std::linalg accessors and transposed_layout

docs/libcudacxx/standard_api.rst

@@ -112,7 +112,14 @@ Feature availability:
 
 -  C++26 ``std::dims`` is available in C++14.
 
--  C++23 ``forward_like``, ``to_underlying`` and ``unreachable`` from ``<utility>`` are available in C++11.
+-  C++26 ``std::linalg`` accessors, transposed layout, and related functions are available in C++17.
+
+   - ``scaled()`` and ``scaled_accessor``
+   - ``conjugated()`` and ``conjugated_accessor``
+   - ``transposed()`` and ``layout_transpose``
+   - ``conjugate_transposed()``
+
+-  C++23 ``forward_like``, ``to_underlying``, and ``unreachable`` from ``<utility>`` are available in C++11.
 
 -  C++23 ``is_scoped_enum`` in ``<type_traits>`` is available in C++11.
 

docs/libcudacxx/standard_api/numerics_library.rst

@@ -10,6 +10,7 @@ Numerics Library
    numerics_library/bit
    numerics_library/complex
    numerics_library/numeric
+   numerics_library/linalg
 
 Any Standard C++ header not listed below is omitted.
 

docs/libcudacxx/standard_api/numerics_library/linalg.rst

@@ -0,0 +1,31 @@
+.. _libcudacxx-standard-api-numerics-linalg:
+
+``<cuda/std/linalg>``
+============================================
+
+Provided functionalities
+------------------------
+
+- ``scaled()`` `std::linalg::scaled <https://en.cppreference.com/w/cpp/numeric/linalg/scaled>`_
+- ``scaled_accessor`` `std::linalg::scaled_accessor <https://en.cppreference.com/w/cpp/numeric/linalg/scaled_accessor>`_
+- ``conjugated()`` `std::linalg::conjugated <https://en.cppreference.com/w/cpp/numeric/linalg/conjugated>`_
+- ``conjugated_accessor`` `std::linalg::conjugated_accessor <https://en.cppreference.com/w/cpp/numeric/linalg/conjugated_accessor>`_
+- ``transposed()`` `std::linalg::transposed <https://en.cppreference.com/w/cpp/numeric/linalg/transposed>`_
+- ``layout_transpose`` `std::linalg::layout_transpose <https://en.cppreference.com/w/cpp/numeric/linalg/layout_transpose>`_
+- ``conjugate_transposed()`` `std::linalg::conjugate_transposed <https://en.cppreference.com/w/cpp/numeric/linalg/conjugate_transposed>`_
+
+Extensions
+----------
+
+-  C++26 ``std::linalg`` accessors, transposed layout, and related functions are available in C++17
+
+Omissions
+---------
+
+-  Currently we do not expose any BLAS functions and layouts.
+
+Restrictions
+------------
+
+-  On device no exceptions are thrown in case of a bad access.
+-  MSVC is only supported with C++20

libcudacxx/include/cuda/std/__linalg/conj_if_needed.h

@@ -0,0 +1,78 @@
+//@HEADER
+// ************************************************************************
+//
+//                        Kokkos v. 4.0
+//       Copyright (2022) National Technology & Engineering
+//               Solutions of Sandia, LLC (NTESS).
+//
+// Under the terms of Contract DE-NA0003525 with NTESS,
+// the U.S. Government retains certain rights in this software.
+//
+// Part of Kokkos, under the Apache License v2.0 with LLVM Exceptions.
+// See https://kokkos.org/LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// ************************************************************************
+//@HEADER
+
+#ifndef _LIBCUDACXX___LINALG_CONJUGATE_IF_NEEDED_HPP
+#define _LIBCUDACXX___LINALG_CONJUGATE_IF_NEEDED_HPP
+
+#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/version>
+
+#if defined(__cccl_lib_mdspan) && _CCCL_STD_VER >= 2017
+
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__type_traits/is_arithmetic.h>
+#  include <cuda/std/complex>
+
+_LIBCUDACXX_BEGIN_NAMESPACE_STD
+
+namespace linalg
+{
+
+_LIBCUDACXX_BEGIN_NAMESPACE_CPO(__conj_if_needed)
+
+template <class _Type>
+_CCCL_CONCEPT _HasConj = _CCCL_REQUIRES_EXPR((_Type), _Type __a)(static_cast<void>(_CUDA_VSTD::conj(__a)));
+
+struct __conj_if_needed
+{
+  template <class _Type>
+  _LIBCUDACXX_HIDE_FROM_ABI constexpr auto operator()(const _Type& __t) const
+  {
+    if constexpr (is_arithmetic_v<_Type> || !_HasConj<_Type>)
+    {
+      return __t;
+    }
+    else
+    {
+      return _CUDA_VSTD::conj(__t);
+    }
+    _CCCL_UNREACHABLE();
+  }
+};
+
+_LIBCUDACXX_END_NAMESPACE_CPO
+
+inline namespace __cpo
+{
+_CCCL_GLOBAL_CONSTANT auto conj_if_needed = __conj_if_needed::__conj_if_needed{};
+
+} // namespace __cpo
+} // end namespace linalg
+
+_LIBCUDACXX_END_NAMESPACE_STD
+
+#endif // defined(__cccl_lib_mdspan) && _CCCL_STD_VER >= 2017
+#endif // _LIBCUDACXX___LINALG_CONJUGATED_HPP

libcudacxx/include/cuda/std/__linalg/conjugate_transposed.h

@@ -0,0 +1,55 @@
+//@HEADER
+// ************************************************************************
+//
+//                        Kokkos v. 4.0
+//       Copyright (2022) National Technology & Engineering
+//               Solutions of Sandia, LLC (NTESS).
+//
+// Under the terms of Contract DE-NA0003525 with NTESS,
+// the U.S. Government retains certain rights in this software.
+//
+// Part of Kokkos, under the Apache License v2.0 with LLVM Exceptions.
+// See https://kokkos.org/LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// ************************************************************************
+//@HEADER
+
+#ifndef _LIBCUDACXX___LINALG_CONJUGATE_TRANSPOSED_HPP
+#define _LIBCUDACXX___LINALG_CONJUGATE_TRANSPOSED_HPP
+
+#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/version>
+
+#if defined(__cccl_lib_mdspan) && _CCCL_STD_VER >= 2017
+
+#  include <cuda/std/__linalg/conjugated.h>
+#  include <cuda/std/__linalg/transposed.h>
+
+_LIBCUDACXX_BEGIN_NAMESPACE_STD
+
+namespace linalg
+{
+
+template <class _ElementType, class _Extents, class _Layout, class _Accessor>
+_CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI constexpr auto
+conjugate_transposed(mdspan<_ElementType, _Extents, _Layout, _Accessor> __a)
+{
+  return conjugated(transposed(__a));
+}
+
+} // end namespace linalg
+
+_LIBCUDACXX_END_NAMESPACE_STD
+
+#endif // defined(__cccl_lib_mdspan) && _CCCL_STD_VER >= 2017
+#endif // _LIBCUDACXX___LINALG_CONJUGATE_TRANSPOSED_HPP

libcudacxx/include/cuda/std/__linalg/conjugated.h

@@ -0,0 +1,141 @@
+//@HEADER
+// ************************************************************************
+//
+//                        Kokkos v. 4.0
+//       Copyright (2022) National Technology & Engineering
+//               Solutions of Sandia, LLC (NTESS).
+//
+// Under the terms of Contract DE-NA0003525 with NTESS,
+// the U.S. Government retains certain rights in this software.
+//
+// Part of Kokkos, under the Apache License v2.0 with LLVM Exceptions.
+// See https://kokkos.org/LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// ************************************************************************
+//@HEADER
+
+#ifndef _LIBCUDACXX___LINALG_CONJUGATED_HPP
+#define _LIBCUDACXX___LINALG_CONJUGATED_HPP
+
+#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/version>
+
+#if defined(__cccl_lib_mdspan) && _CCCL_STD_VER >= 2017
+
+#  include <cuda/std/__linalg/conj_if_needed.h>
+#  include <cuda/std/__type_traits/add_const.h>
+#  include <cuda/std/__type_traits/is_arithmetic.h>
+#  include <cuda/std/__type_traits/remove_const.h>
+#  include <cuda/std/__utility/declval.h>
+#  include <cuda/std/mdspan>
+
+_LIBCUDACXX_BEGIN_NAMESPACE_STD
+
+namespace linalg
+{
+
+template <class _NestedAccessor>
+class conjugated_accessor
+{
+private:
+  using __nested_element_type = typename _NestedAccessor::element_type;
+  using __nc_result_type      = decltype(conj_if_needed(_CUDA_VSTD::declval<__nested_element_type>()));
+
+public:
+  using element_type     = add_const_t<__nc_result_type>;
+  using reference        = remove_const_t<element_type>;
+  using data_handle_type = typename _NestedAccessor::data_handle_type;
+  using offset_policy    = conjugated_accessor<typename _NestedAccessor::offset_policy>;
+
+  _CCCL_HIDE_FROM_ABI constexpr conjugated_accessor() = default;
+
+  _LIBCUDACXX_HIDE_FROM_ABI constexpr conjugated_accessor(const _NestedAccessor& __acc)
+      : __nested_accessor_(__acc)
+  {}
+
+  _CCCL_TEMPLATE(class _OtherNestedAccessor)
+  _CCCL_REQUIRES(_CCCL_TRAIT(is_constructible, _NestedAccessor, const _OtherNestedAccessor&)
+                   _CCCL_AND _CCCL_TRAIT(is_convertible, _OtherNestedAccessor, _NestedAccessor))
+  _LIBCUDACXX_HIDE_FROM_ABI constexpr conjugated_accessor(const conjugated_accessor<_OtherNestedAccessor>& __other)
+      : __nested_accessor_(__other.nested_accessor())
+  {}
+
+  _CCCL_TEMPLATE(class _OtherNestedAccessor)
+  _CCCL_REQUIRES(_CCCL_TRAIT(is_constructible, _NestedAccessor, const _OtherNestedAccessor&)
+                   _CCCL_AND(!_CCCL_TRAIT(is_convertible, _OtherNestedAccessor, _NestedAccessor)))
+  _LIBCUDACXX_HIDE_FROM_ABI explicit constexpr conjugated_accessor(
+    const conjugated_accessor<_OtherNestedAccessor>& __other)
+      : __nested_accessor_(__other.nested_accessor())
+  {}
+
+  _LIBCUDACXX_HIDE_FROM_ABI constexpr reference access(data_handle_type __p, size_t __i) const noexcept
+  {
+    return conj_if_needed(__nested_element_type(__nested_accessor_.access(__p, __i)));
+  }
+
+  _CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI constexpr typename offset_policy::data_handle_type
+  offset(data_handle_type __p, size_t __i) const noexcept
+  {
+    return __nested_accessor_.offset(__p, __i);
+  }
+
+  _CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI constexpr const _NestedAccessor& nested_accessor() const noexcept
+  {
+    return __nested_accessor_;
+  }
+
+private:
+  _NestedAccessor __nested_accessor_;
+};
+
+template <class _ElementType, class _Extents, class _Layout, class _Accessor>
+_CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI constexpr auto
+conjugated(mdspan<_ElementType, _Extents, _Layout, _Accessor> __a)
+{
+  using __value_type = typename decltype(__a)::value_type;
+  // Current status of [linalg] only optimizes if _Accessor is conjugated_accessor<_Accessor> for some _Accessor.
+  // There's __a separate specialization for that case below.
+
+  // P3050 optimizes conjugated's accessor type for when we know that it can't be complex: arithmetic types,
+  // and types for which `conj` is not ADL-findable.
+  if constexpr (is_arithmetic_v<__value_type> || !__conj_if_needed::_HasConj<__value_type>)
+  {
+    return mdspan<_ElementType, _Extents, _Layout, _Accessor>(__a.data_handle(), __a.mapping(), __a.accessor());
+  }
+  else
+  {
+    using __return_element_type  = typename conjugated_accessor<_Accessor>::element_type;
+    using __return_accessor_type = conjugated_accessor<_Accessor>;
+    return mdspan<__return_element_type, _Extents, _Layout, __return_accessor_type>{
+      __a.data_handle(), __a.mapping(), __return_accessor_type(__a.accessor())};
+  }
+  _CCCL_UNREACHABLE();
+}
+
+// Conjugation is self-annihilating
+template <class _ElementType, class _Extents, class _Layout, class _NestedAccessor>
+_CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI constexpr auto
+conjugated(mdspan<_ElementType, _Extents, _Layout, conjugated_accessor<_NestedAccessor>> __a)
+{
+  using __return_element_type  = typename _NestedAccessor::element_type;
+  using __return_accessor_type = _NestedAccessor;
+  return mdspan<__return_element_type, _Extents, _Layout, __return_accessor_type>(
+    __a.data_handle(), __a.mapping(), __a.accessor().nested_accessor());
+}
+
+} // end namespace linalg
+
+_LIBCUDACXX_END_NAMESPACE_STD
+
+#endif // defined(__cccl_lib_mdspan) && _CCCL_STD_VER >= 2017
+#endif // _LIBCUDACXX___LINALG_CONJUGATED_HPP