Replace _LIBCUDACXX_CPO_ACCESSIBILITY with _CCCL_GLOBAL_CONSTANT

NVIDIA · Jul 30, 2024 · e4d1d1f · e4d1d1f
1 parent 59d7a4b
commit e4d1d1f
Show file tree

Hide file tree

Showing 31 changed files with 83 additions and 76 deletions.
diff --git a/cudax/include/cuda/experimental/__hierarchy/hierarchy_levels.cuh b/cudax/include/cuda/experimental/__hierarchy/hierarchy_levels.cuh
@@ -140,7 +140,7 @@ struct grid_level
   using allowed_above = allowed_levels<>;
   using allowed_below = allowed_levels<block_level, cluster_level>;
 };
-_LIBCUDACXX_CPO_ACCESSIBILITY grid_level grid;
+_CCCL_GLOBAL_CONSTANT grid_level grid;
 
 /**
  * @brief Type representing the cluster level in CUDA thread hierarchy
@@ -158,7 +158,7 @@ struct cluster_level
   using allowed_above = allowed_levels<grid_level>;
   using allowed_below = allowed_levels<block_level>;
 };
-_LIBCUDACXX_CPO_ACCESSIBILITY cluster_level cluster;
+_CCCL_GLOBAL_CONSTANT cluster_level cluster;
 
 /**
  * @brief Type representing the block level in CUDA thread hierarchy
@@ -176,7 +176,7 @@ struct block_level
   using allowed_above = allowed_levels<grid_level, cluster_level>;
   using allowed_below = allowed_levels<thread_level>;
 };
-_LIBCUDACXX_CPO_ACCESSIBILITY block_level block;
+_CCCL_GLOBAL_CONSTANT block_level block;
 
 /**
  * @brief Type representing the thread level in CUDA thread hierarchy
@@ -194,7 +194,7 @@ struct thread_level
   using allowed_above = allowed_levels<block_level>;
   using allowed_below = allowed_levels<>;
 };
-_LIBCUDACXX_CPO_ACCESSIBILITY thread_level thread;
+_CCCL_GLOBAL_CONSTANT thread_level thread;
 
 template <typename Level>
 constexpr bool is_core_cuda_hierarchy_level =

diff --git a/libcudacxx/include/cuda/__memory_resource/get_property.h b/libcudacxx/include/cuda/__memory_resource/get_property.h
@@ -176,7 +176,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto get_property = __get_property::__fn{};
+_CCCL_GLOBAL_CONSTANT auto get_property = __get_property::__fn{};
 } // namespace __cpo
 
 _LIBCUDACXX_END_NAMESPACE_CUDA

diff --git a/libcudacxx/include/cuda/std/__cccl/dialect.h b/libcudacxx/include/cuda/std/__cccl/dialect.h
@@ -90,4 +90,25 @@
 #  define _CCCL_TRAIT(__TRAIT, ...) __TRAIT<__VA_ARGS__>::value
 #endif // _CCCL_STD_VER <= 2014
 
+// In nvcc prior to 11.3 global variables could not be marked constexpr
+#if defined(_CCCL_CUDACC_BELOW_11_3)
+#  define _CCCL_CONSTEXPR_GLOBAL const
+#else // ^^^ _CCCL_CUDACC_BELOW_11_3 ^^^ / vvv !_CCCL_CUDACC_BELOW_11_3 vvv
+#  define _CCCL_CONSTEXPR_GLOBAL constexpr
+#endif // !_CCCL_CUDACC_BELOW_11_3
+
+// Inline variables are only available from C++17 onwards
+#if _CCCL_STD_VER >= 2017 && defined(__cpp_inline_variables) && (__cpp_inline_variables >= 201606L)
+#  define _CCCL_INLINE_VAR inline
+#else // ^^^ C++14 ^^^ / vvv C++17 vvv
+#  define _CCCL_INLINE_VAR
+#endif // _CCCL_STD_VER <= 2014
+
+// We need to treat host and device separately
+#if defined(__CUDA_ARCH__)
+#  define _CCCL_GLOBAL_CONSTANT _CCCL_DEVICE _CCCL_CONSTEXPR_GLOBAL
+#else // ^^^ __CUDA_ARCH__ ^^^ / vvv !__CUDA_ARCH__ vvv
+#  define _CCCL_GLOBAL_CONSTANT _CCCL_INLINE_VAR constexpr
+#endif // __CUDA_ARCH__
+
 #endif // __CCCL_DIALECT_H
diff --git a/libcudacxx/include/cuda/std/__concepts/swappable.h b/libcudacxx/include/cuda/std/__concepts/swappable.h
@@ -158,7 +158,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto swap = __swap::__fn{};
+_CCCL_GLOBAL_CONSTANT auto swap = __swap::__fn{};
 } // namespace __cpo
 _LIBCUDACXX_END_NAMESPACE_RANGES
 

diff --git a/libcudacxx/include/cuda/std/__expected/unexpect.h b/libcudacxx/include/cuda/std/__expected/unexpect.h
@@ -28,7 +28,7 @@ struct unexpect_t
   explicit unexpect_t() = default;
 };
 
-_LIBCUDACXX_CPO_ACCESSIBILITY unexpect_t unexpect{};
+_CCCL_GLOBAL_CONSTANT unexpect_t unexpect{};
 
 _LIBCUDACXX_END_NAMESPACE_STD
 

diff --git a/libcudacxx/include/cuda/std/__iterator/access.h b/libcudacxx/include/cuda/std/__iterator/access.h
@@ -53,7 +53,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto begin = __begin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto begin = __begin::__fn{};
 } // namespace __cpo
 
 namespace __end
@@ -84,7 +84,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto end = __end::__fn{};
+_CCCL_GLOBAL_CONSTANT auto end = __end::__fn{};
 } // namespace __cpo
 
 #if _CCCL_STD_VER >= 2014
@@ -104,7 +104,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto cbegin = __cbegin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto cbegin = __cbegin::__fn{};
 } // namespace __cpo
 
 namespace __cend
@@ -122,7 +122,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto cend = __cend::__fn{};
+_CCCL_GLOBAL_CONSTANT auto cend = __cend::__fn{};
 } // namespace __cpo
 
 #endif // _CCCL_STD_VER >= 2014

diff --git a/libcudacxx/include/cuda/std/__iterator/advance.h b/libcudacxx/include/cuda/std/__iterator/advance.h
@@ -238,7 +238,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto advance = __advance::__fn{};
+_CCCL_GLOBAL_CONSTANT auto advance = __advance::__fn{};
 } // namespace __cpo
 
 _LIBCUDACXX_END_NAMESPACE_RANGES

diff --git a/libcudacxx/include/cuda/std/__iterator/distance.h b/libcudacxx/include/cuda/std/__iterator/distance.h
@@ -116,7 +116,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto distance = __distance::__fn{};
+_CCCL_GLOBAL_CONSTANT auto distance = __distance::__fn{};
 } // namespace __cpo
 _LIBCUDACXX_END_NAMESPACE_RANGES
 

diff --git a/libcudacxx/include/cuda/std/__iterator/iter_move.h b/libcudacxx/include/cuda/std/__iterator/iter_move.h
@@ -119,7 +119,7 @@ struct __fn
 _LIBCUDACXX_END_NAMESPACE_CPO
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto iter_move = __iter_move::__fn{};
+_CCCL_GLOBAL_CONSTANT auto iter_move = __iter_move::__fn{};
 } // namespace __cpo
 _LIBCUDACXX_END_NAMESPACE_RANGES
 

diff --git a/libcudacxx/include/cuda/std/__iterator/iter_swap.h b/libcudacxx/include/cuda/std/__iterator/iter_swap.h
@@ -119,7 +119,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto iter_swap = __iter_swap::__fn{};
+_CCCL_GLOBAL_CONSTANT auto iter_swap = __iter_swap::__fn{};
 } // namespace __cpo
 _LIBCUDACXX_END_NAMESPACE_RANGES
 

diff --git a/libcudacxx/include/cuda/std/__iterator/next.h b/libcudacxx/include/cuda/std/__iterator/next.h
@@ -90,7 +90,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto next = __next::__fn{};
+_CCCL_GLOBAL_CONSTANT auto next = __next::__fn{};
 } // namespace __cpo
 _LIBCUDACXX_END_NAMESPACE_RANGES
 

diff --git a/libcudacxx/include/cuda/std/__iterator/prev.h b/libcudacxx/include/cuda/std/__iterator/prev.h
@@ -81,7 +81,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto prev = __prev::__fn{};
+_CCCL_GLOBAL_CONSTANT auto prev = __prev::__fn{};
 } // namespace __cpo
 _LIBCUDACXX_END_NAMESPACE_RANGES
 

diff --git a/libcudacxx/include/cuda/std/__iterator/reverse_access.h b/libcudacxx/include/cuda/std/__iterator/reverse_access.h
@@ -65,7 +65,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto rbegin = __rbegin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto rbegin = __rbegin::__fn{};
 } // namespace __cpo
 
 namespace __rend
@@ -104,7 +104,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto rend = __rend::__fn{};
+_CCCL_GLOBAL_CONSTANT auto rend = __rend::__fn{};
 } // namespace __cpo
 
 namespace __crbegin
@@ -122,7 +122,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto crbegin = __crbegin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto crbegin = __crbegin::__fn{};
 } // namespace __cpo
 
 namespace __crend
@@ -140,7 +140,7 @@ struct __fn
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto crend = __crend::__fn{};
+_CCCL_GLOBAL_CONSTANT auto crend = __crend::__fn{};
 } // namespace __cpo
 
 #endif // _CCCL_STD_VER >= 2014

diff --git a/libcudacxx/include/cuda/std/__iterator/unreachable_sentinel.h b/libcudacxx/include/cuda/std/__iterator/unreachable_sentinel.h
@@ -82,7 +82,7 @@ struct unreachable_sentinel_t : __unreachable_sentinel_detail::__unreachable_bas
 
 _LIBCUDACXX_END_NAMESPACE_RANGES_ABI
 
-_LIBCUDACXX_CPO_ACCESSIBILITY unreachable_sentinel_t unreachable_sentinel{};
+_CCCL_GLOBAL_CONSTANT unreachable_sentinel_t unreachable_sentinel{};
 _LIBCUDACXX_END_NAMESPACE_STD
 
 #endif // _CCCL_STD_VER > 2014

diff --git a/libcudacxx/include/cuda/std/__mdspan/standard_layout_static_array.h b/libcudacxx/include/cuda/std/__mdspan/standard_layout_static_array.h
@@ -77,12 +77,11 @@ namespace __detail
 
 struct __construct_psa_from_dynamic_exts_values_tag_t
 {};
-_LIBCUDACXX_CPO_ACCESSIBILITY __construct_psa_from_dynamic_exts_values_tag_t
-  __construct_psa_from_dynamic_exts_values_tag;
+_CCCL_GLOBAL_CONSTANT __construct_psa_from_dynamic_exts_values_tag_t __construct_psa_from_dynamic_exts_values_tag;
 
 struct __construct_psa_from_all_exts_values_tag_t
 {};
-_LIBCUDACXX_CPO_ACCESSIBILITY __construct_psa_from_all_exts_values_tag_t __construct_psa_from_all_exts_values_tag;
+_CCCL_GLOBAL_CONSTANT __construct_psa_from_all_exts_values_tag_t __construct_psa_from_all_exts_values_tag;
 
 struct __construct_psa_from_all_exts_array_tag_t
 {};

diff --git a/libcudacxx/include/cuda/std/__ranges/access.h b/libcudacxx/include/cuda/std/__ranges/access.h
@@ -125,7 +125,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto begin = __begin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto begin = __begin::__fn{};
 } // namespace __cpo
 
 // [range.range]
@@ -213,7 +213,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto end = __end::__fn{};
+_CCCL_GLOBAL_CONSTANT auto end = __end::__fn{};
 } // namespace __cpo
 
 // [range.access.cbegin]
@@ -243,7 +243,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto cbegin = __cbegin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto cbegin = __cbegin::__fn{};
 } // namespace __cpo
 
 // [range.access.cend]
@@ -273,7 +273,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto cend = __cend::__fn{};
+_CCCL_GLOBAL_CONSTANT auto cend = __cend::__fn{};
 } // namespace __cpo
 #endif // _CCCL_STD_VER > 2014 && !_CCCL_COMPILER_MSVC_2017
 

diff --git a/libcudacxx/include/cuda/std/__ranges/data.h b/libcudacxx/include/cuda/std/__ranges/data.h
@@ -97,7 +97,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto data = __data::__fn{};
+_CCCL_GLOBAL_CONSTANT auto data = __data::__fn{};
 } // namespace __cpo
 
 // [range.prim.cdata]
@@ -127,7 +127,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto cdata = __cdata::__fn{};
+_CCCL_GLOBAL_CONSTANT auto cdata = __cdata::__fn{};
 } // namespace __cpo
 
 #endif // _CCCL_STD_VER >= 2017 && !_CCCL_COMPILER_MSVC_2017

diff --git a/libcudacxx/include/cuda/std/__ranges/empty.h b/libcudacxx/include/cuda/std/__ranges/empty.h
@@ -102,7 +102,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto empty = __empty::__fn{};
+_CCCL_GLOBAL_CONSTANT auto empty = __empty::__fn{};
 } // namespace __cpo
 
 #endif // _CCCL_STD_VER >= 2017 && !_CCCL_COMPILER_MSVC_2017

diff --git a/libcudacxx/include/cuda/std/__ranges/rbegin.h b/libcudacxx/include/cuda/std/__ranges/rbegin.h
@@ -130,7 +130,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto rbegin = __rbegin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto rbegin = __rbegin::__fn{};
 } // namespace __cpo
 
 // [range.access.crbegin]
@@ -160,7 +160,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto crbegin = __crbegin::__fn{};
+_CCCL_GLOBAL_CONSTANT auto crbegin = __crbegin::__fn{};
 } // namespace __cpo
 
 #endif // _CCCL_STD_VER >= 2017 && && !_CCCL_COMPILER_MSVC_2017

diff --git a/libcudacxx/include/cuda/std/__ranges/rend.h b/libcudacxx/include/cuda/std/__ranges/rend.h
@@ -137,7 +137,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto rend = __rend::__fn{};
+_CCCL_GLOBAL_CONSTANT auto rend = __rend::__fn{};
 } // namespace __cpo
 
 // [range.access.crend]
@@ -167,7 +167,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto crend = __crend::__fn{};
+_CCCL_GLOBAL_CONSTANT auto crend = __crend::__fn{};
 } // namespace __cpo
 
 #endif // _CCCL_STD_VER >= 2017 && !_CCCL_COMPILER_MSVC_2017

diff --git a/libcudacxx/include/cuda/std/__ranges/size.h b/libcudacxx/include/cuda/std/__ranges/size.h
@@ -157,7 +157,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto size = __size::__fn{};
+_CCCL_GLOBAL_CONSTANT auto size = __size::__fn{};
 } // namespace __cpo
 
 // [range.prim.ssize]
@@ -198,7 +198,7 @@ _LIBCUDACXX_END_NAMESPACE_CPO
 
 inline namespace __cpo
 {
-_LIBCUDACXX_CPO_ACCESSIBILITY auto ssize = __ssize::__fn{};
+_CCCL_GLOBAL_CONSTANT auto ssize = __ssize::__fn{};
 } // namespace __cpo
 
 #endif // _CCCL_STD_VER >= 2017 && !_CCCL_COMPILER_MSVC_2017

diff --git a/libcudacxx/include/cuda/std/__utility/in_place.h b/libcudacxx/include/cuda/std/__utility/in_place.h
@@ -32,7 +32,7 @@ struct _LIBCUDACXX_TYPE_VIS in_place_t
 {
   explicit in_place_t() = default;
 };
-_LIBCUDACXX_CPO_ACCESSIBILITY in_place_t in_place{};
+_CCCL_GLOBAL_CONSTANT in_place_t in_place{};
 
 template <class _Tp>
 struct _LIBCUDACXX_TEMPLATE_VIS in_place_type_t

diff --git a/libcudacxx/include/cuda/std/detail/libcxx/include/__config b/libcudacxx/include/cuda/std/detail/libcxx/include/__config
@@ -1743,19 +1743,6 @@ __sanitizer_annotate_contiguous_container(const void*, const void*, const void*,
 #    define _LIBCUDACXX_CTAD_SUPPORTED_FOR_TYPE(_ClassName) static_assert(true, "")
 #  endif
 
-#  if (defined(__CUDACC_VER_MAJOR__) && __CUDACC_VER_MAJOR__ <= 11) \
-    && (defined(__CUDACC_VER_MINOR__) && __CUDACC_VER_MINOR__ <= 2)
-#    define _LIBCUDACXX_CONSTEXPR_GLOBAL const
-#  else
-#    define _LIBCUDACXX_CONSTEXPR_GLOBAL constexpr
-#  endif
-
-#  if defined(__CUDA_ARCH__)
-#    define _LIBCUDACXX_CPO_ACCESSIBILITY _CCCL_DEVICE _LIBCUDACXX_CONSTEXPR_GLOBAL
-#  else
-#    define _LIBCUDACXX_CPO_ACCESSIBILITY _LIBCUDACXX_INLINE_VAR constexpr
-#  endif
-
 // Older nvcc do not handle the constraint of `construct_at` in earlier std modes
 // So to preserve our performance optimization we default to the unconstrained
 // `__construct_at` and only in C++20 use `construct_at`

diff --git a/libcudacxx/include/cuda/std/detail/libcxx/include/optional b/libcudacxx/include/cuda/std/detail/libcxx/include/optional
@@ -265,7 +265,7 @@ struct nullopt_t
   _LIBCUDACXX_INLINE_VISIBILITY constexpr explicit nullopt_t(__secret_tag, __secret_tag) noexcept {}
 };
 
-_LIBCUDACXX_CPO_ACCESSIBILITY nullopt_t nullopt{nullopt_t::__secret_tag{}, nullopt_t::__secret_tag{}};
+_CCCL_GLOBAL_CONSTANT nullopt_t nullopt{nullopt_t::__secret_tag{}, nullopt_t::__secret_tag{}};
 
 struct __optional_construct_from_invoke_tag
 {};