Add bound checks for each dimension of mdspan

docs/libcudacxx/standard_api.rst

@@ -103,15 +103,15 @@ Feature availability:
 
 -  C++23 ``invoke_r`` in ``<functional>`` is available in C++14.
 
--  C++23 ``<mdspan>`` is available in C++17.
+-  C++23 ``<mdspan>`` is available in C++14.
 
-   -  ``mdspan`` is feature complete in C++17 onwards.
+   -  ``mdspan`` is feature complete in C++14 onwards.
    -  ``mdspan`` on msvc is only supported in C++20 and onwards.
 
+-  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++23 ``is_scoped_enum`` in ``<type_traits>`` is available in C++11.
 
--  C++26 ``std::dims`` is available in C++17.
-
 -  C++26 tuple protocol to ``complex`` is available in C++11.

docs/libcudacxx/standard_api/container_library/mdspan.rst

@@ -6,8 +6,9 @@
 Extensions
 ----------
 
--  All features of ``<mdspan>`` are made available in C++17 onwards
--  C++26 ``std::dims`` is made available in C++17 onwards
+-  All features of ``<mdspan>`` are made available in C++14 onwards
+-  The C++23 multidimensional ``operator[]`` is replaced with ``operator()``
+-  C++26 ``std::dims`` is made available in C++14 onwards
 
 Restrictions
 ------------

libcudacxx/include/cuda/std/__mdspan/mdspan.h

@@ -54,6 +54,8 @@
 #  pragma system_header
 #endif // no system header
 
+#include <cuda/std/__algorithm/all_of.h>
+#include <cuda/std/__functional/identity.h>
 #include <cuda/std/__mdspan/compressed_pair.h>
 #include <cuda/std/__mdspan/default_accessor.h>
 #include <cuda/std/__mdspan/extents.h>
@@ -64,6 +66,7 @@
 #include <cuda/std/__type_traits/is_convertible.h>
 #include <cuda/std/__type_traits/is_default_constructible.h>
 #include <cuda/std/__type_traits/is_nothrow_constructible.h>
+#include <cuda/std/__type_traits/is_signed.h>
 #include <cuda/std/__type_traits/rank.h>
 #include <cuda/std/__type_traits/remove_all_extents.h>
 #include <cuda/std/__type_traits/remove_cv.h>
@@ -79,6 +82,8 @@ _LIBCUDACXX_BEGIN_NAMESPACE_STD
 
 #if _CCCL_STD_VER >= 2014
 
+_CCCL_NV_DIAG_SUPPRESS(186) // pointless comparison of unsigned integer with zero
+
 template <class _ElementType,
           class _Extents,
           class _LayoutPolicy   = layout_right,
@@ -108,28 +113,45 @@ class mdspan
       return (__self.rank() > 0)
           && __MDSPAN_FOLD_OR((__self.__mapping_ref().extents().template __extent<_Idxs>() == index_type(0)));
     }
+
+    template <class... _SizeTypes>
+    _CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI static constexpr bool
+    __check_index(_Extents const& exts, _SizeTypes... __indices)
+    {
+#  if _CCCL_STD_VER >= 2017
+      return (((is_unsigned_v<index_type> ? true : static_cast<index_type>(__indices) >= 0)
+               && static_cast<index_type>(__indices) < exts.extent(_Idxs))
+              && ...);
+#  else
+      return true;
+#  endif // _CCCL_STD_VER >= 2017
+    }
+
     template <class... _SizeTypes>
     _CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI static constexpr index_type
     __index(mdspan const& __self, _SizeTypes... __indices) noexcept
     {
+      _CCCL_ASSERT(__check_index(__self.__mapping_ref().extents(), __indices...),
+                   "cuda::std::mdspan subscript out of range!");
       const index_type __res = __self.__mapping_ref()(index_type(__indices)...);
-      _CCCL_ASSERT(__res < __self.__mapping_ref().required_span_size(), "cuda::std::mdspan subscript out of range!");
       return __res;
     }
     template <class _SizeType, size_t _Np>
     _CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI static constexpr index_type
     __index(mdspan const& __self, const _CUDA_VSTD::array<_SizeType, _Np>& __indices) noexcept
     {
+      _CCCL_ASSERT(__check_index(__self.__mapping_ref().extents(), __indices[_Idxs]...),
+                   "cuda::std::mdspan subscript out of range!");
       const index_type __res = __self.__mapping_ref()(__indices[_Idxs]...);
-      _CCCL_ASSERT(__res < __self.__mapping_ref().required_span_size(), "cuda::std::mdspan subscript out of range!");
       return __res;
     }
     template <class _SizeType, size_t _Np>
     _CCCL_NODISCARD _LIBCUDACXX_HIDE_FROM_ABI static constexpr index_type
     __index(mdspan const& __self, const _CUDA_VSTD::span<_SizeType, _Np>& __indices) noexcept
     {
+      _CCCL_ASSERT(__check_index(__self.__mapping_ref().extents(), __indices[_Idxs]...),
+                   "cuda::std::mdspan subscript out of range!");
       const index_type __res = __self.__mapping_ref()(__indices[_Idxs]...);
-      _CCCL_ASSERT(__res < __self.__mapping_ref().required_span_size(), "cuda::std::mdspan subscript out of range!");
       return __res;
     }
   };
@@ -488,6 +510,8 @@ _CCCL_HOST_DEVICE mdspan(const typename _AccessorType::data_handle_type, const _
             _AccessorType>;
 #  endif // __MDSPAN_USE_CLASS_TEMPLATE_ARGUMENT_DEDUCTION
 
+_CCCL_NV_DIAG_DEFAULT(186)
+
 #endif // _CCCL_STD_VER >= 2014
 
 _LIBCUDACXX_END_NAMESPACE_STD