |
Contact: | john.michopoulos@nrl.navy.mil |
Audience: | Library Evolution Working Group (LEWG) |
URL: | https://github.com/kokkos/array_ref/blob/master/proposals/P0009.rst |
Original multidimensional array reference paper with motivation, specification, and examples.
LEWG feedback view is not an acceptable name, bikeshed names: view , span , array_ref , slice , array_view , ref , array_span , basic_span , object_span , field
Revised with renaming from view to array_ref and allow unbounded rank through variadic arguments.
Adding details for extensibility of layout mapping.
Move motivation, examples, and relaxed incomplete array type proposal to separate papers.
- P0331 : Motivation and Examples for Polymorphic Multidimensional Array
- P0332 : Relaxed Incomplete Multidimensional Array Type Declaration
Oulu-2016 LEWG strawpoll: Move iterator from this paper to a subsequent paper.
Oulu-2016 LEWG feedback: http://wiki.edg.com/bin/view/Wg21oulu/P0009
- Array extents specification mechanism options are either-or, not both.
- List potential names for LEWG and/or LWG todo bikeshedding.
- Clearly & concisely note difference between multidimensional array versus language's array-of-array-of-array...
- Actual specification of reference type (and others), not "typically is" vagueness.
- Future directions / extensibility section regarding Properties...
The domain space specification preferred and undesirable mechanisms changed from accepting both to accepting only one.
Tighten up domain, codomain, and domain -> codomain mapping specifications.
Consistently use extent and extents for the multidimensional index space.
More LEWG name bikeshedding: sci_span , numeric_span , multidimensional_span , multidim_span , md_span , vla_span , multispan , multi_span
Changes from P0009r3:
- Rename to mdspan, multidimensional span, to align with P0122r5 span.
- Move preferred array extents mechanism to appendix
- Align with P0122r5 span
- Expose codomain as a std::span
- Elaborate layout mapping
Requested full-LEWG straw polls at Albuquerque Nov'2017 meeting:
- Acceptability of exclusively using signed integer index_type and omitting the traditional container unsigned integer size_type.
- Embedding the domain-to-codomain mapping observers within the mdspan class or moving these into a mdspan::mapping class.
- Given authors' update in response to previous two straw polls, is this ready to advance to LWG?
- If not, what specific modifications are required for consensus?
The proposed polymorphic multidimensional array reference (mdspan) defines types and functions for mapping indices from a multidimensional index space (the domain) to members of a contiguous span of objects (the codomain). A multidimensional index space is defined as the Cartesian product of integer extents, [0..N0) X [0..N1) X [0..N2) ...
This layout mapping is one property of the mdspan that may be specified through a template parameter. The intent is that properties* are an extensible set of options for multi-index mapping and member access. For example, bounds checking the input multi-index versus the multdimensional extents or accessing members through an atomic interface. The recent Accessors paper (P0367) introduces a rich set of potential access properties.
A multidimensional array is not an array-of-array-of-array-of-array...
The multidimensional array abstraction has been fundamental to numerical computations for over five decades. However, the C/C++ language provides only a one dimensional array abstraction which can be composed into array-of-array-of-array... types. While such types have some similarity to multidimensional arrays they do not provide adequate multidimensional array functionality of this proposal. Two critical functionality differences are (1) multiple dynamic extents and (2) polymorphic mapping of multi-indices to member objects.
namespace std { namespace experimental { inline constexpr ptrdiff_t dynamic_extent = -1 ; // Revise to add inline template< typename DataType , typename ... Properties > class mdspan ; template< ptrdiff_t ... StaticExtents > class extents ; class layout_right ; class layout_left ; class layout_stride ; template< ptrdiff_t ... LHS , ptrdiff_t ... RHS > constexpr bool operator == ( extents<LHS...> const & lhs , extents<RHS...> const & rhs ) ; template< ptrdiff_t ... LHS , ptrdiff_t ... RHS > constexpr bool operator != ( extents<LHS...> const & lhs , extents<RHS...> const & rhs ) ; // return type of subspan free function is an mdspan template< typename DataType , typename ... Properties , typename ... SliceSpecifiers > // for exposition only: detail::subspan_deduction_t< mdspan<DataType,Properties...>,SliceSpecifiers...> subspan( mdspan< DataType, Properties ... > const & , SliceSpecifiers ... ) noexcept; // tag supporting subspan struct all_type {}; inline constexpr all_type all = all_type{}; }}
The mdspan class maps a multi-index within a multi-index domain to a reference to the codomain, defined by a span of objects.
The subspan free function generates an mdspan with a domain contained within the input mdspan domain and codomain contained within the input mdspan codomain.
The detail::subspan_deduction_t template class is not proposed and appears for exposition only. An implementation metafunction of this form is necessary to deduce the specific mdspan return type of the subspan function.
namespace std { namespace experimental { template <typename DataType, typename... Properties> class mdspan { public: // domain and codomain types using element_type = typename remove_all_extents_t<DataType> ; using value_type = typename remove_cv_t< element_type > ; using index_type = ptrdiff_t ; using difference_type = ptrdiff_t ; using pointer = element_type * ; using reference = element_type & ; // Standard constructors, assignments, and destructor ~mdspan() noexcept ; constexpr mdspan() noexcept ; constexpr mdspan(mdspan&&) noexcept = default ; constexpr mdspan(mdspan const&) noexcept = default ; mdspan& operator=(mdspan&&) noexcept = default ; mdspan& operator=(mdspan const&) noexcept = default ; // Constructor and assignment for assignable mdspan template <typename UType, typename ... UProp> constexpr mdspan( mdspan<UType, UProp...> const& ) noexcept; template <typename UType, typename ... UProp> mdspan& operator=( mdspan<UType, UProp...> const& ) noexcept; // Wrapping constructors constexpr mdspan(nullptr_t) noexcept; template< class ... IndexType > explicit constexpr mdspan(pointer, IndexType ... DynamicExtents ) noexcept; template< class ... IndexType > explicit constexpr mdspan(std::span<element_type>, IndexType ... DynamicExtents ) noexcept; template< class IndexType , size_t N > explicit constexpr msspan(pointer, std::array<IndexType,N> const & DynamicExtents ) noexcept ; template< class IndexType , size_t N > explicit constexpr msspan(std::span<element_type>, std::array<IndexType,N> const & DynamicExtents ) noexcept ; // mapping domain multi-index to access codomain member reference operator[]( index_type ) const noexcept; // requires rank() == 1 template< class ... IndexType > reference operator()( IndexType ... indices ) const noexcept; template< class IndexType , size_t N > reference operator()( std::array<IndexType,N> const & indices ) const noexcept; // observers of the index space domain static constexpr int rank() noexecept ; static constexpr int rank_dynamic() noexecept ; static constexpr index_type static_extent(int) noexecept ; constexpr index_type extent(int) const noexecept ; constexpr index_type size() const noexecept ; // observers of the codomain: constexpr std::span<element_type> span() const noexecept ; template< class ... IndexType > static constexpr index_type required_span_size( IndexType ... DynamicExtents ); template< class ... IndexType , size_t N > static constexpr index_type required_span_size( std::array<IndexType,N> const & DynamicExtents ); // observers of the mapping : domain -> codomain using layout = /* extracted from Properties... */ ; static constexpr bool is_always_unique = /* layout */ ; static constexpr bool is_always_contiguous = /* layout */ ; static constexpr bool is_always_strided = /* layout */ ; constexpr bool is_unique() const ; constexpr bool is_contiguous() const ; constexpr bool is_strided() const ; constexpr index_type stride(int) const ; private: // exposition only typename layout::mapping< StaticExtents... > mapping ; pointer_type ptr ; }; }}
template <typename DataType, typename... Properties> class mdspan
DataType
Requires: Is a non-array type denoting the element type of the array.
Properties...
Effects: The domain index space rank, static extents, and identification of dynamic extents is determined from the extents member of the property pack. The domain to codomain mapping is determined by the remaining members of the property pack
[Note: If std::is_const<element_type> then references to codomain members are const. Extensions to access properties may cause reference to become a proxy type. --end note]
[Note: Integral types for dimensions and indexing are signed integers to avoid casting unsigned-to-signed for loop bounds and improve opportunities for optimizing loops. --end note]
The multi-index domain space is the Cartesian product of the extents: [0..extent(0)) X [0..extent(1)) X ... X [0..extent(rank()-1)). Each extent may be statically (at compile time) or dynamically (at runtime) specified.
static constexpr int rank();
Returns: Rank of the multi-index domain.
static constexpr int rank_dynamic();
Returns: number of extents that are dynamic.
static constexpr index_type static_extent(int r);
Requires: 0 <= r
Returns: If 0 <= r < rank() the statically specified extent. A statically declared extent of dynamic_extent denotes that the extent is dynamic. If rank() <= r then static_extent(r) == 1.
constexpr index_type extent(int r) const ;
Requires: 0 <= r
Returns: If 0 <= r < rank() the extent of coordinate r. If rank() <= r then extent(r) == 1.
constexpr index_type size() const ;
Returns: product of extent(r) where 0 <= r < rank().
Given mdspan x then:
int d = 0 ; index_type s = 1 ; for ( int i = 0 ; i < x.rank() ; ++i ) { if ( x.static_extent(i) == std::dynamic_extent ) { ++d ; } s *= x.extent(i); } assert( d == x.rank_dynamic() ); assert( s == x.size() );
Not all members of the codomain may be accessible through the layout mapping; i.e., the range of the mapping is contained within the codomain but may not be equal to the codomain.
constexpr std::span<element_type> span() const ;
Returns: An std::span for the codomain.
Requires:
- rank_dynamic() <= sizeof...(DynamicExtents)
- is_integral_type_v<IndexType>...
- Denote the ith coordinate of DynamicExtents... as denoted as DynamicExtents[ith] then:
- 0 <= DynamicExtents[ith] for 0 <= ith < rank_dynanic()
- 1 == DynamicExtents[ith] for rank_dynamic() < ith
Returns: The minimum size of the codomain to support the multi-index domain defined by the merging of DynamicExents with StaticExtents.
Requires:
- rank_dynamic() <= N
- is_integral_type_v<IndexType>...
- 0 <= DynamicExtents[ith] for 0 <= ith < rank_dynanic()
- 1 == DynamicExtents[ith] for rank_dynamic() < ith
Returns: The minimum size of the codomain to support the multi-index domain defined by the merging of DynamicExents with StaticExtents.
constexpr mdspan();
Effect: Construct a null mdspan with codomain span() == std::span<element_type>() and extent(r) == 0 for all dynamic extents.
Requires: Given using V = mdspan<DataType,Properties...> and using U = mdspan<UType,UProperties...> then
is_assignable<V::pointer,U::pointer> ,V::rank() == U::rank() ,V::static_extent(r) == U::static_extent(r) or V::static_extent(r) == std::dynamic_extent for 0 <= r < V::rank() ,compatibility of layout mappingEffect: * this has equal domain, equal codomain, and equivalent mapping.
Requires:
- sizeof...(DynamicExtents) == rank_dynamic()
- is_integral_type_v<IndexType>...
- The ith coordinate of DynamicExtents..., denoted as DynamicExtents[ith], is 0 <= DynamicExtents[ith].
- The span of elements denoted by [ ptr , ptr + required_span_size(DynamicExtents...) ), shall be a valid contiguous span of elements.
Effects: This wrapping constructor constructs * this with domain's dynamic extents equal to DynamicExtents... and codomain equal to std::span<element_type>( ptr , required_span_size(DynamicExtents...) )
Requires:
- N == rank_dynamic()
- is_integral_type_v<IndexType>...
- 0 <= DynamicExtents[ith]
- The span of elements denoted by [ ptr , ptr + required_span_size(DynamicExtents) ), shall be a valid contiguous span of elements.
Effects: This wrapping constructor constructs * this with domain's dynamic extents equal to DynamicExtents[ith]. and codomain equal to std::span<element_type>( ptr , required_span_size(DynamicExtents) )
Requires:
- sizeof...(DynamicExtents) == rank_dynamic()
- is_integral_type_v<IndexType>...
- The ith coordinate of DynamicExtents..., denoted as DynamicExtents[ith], is 0 <= DynamicExtents[ith]
- required_span_size(DynamicExtents...) <= s.size()
Effects: This wrapping constructor constructs * this with domain's dynamic extents equal to DynamicExtents... and codomain equal to std::span<element_type>( ptr , required_span_size(DynamicExtents...) )
Requires:
- N == rank_dynamic()
- is_integral_type_v<IndexType>...
- 0 <= DynamicExtents[ith]
- required_span_size(DynamicExtents) <= s.size()
Effects: This wrapping constructor constructs * this with domain's dynamic extents equal to DynamicExtents[ith] and codomain equal to std::span<element_type>( ptr , required_span_size(DynamicExtents[ith]) )
reference operator[]( index_type index ) const noexcept
Requires: rank() == 1 and 0 <= i < extent(0)
Returns: A reference to the element mapped to by index.
Requires: indices is a multi-index in the domain:
- rank() <= sizeof...(IndexType)
- The ith coordinate of indices..., denoted as indices[ith], is in the domain: 0 <= indices[ith] < extent(ith).
- [Note: Because extent(ith) == 1 for rank() <= ith then extra zero-value indices are valid. --end note]
Returns: A reference to the element mapped to by indices....
Remark: Optimization of the mapping operator is a critical feature of the multidimensional array implementation. Recommended optimizations include:
- Rank-specific overloads to better enable optimization of the member access operator.
- Inlining of a constexpr multi-index mapping expression that is not included in an optimizer's inlining budget.
- Compile-time evaluation statically determined portions of multi-index mapping expression.
Requires: indices is a multi-index in the domain:
- rank() <= N
- 0 <= indices[ith] < extent(ith).
- [Note: Because extent(ith) == 1 for rank() <= ith then extra zero-value indices are valid. --end note]
Returns: A reference to the element mapped to by indices....
Remark: Optimization of the mapping operator is a critical feature of the multidimensional array implementation. Recommended optimizations include:
- Rank-specific overloads to better enable optimization of the member access operator.
- Inlining of a constexpr multi-index mapping expression that is not included in an optimizer's inlining budget.
- Compile-time evaluation statically determined portions of multi-index mapping expression.
using layout = /* implmentation deduces from Properties... */ ;
Identification of the layout mapping. If Properties... does not include a layout property then layout is layout_right denoting the traditional C/C++ mapping.
A layout mapping is unique if each multi-index in the domain is mapped to a unique member in the codomain.
A layout mapping is contiguous if the codomain elements accessed through the layout mapping form a contiguous span.
A layout mapping that is both unique and contiguous is bijective and has size() == span().size().
A strided layout has constant striding between multi-index coordinates. Let A be an mdspan and indices_V... and indices_U... be multi-indices in the domain space such that all coordinates are equal except for the ith coordinate where indices_V[ith] = indices_U[ith] + 1. Then stride(ith) = distance(& A(indices_V...) - & A( indices_U... ) is constant for all coordinates.
Requires: is_strided().
Returns: When r < rank() the distance between members when the index of coordinate r is incremented by one, otherwise 0.
The mapping observers could be provided instead through a mapping type and object.
template <typename DataType, typename... Properties> class mdspan { public: using layout = /* extracted from Properties... */ ; using mapping = /* implementation defined */ ; const mapping & map() const noexcept ; struct mapping /* for exposition */ { using index_type = ptrdiff_t ; // observers of domain: static constexpr int rank() noexcept; static constexpr int rank_dynamic() noexcept; static constexpr index_type static_extent(int) noexcept; constexpr index_type extent(int) const noexcept; constexpr index_type size() const noexcept; // observers of the codomain: [0..span_size()) constexpr index_type span_size() const noexcept; // observers of the mapping from domain to codomain static constexpr bool is_always_unique = /* deduced */ ; static constexpr bool is_always_contiguous = /* deduced */ ; static constexpr bool is_always_strided = /* deduced */ ; constexpr bool is_unique() const noexcept; constexpr bool is_contiguous() const noexcept; constexpr bool is_strided() noexcept; constexpr index_type stride(int) const ; }; };
One of the valid members of an mdspan Properties... pack is an instantiation of template class extents. This property declares the rank and static extents of the mdspan type. Example:
using tensor = std::mdspan<double,std::extents<std::dynamic_extent,std::dynamic_extent,std::dynamic_extent>> ;
Note: A preferred, concise, and intuitive syntax for declaring the multidimensional index space of an mdspan is proposed in P0332.
namespace std { namespace experimental { template< ptrdiff_t ... StaticExtents > class extents { public: using index_type = ptrdiff_t ; // observers of the index space domain: // [0..extent(0))X[0..extent(1))X...X[0..extent(rank()-1)) static constexpr int rank() noexcept ; static constexpr int rank_dynamic() noexcept ; static constexpr index_type static_extent(int) noexcept ; constexpr index_type extent(int) const noexcept ; constexpr index_type size() const noexcept ; // constructors/assignment/destructor ~extents() = default ; constexpr extents(); constexpr extents(extents const &) = default ; constexpr extents(extents &&) = default ; extents & operator = (extents const &) noexcept = default ; extents & operator = (extents &&) noexcept = default ; template< class ... IndexType > constexpr extents( IndexType ... DynamicExtents ) noexcept ; }; }}
The detail::subspan_deduction_t is for exposition only to indicate that the implementation will require a metafunction to deduce the resulting mdspan type from U and slices....
Let the ith member of slices... be denoted by slices[ith].
Let an integral range be denoted by any of the following.
- an initializer_list<T> of integral type T and size 2
- a pair<T,T> of integral type T
- a tuple<T,T> of integral type T
- an array<T,2> of integral type T
- all to denote the range [0 .. U.extent(ith))
If slices[ith] is an integral range then let begin(slices[ith]) be the beginning of the integral range end(slices[ith]) be the end of the integral range. If slices[ith] is an integral value then let begin(slices[ith]) == slices[ith] and end(slices[ith]) == slices[ith]+1.
Requires:
- U.rank() == sizeof...(slices).
- Each member of the slices... pack is either an integral range or an integral value.
- 0 <= begin(slices[ith]) <= end(slices[ith]) <= U.extent(ith).
Returns: An mdspan V with a domain contained within the domain of U , codomain contained within the codomain of U , V.rank() is the number of integral ranges in slices... , U( begin(slices)... ) refers to the same codomain member refered to by the mapping the zero-index of V , each integral value in slices... contracts the corresponding extent of U.
Example:
// given U.rank() == 4 void foo( mdspan< DataType , Properties ... > const & U ) { auto V = subspan( U , make_pair(1,U.extent(0)-1) , 1 , make_pair(2,U.extent(2) , 2 ); assert( V.extent(0) == U.extent(0) - 2 ); assert( V.extent(1) == U.extent(2) - 2 ); assert( & V(0,0) == U(1,1,2,2) ); assert( & V(1,0) == U(2,1,2,2) ); assert( & V(0,1) == U(1,1,3,2) ); }
The proposed initializer_list, pair, tuple, and array slice specifier types define dynamic extents. When the all slice specifier references a static extent then the subspan's corresponding extent should be static as well. When the extent of a slice specifier is statically known there should be a slice specifier type to explicitly express this knowledge. Such a static extent slice specifier type is to-be-done.
An mdspan maps multi-indices from the domain to reference elements in the codomain by composing a layout mapping with a span of elements. The layout mapping is an extension point such that an mdspan may be instantiated with non-standard layout mappings.
The layout_right property denotes the C/C++ standard multidimensional array index mapping where the right-most extent is stride one and strides increase right-to-left as the product of extents.
The layout_left property denotes the FORTRAN standard multidimensional array index mapping where the left-most extent is stride one and strides increase left-to-right as the product of extents.
The layout_stride property denotes a multidimensional array index mapping with arbitrary strides for each extent. This is the layout for subarrays that are not contiguous.
The three standard layouts have the following layout mapping traits.
layout_right ; i.e., the C/C++ standard layout
is_always_unique == trueis_always_contiguous == trueis_always_strided == trueWhen 0 < rank() then stride(rank()-1) == 1 .When 1 < rank() then stride(r-1) = stride(r) * extent(r) for 0 < r < rank() ..For rank-two arrays (a.k.a., matrices) this is also known as row major layout.
layout_left ; i.e., the FORTRAN standard layout
is_always_unique == trueis_always_contiguous == trueis_always_strided == trueWhen 0 < rank() then stride(0) == 1 .When 1 < rank() then stride(r) = stride(r-1) * extent(r-1) for 0 < r < rank() ..For rank-two arrays (a.k.a., matrices) this is also known as column major layout.
layout_stride ; i.e., an arbitrary strided layout
is_always_unique == falseis_always_contiguous == falseis_always_strided == true
A layout class conforms to the following interface such that an mdspan can compose the layout mapping with its mdspan codomain member reference generation.
class layout_concept /* exposition only */ { public: template< ptrdiff_t ... StaticExtents > class mapping { public: // domain types using index_type = ptrdiff_t ; // constructors, copy, assignment, and destructor ~mapping() noexcept = default ; constexpr mapping() noexcept = default ; constexpr mapping(mapping const&) noexcept = default ; mapping& operator=(mapping const&) noexcept = default ; // observers of domain static constexpr int rank() noexcept; static constexpr int rank_dynamic() noexcept; static constexpr index_type static_extent(int) noexcept; constexpr index_type extent(int) const noexcept; constexpr index_type size() const noexcept; // observers of the codomain: [0..span_size()) constexpr index_type span_size() const noexcept; // observers of the mapping from domain to codomain static constexpr bool is_always_unique = /* deduced */ ; static constexpr bool is_always_contiguous = /* deduced */ ; static constexpr bool is_always_strided = /* deduced */ ; constexpr bool is_unique() const noexcept; constexpr bool is_contiguous() const noexcept; constexpr bool is_strided() noexcept; constexpr index_type stride(int) const noexcept; // mapping domain index to access codomain element template< class ... IndexType > constexpr index_type operator()( IndexType ... indices ) const noexcept; }; };
template< ptrdiff_t ... StaticExtents > class mapping
Requires: Let StaticExtents[ith] be the ith member of the pack. StaticExtents[ith] == std::dynamic_extent or 0 <= StaticExtents[ith].
Effects: Defines the domain index space where rank() == sizeof...(StaticExtents) and each StaticExtents[ith] == std::dynamic_extent denotes that ith extent coordinate is a dynamic extent.
constexpr mapping();
Effects: If static_extent(i) != std::dynamic_extent then extent(i) == static_extent(i) otherwise extent(i) == 0.
Constructors, assignment operators, and destructor requires and effects correspond to the corresponding members of mdspan .
Domain, codomain, and mapping observers requires and effects correspond to the corresponding members of mdspan .
Requires: rank() == sizeof...(indices) and 0 <= indices[ith] < extent(ith).
Returns: Layout mapping of indices... to codomain.
The proposed declaration mechanism for the multi-index domain space is verbose and unwieldy.
using tensor = std::mdspan<double,std::extents<std::dynamic_extent,std::dynamic_extent,std::dynamic_extent>> ;
The preferred mechanism is compact, is intuitive, LEWG has staw-polled strong preference, and users have voiced strong expressed preference.
using tensor = mdspan<double[][][]> ;
However, this syntax requires the trivial, non-functional language change proposed in P0332 to relax the definition of an incomplete array type.
Precedence:
There is precedence for using incomplete array types for dynamic extents.
- std::shared_ptr<T[]> and std::unique_ptr<T[]> denote a dynamic extent array through the incomplete type T[]
- P0674 denotes make_shared<T[][N1][N2]> to allocate a shared_ptr to a C style multidimensional array.
DataType
Requires: Is a complete or incomplete array type (8.3.4.p3). Each omitted static extent in the incomplete array type, [], denotes a dynamic extent.
using element_type = std::remove_all_extents<DataType>::type ;constexpr int rank() const { return std::rank<DataType>::value ; }static_extent(i) is std::extent_v<DataType,i>A dynamic extent is denoted when std::extent_v<DataType,i> == 0.The need for the magic number std::dynamic_extent is removed.
A minor revision of P0122 span is proposed in P0456 that would more closely align span with mdspan and enable span to have a similar extensibility for access properties.
template< typename DataType , class ... Properties > class span { public: // change element_type declaration: using element_type = std::remove_extent_t< DataType > ; };
Given P0456 the proposed span and mdspan could be merged into a single template class by simply requiring that all members specific to a one-dimensional span Requre that rank() == 1.
The reference type may be a proxy for accessing an element_type object. For example, if an atomic_access property were defined with the meaning that all access operations on codomain objects are atomic then the reference type must be an atomic reference type (paper P0019).
mdspan<int[],atomic_access> a( ptr , N ); static_assert( std::is_same_v< delctype(a(i)) , atomic_ref<int> > );
namespace std { namespace experimental { // bounds checking property template< bool Enable > struct bounds_check_if ; using bounds_check = bounds_check_if< true > ; }}
When mdspan Properties... includes bounds_check_if<true> then the mapping operators mdspan::operator() and mdspan::operator[] verify that each index is valid, 0 <= index[ith] < extent(ith). Verification failure shall be reported.