backport unreachable

cub/cub/agent/agent_sub_warp_merge_sort.cuh

@@ -45,6 +45,8 @@
 
 #include <thrust/system/cuda/detail/core/util.h>
 
+#include <cuda/std/utility>
+
 #include <nv/target>
 
 CUB_NAMESPACE_BEGIN
@@ -120,7 +122,7 @@ class AgentSubWarpSort
       {
         return lhs < rhs;
       }
-      _CCCL_UNREACHABLE();
+      ::cuda::std::unreachable();
     }
 
 #if defined(__CUDA_FP16_TYPES_EXIST__)
@@ -135,7 +137,7 @@ class AgentSubWarpSort
       {
         NV_IF_TARGET(NV_PROVIDES_SM_53, (return __hlt(lhs, rhs);), (return __half2float(lhs) < __half2float(rhs);));
       }
-      _CCCL_UNREACHABLE();
+      ::cuda::std::unreachable();
     }
 #endif // __CUDA_FP16_TYPES_EXIST__
   };

cub/cub/thread/thread_reduce.cuh

@@ -53,6 +53,7 @@
 #include <cuda/std/cassert> // assert
 #include <cuda/std/cstdint> // uint16_t
 #include <cuda/std/functional> // cuda::std::plus
+#include <cuda/std/utility> // unreachable
 
 CUB_NAMESPACE_BEGIN
 
@@ -522,7 +523,7 @@ ThreadReduceSimd(const Input& input, ReductionOp reduction_op) -> ::cuda::std::r
     }
     return unsafe_bitcast<UnpackedType>(result)[0];
   }
-  _CCCL_UNREACHABLE(); // nvcc 11.x warning workaround (never reached)
+  ::cuda::std::unreachable(); // nvcc 11.x warning workaround (never reached)
 }
 
 } // namespace internal

cub/cub/util_device.cuh

@@ -730,7 +730,7 @@ private:
   template <int, typename FunctorT>
   CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE static cudaError_t invoke_static(FunctorT&, ::cuda::std::false_type)
   {
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 
   template <typename FunctorT>
@@ -742,7 +742,7 @@ private:
   template <typename FunctorT>
   CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE static cudaError_t do_invoke(FunctorT&, ::cuda::std::false_type)
   {
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 };
 
@@ -773,7 +773,7 @@ private:
   template <int, typename FunctorT>
   CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE static cudaError_t invoke_static(FunctorT&, ::cuda::std::false_type)
   {
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 };
 

cub/test/catch2_test_device_histogram.cu

@@ -33,6 +33,7 @@
 #include <cuda/std/array>
 #include <cuda/std/bit>
 #include <cuda/std/type_traits>
+#include <cuda/std/utility>
 
 #include <algorithm>
 #include <limits>
@@ -305,7 +306,7 @@ void test_even_and_range(LevelT max_level, int max_level_count, OffsetT width, O
       {
         return static_cast<int>((sample - min) * fp_scales[channel]);
       }
-      _CCCL_UNREACHABLE();
+      ::cuda::std::unreachable();
     };
     auto h_histogram = compute_reference_result<Channels, CounterT>(
       h_samples, sample_to_bin_index, num_levels, width, height, row_pitch);

cudax/include/cuda/experimental/__hierarchy/hierarchy_dimensions.cuh

@@ -17,6 +17,7 @@
 #include <cuda/std/__type_traits/type_list.h>
 #include <cuda/std/__utility/declval.h>
 #include <cuda/std/__utility/integer_sequence.h>
+#include <cuda/std/__utility/unreachable.h>
 #include <cuda/std/span>
 #include <cuda/std/tuple>
 
@@ -79,7 +80,7 @@ _CCCL_NODISCARD _CUDAX_API constexpr auto fool_compiler(const dimensions<T, Exte
   {
     return ex;
   }
-  _CCCL_UNREACHABLE();
+  ::cuda::std::unreachable();
 }
 
 template <typename QueryLevel, typename Hierarchy>
@@ -118,7 +119,7 @@ struct get_level_helper
     {
       return (*this)(levels...);
     }
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 };
 } // namespace detail

cudax/include/cuda/experimental/__hierarchy/hierarchy_levels.cuh

@@ -12,6 +12,7 @@
 #define _CUDAX__HIERARCHY_HIERARCHY_LEVELS
 
 #include <cuda/std/__type_traits/type_list.h>
+#include <cuda/std/__utility/unreachable.h>
 
 #include <cuda/experimental/__hierarchy/dimensions.cuh>
 
@@ -285,7 +286,7 @@ template <typename Unit, typename Level>
     return dims_product<typename Level::product_type>(
       extents_impl<SplitLevel, Level>(), extents_impl<Unit, SplitLevel>());
   }
-  _CCCL_UNREACHABLE();
+  ::cuda::std::unreachable();
 }
 
 template <typename Unit, typename Level>
@@ -302,7 +303,7 @@ template <typename Unit, typename Level>
       dims_product<typename Level::product_type>(index_impl<SplitLevel, Level>(), extents_impl<Unit, SplitLevel>()),
       index_impl<Unit, SplitLevel>());
   }
-  _CCCL_UNREACHABLE();
+  ::cuda::std::unreachable();
 }
 } // namespace detail
 

libcudacxx/include/cuda/__cccl_config

@@ -25,7 +25,6 @@
 #include <cuda/std/__cccl/rtti.h> // IWYU pragma: export
 #include <cuda/std/__cccl/sequence_access.h> // IWYU pragma: export
 #include <cuda/std/__cccl/system_header.h> // IWYU pragma: export
-#include <cuda/std/__cccl/unreachable.h> // IWYU pragma: export
 #include <cuda/std/__cccl/version.h> // IWYU pragma: export
 #include <cuda/std/__cccl/visibility.h> // IWYU pragma: export
 

libcudacxx/include/cuda/__memcpy_async/memcpy_async_tx.h

@@ -34,6 +34,7 @@
 #    include <cuda/__ptx/ptx_helper_functions.h>
 #    include <cuda/std/__atomic/scopes.h>
 #    include <cuda/std/__type_traits/is_trivially_copyable.h>
+#    include <cuda/std/__utility/unreachable.h>
 #    include <cuda/std/cstdint>
 
 _LIBCUDACXX_BEGIN_NAMESPACE_CUDA_DEVICE
@@ -75,7 +76,7 @@ _CCCL_DEVICE inline async_contract_fulfillment memcpy_async_tx(
         // or from shared to remote cluster dsmem. To copy to remote
         // dsmem, we need to arrive on a cluster-scoped barrier, which
         // is not yet implemented. So we trap in this case as well.
-        _CCCL_UNREACHABLE();
+        ::cuda::std::unreachable();
       }),
     (__cuda_ptx_memcpy_async_tx_is_not_supported_before_SM_90__();));
 

libcudacxx/include/cuda/__memcpy_async/memcpy_completion.h

@@ -29,6 +29,7 @@
 #include <cuda/__memcpy_async/is_local_smem_barrier.h>
 #include <cuda/__memcpy_async/try_get_barrier_handle.h>
 #include <cuda/std/__atomic/scopes.h>
+#include <cuda/std/__utility/unreachable.h>
 #include <cuda/std/cstdint>
 
 #if defined(_CCCL_CUDA_COMPILER)
@@ -84,7 +85,7 @@ struct __memcpy_completion_impl
         // This completion mechanism should not be used with a shared
         // memory barrier. Or at least, we do not currently envision
         // bulk group to be used with shared memory barriers.
-        _CCCL_UNREACHABLE();
+        ::cuda::std::unreachable();
       case __completion_mechanism::__mbarrier_complete_tx:
 #if __cccl_ptx_isa >= 800
         // Pre-sm90, the mbarrier_complete_tx completion mechanism is not available.
@@ -100,7 +101,7 @@ struct __memcpy_completion_impl
         return async_contract_fulfillment::none;
       default:
         // Get rid of "control reaches end of non-void function":
-        _CCCL_UNREACHABLE();
+        ::cuda::std::unreachable();
     }
   }
 
@@ -129,16 +130,16 @@ struct __memcpy_completion_impl
         return async_contract_fulfillment::async;
       case __completion_mechanism::__mbarrier_complete_tx:
         // Non-smem barriers do not have an mbarrier_complete_tx mechanism..
-        _CCCL_UNREACHABLE();
+        ::cuda::std::unreachable();
       case __completion_mechanism::__async_bulk_group:
         // This completion mechanism is currently not expected to be used with barriers.
-        _CCCL_UNREACHABLE();
+        ::cuda::std::unreachable();
       case __completion_mechanism::__sync:
         // sync: In this case, we do not need to do anything.
         return async_contract_fulfillment::none;
       default:
         // Get rid of "control reaches end of non-void function":
-        _CCCL_UNREACHABLE();
+        ::cuda::std::unreachable();
     }
   }
 
@@ -158,7 +159,7 @@ struct __memcpy_completion_impl
         return async_contract_fulfillment::none;
       default:
         // Get rid of "control reaches end of non-void function":
-        _CCCL_UNREACHABLE();
+        ::cuda::std::unreachable();
     }
   }
 };

libcudacxx/include/cuda/std/__exception/terminate.h

@@ -22,6 +22,7 @@
 #  pragma system_header
 #endif // no system header
 
+#include <cuda/std/__utility/unreachable.h>
 #include <cuda/std/cstdlib> // ::exit
 
 _CCCL_DIAG_PUSH
@@ -32,7 +33,7 @@ _LIBCUDACXX_BEGIN_NAMESPACE_STD_NOVERSION // purposefully not using versioning n
 _CCCL_NORETURN _LIBCUDACXX_HIDE_FROM_ABI void __cccl_terminate() noexcept
 {
   NV_IF_ELSE_TARGET(NV_IS_HOST, (::exit(-1);), (__trap();))
-  _CCCL_UNREACHABLE();
+  ::cuda::std::unreachable();
 }
 
 #if 0 // Expose once atomic is universally available
@@ -59,7 +60,7 @@ _LIBCUDACXX_HIDE_FROM_ABI  terminate_handler get_terminate() noexcept
 _CCCL_NORETURN _LIBCUDACXX_HIDE_FROM_ABI void terminate() noexcept
 {
   __cccl_terminate();
-  _CCCL_UNREACHABLE();
+  ::cuda::std::unreachable();
 }
 
 _LIBCUDACXX_END_NAMESPACE_STD_NOVERSION

libcudacxx/include/cuda/std/__iterator/advance.h

@@ -28,6 +28,7 @@
 #include <cuda/std/__iterator/iterator_traits.h>
 #include <cuda/std/__utility/convert_to_integral.h>
 #include <cuda/std/__utility/move.h>
+#include <cuda/std/__utility/unreachable.h>
 
 _LIBCUDACXX_BEGIN_NAMESPACE_STD
 
@@ -226,7 +227,7 @@ struct __fn
       }
       return __n;
     }
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 };
 _LIBCUDACXX_END_NAMESPACE_CPO

libcudacxx/include/cuda/std/__iterator/distance.h

@@ -29,6 +29,7 @@
 #include <cuda/std/__ranges/size.h>
 #include <cuda/std/__type_traits/decay.h>
 #include <cuda/std/__type_traits/remove_cvref.h>
+#include <cuda/std/__utility/unreachable.h>
 
 _LIBCUDACXX_BEGIN_NAMESPACE_STD
 
@@ -92,7 +93,7 @@ struct __fn
     {
       return __last - decay_t<_Ip>(__first);
     }
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 
   _LIBCUDACXX_TEMPLATE(class _Rp)
@@ -107,7 +108,7 @@ struct __fn
     {
       return operator()(_CUDA_VRANGES::begin(__r), _CUDA_VRANGES::end(__r));
     }
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 };
 _LIBCUDACXX_END_NAMESPACE_CPO

libcudacxx/include/cuda/std/__iterator/move_iterator.h

@@ -41,6 +41,7 @@
 #include <cuda/std/__type_traits/is_reference.h>
 #include <cuda/std/__type_traits/remove_reference.h>
 #include <cuda/std/__utility/move.h>
+#include <cuda/std/__utility/unreachable.h>
 
 _LIBCUDACXX_BEGIN_NAMESPACE_STD
 
@@ -126,7 +127,7 @@ class _CCCL_TYPE_VISIBILITY_DEFAULT move_iterator
     {
       return input_iterator_tag{};
     }
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 #  endif // !_CCCL_COMPILER_MSVC_2017
 #endif // _CCCL_STD_VER >= 2017

libcudacxx/include/cuda/std/__ranges/size.h

@@ -32,6 +32,7 @@
 #include <cuda/std/__type_traits/remove_cvref.h>
 #include <cuda/std/__utility/auto_cast.h>
 #include <cuda/std/__utility/declval.h>
+#include <cuda/std/__utility/unreachable.h>
 #include <cuda/std/cstddef>
 
 _LIBCUDACXX_BEGIN_NAMESPACE_RANGES
@@ -191,7 +192,7 @@ struct __fn
     {
       return static_cast<_Signed>(_CUDA_VRANGES::size(__t));
     }
-    _CCCL_UNREACHABLE();
+    ::cuda::std::unreachable();
   }
 };
 _LIBCUDACXX_END_NAMESPACE_CPO