diff --git a/libcudacxx/cmake/LibcudacxxPublicHeaderTestingHost.cmake b/libcudacxx/cmake/LibcudacxxPublicHeaderTestingHost.cmake
index 26e3c1fe1609..d37367332d54 100644
--- a/libcudacxx/cmake/LibcudacxxPublicHeaderTestingHost.cmake
+++ b/libcudacxx/cmake/LibcudacxxPublicHeaderTestingHost.cmake
@@ -6,6 +6,7 @@
 
 # Meta target for all configs' header builds:
 add_custom_target(libcudacxx.test.public_headers_host_only)
+add_custom_target(libcudacxx.test.public_headers_host_only_with_ctk)
 
 # Grep all public headers
 file(
@@ -13,18 +14,8 @@ file(
   LIST_DIRECTORIES false
   RELATIVE "${libcudacxx_SOURCE_DIR}/include"
   CONFIGURE_DEPENDS
+  "${libcudacxx_SOURCE_DIR}/include/cuda/*"
   "${libcudacxx_SOURCE_DIR}/include/cuda/std/*"
-  # Add some files we expect to work in host only compilation
-  "${libcudacxx_SOURCE_DIR}/include/cuda/bit"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/cmath"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/functional"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/iterator"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/mdspan"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/memory"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/numeric"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/type_traits"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/utility"
-  "${libcudacxx_SOURCE_DIR}/include/cuda/version"
 )
 
 set(public_host_header_cxx_compile_options)
@@ -63,6 +54,38 @@ function(libcudacxx_create_public_header_test_host header_name headertest_src)
   )
 endfunction()
 
+function(
+  libcudacxx_create_public_header_test_host_with_ctk
+  header_name
+  headertest_src
+)
+  # Create the default target for that file
+  add_library(
+    public_headers_host_only_with_ctk_${header_name}
+    SHARED
+    "${headertest_src}.cpp"
+  )
+  cccl_configure_target(public_headers_host_only_with_ctk_${header_name})
+  target_compile_definitions(
+    public_headers_host_only_with_ctk_${header_name}
+    PRIVATE #
+      ${public_host_header_cxx_compile_definitions}
+      _CCCL_HEADER_TEST
+  )
+  target_compile_options(
+    public_headers_host_only_with_ctk_${header_name}
+    PRIVATE ${public_host_header_cxx_compile_options}
+  )
+  target_link_libraries(
+    public_headers_host_only_with_ctk_${header_name}
+    PUBLIC libcudacxx.compiler_interface CUDA::cudart
+  )
+  add_dependencies(
+    libcudacxx.test.public_headers_host_only_with_ctk
+    public_headers_host_only_with_ctk_${header_name}
+  )
+endfunction()
+
 function(libcudacxx_add_public_headers_host_only header)
   # ${header} contains the "/" from the subfolder, replace by "_" for actual names
   string(REPLACE "/" "_" header_name "${header}")
@@ -76,6 +99,7 @@ function(libcudacxx_add_public_headers_host_only header)
 
   # Create the default target for that file
   libcudacxx_create_public_header_test_host(${header_name} ${headertest_src})
+  libcudacxx_create_public_header_test_host_with_ctk(${header_name} ${headertest_src})
 endfunction()
 
 foreach (header IN LISTS public_headers_host_only)
diff --git a/libcudacxx/include/cuda/__container/buffer.h b/libcudacxx/include/cuda/__container/buffer.h
index 2f1f50e3e796..9b91010f3fa1 100644
--- a/libcudacxx/include/cuda/__container/buffer.h
+++ b/libcudacxx/include/cuda/__container/buffer.h
@@ -21,35 +21,37 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_HAS_CUDA_COMPILER()
-#  include <cub/device/device_transform.cuh>
-#endif // _CCCL_HAS_CUDA_COMPILER()
-
-#include <cuda/__container/heterogeneous_iterator.h>
-#include <cuda/__container/uninitialized_async_buffer.h>
-#include <cuda/__launch/host_launch.h>
-#include <cuda/__memory_resource/any_resource.h>
-#include <cuda/__memory_resource/get_memory_resource.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/synchronous_resource_adapter.h>
-#include <cuda/__runtime/ensure_current_context.h>
-#include <cuda/__stream/get_stream.h>
-#include <cuda/std/__execution/env.h>
-#include <cuda/std/__iterator/concepts.h>
-#include <cuda/std/__iterator/distance.h>
-#include <cuda/std/__iterator/reverse_iterator.h>
-#include <cuda/std/__memory/uninitialized_algorithms.h>
-#include <cuda/std/__ranges/access.h>
-#include <cuda/std/__ranges/concepts.h>
-#include <cuda/std/__ranges/size.h>
-#include <cuda/std/__ranges/unwrap_end.h>
-#include <cuda/std/__type_traits/is_trivially_copyable.h>
-#include <cuda/std/__utility/forward.h>
-#include <cuda/std/__utility/move.h>
-#include <cuda/std/cstdint>
-#include <cuda/std/initializer_list>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  if _CCCL_HAS_CUDA_COMPILER()
+#    include <cub/device/device_transform.cuh>
+#  endif // _CCCL_HAS_CUDA_COMPILER()
+
+#  include <cuda/__container/heterogeneous_iterator.h>
+#  include <cuda/__container/uninitialized_async_buffer.h>
+#  include <cuda/__launch/host_launch.h>
+#  include <cuda/__memory_resource/any_resource.h>
+#  include <cuda/__memory_resource/get_memory_resource.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__memory_resource/synchronous_resource_adapter.h>
+#  include <cuda/__runtime/ensure_current_context.h>
+#  include <cuda/__stream/get_stream.h>
+#  include <cuda/std/__execution/env.h>
+#  include <cuda/std/__iterator/concepts.h>
+#  include <cuda/std/__iterator/distance.h>
+#  include <cuda/std/__iterator/reverse_iterator.h>
+#  include <cuda/std/__memory/uninitialized_algorithms.h>
+#  include <cuda/std/__ranges/access.h>
+#  include <cuda/std/__ranges/concepts.h>
+#  include <cuda/std/__ranges/size.h>
+#  include <cuda/std/__ranges/unwrap_end.h>
+#  include <cuda/std/__type_traits/is_trivially_copyable.h>
+#  include <cuda/std/__utility/forward.h>
+#  include <cuda/std/__utility/move.h>
+#  include <cuda/std/cstdint>
+#  include <cuda/std/initializer_list>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 //! @file The \c buffer class provides a container of contiguous memory
 _CCCL_BEGIN_NAMESPACE_CUDA
@@ -309,7 +311,7 @@ class buffer
       __buf_.size());
   }
 
-#ifndef _CCCL_DOXYGEN_INVOKED // doxygen conflates the overloads
+#  ifndef _CCCL_DOXYGEN_INVOKED // doxygen conflates the overloads
   _CCCL_TEMPLATE(class _Range, class _Resource, class _Env = ::cuda::std::execution::env<>)
   _CCCL_REQUIRES(
     ::cuda::mr::synchronous_resource<::cuda::std::decay_t<_Resource>> _CCCL_AND __compatible_range<_Range>
@@ -332,7 +334,7 @@ class buffer
       __unwrapped_begin(),
       __buf_.size());
   }
-#endif // _CCCL_DOXYGEN_INVOKED
+#  endif // _CCCL_DOXYGEN_INVOKED
   //! @}
 
   //! @addtogroup iterators
@@ -447,7 +449,7 @@ class buffer
     return __buf_.data();
   }
 
-#ifndef _CCCL_DOXYGEN_INVOKED
+#  ifndef _CCCL_DOXYGEN_INVOKED
   //! @brief Returns a pointer to the first element of the buffer. If the buffer
   //! is empty, the returned pointer will be null.
   [[nodiscard]] _CCCL_HIDE_FROM_ABI pointer __unwrapped_begin() noexcept
@@ -477,7 +479,7 @@ class buffer
   {
     return __buf_.data() + __buf_.size();
   }
-#endif // _CCCL_DOXYGEN_INVOKED
+#  endif // _CCCL_DOXYGEN_INVOKED
 
   //! @}
 
@@ -683,14 +685,14 @@ __fill_n(cuda::stream_ref __stream, _Tp* __first, ::cuda::std::size_t __count, c
     }
     else
     {
-#if _CCCL_HAS_CUDA_COMPILER()
+#  if _CCCL_HAS_CUDA_COMPILER()
       ::cuda::__ensure_current_context __guard(__stream);
       ::cub::DeviceTransform::Fill(__first, __count, __value, __stream.get());
-#else
+#  else
       static_assert(0,
                     "CUDA compiler is required to initialize a buffer with "
                     "elements larger than 4 bytes");
-#endif
+#  endif
     }
   }
 }
@@ -887,6 +889,8 @@ auto make_buffer(stream_ref __stream, _Resource&& __mr, _Range&& __range, const
 }
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___CONTAINER_BUFFER_H
diff --git a/libcudacxx/include/cuda/__container/heterogeneous_iterator.h b/libcudacxx/include/cuda/__container/heterogeneous_iterator.h
index e33c40f20165..9a62fc277fbe 100644
--- a/libcudacxx/include/cuda/__container/heterogeneous_iterator.h
+++ b/libcudacxx/include/cuda/__container/heterogeneous_iterator.h
@@ -21,17 +21,19 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/std/__iterator/iterator_traits.h>
-#include <cuda/std/__memory/addressof.h>
-#include <cuda/std/__memory/pointer_traits.h>
-#include <cuda/std/__type_traits/is_const.h>
-#include <cuda/std/__type_traits/is_same.h>
-#include <cuda/std/__type_traits/maybe_const.h>
-#include <cuda/std/__type_traits/remove_const.h>
-#include <cuda/std/cstdint>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/std/__iterator/iterator_traits.h>
+#  include <cuda/std/__memory/addressof.h>
+#  include <cuda/std/__memory/pointer_traits.h>
+#  include <cuda/std/__type_traits/is_const.h>
+#  include <cuda/std/__type_traits/is_same.h>
+#  include <cuda/std/__type_traits/maybe_const.h>
+#  include <cuda/std/__type_traits/remove_const.h>
+#  include <cuda/std/cstdint>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 //! @file The \c heterogeneous_iterator class is an iterator that provides typed execution space safety.
 _CCCL_BEGIN_NAMESPACE_CUDA
@@ -290,7 +292,7 @@ class heterogeneous_iterator
     return __temp;
   }
 
-#ifndef _CCCL_DOXYGEN_INVOKED // Do not document
+#  ifndef _CCCL_DOXYGEN_INVOKED // Do not document
   //! @brief Advance a \c heterogeneous_iterator
   //! @param __count The number of elements to advance.
   //! @param __other A heterogeneous_iterator.
@@ -301,7 +303,7 @@ class heterogeneous_iterator
     __other += __count;
     return __other;
   }
-#endif // _CCCL_DOXYGEN_INVOKED
+#  endif // _CCCL_DOXYGEN_INVOKED
 
   //! @brief Advance a \c heterogeneous_iterator by the negative value of \p __count
   //! @param __count The number of elements to advance.
@@ -330,7 +332,7 @@ class heterogeneous_iterator
     return static_cast<difference_type>(this->__ptr_ - __other.__ptr_);
   }
 
-#ifndef _CCCL_DOXYGEN_INVOKED // Do not document
+#  ifndef _CCCL_DOXYGEN_INVOKED // Do not document
   //! @brief Equality comparison between two heterogeneous_iterator
   //! @param __lhs A heterogeneous_iterator.
   //! @param __rhs Another heterogeneous_iterator.
@@ -340,7 +342,7 @@ class heterogeneous_iterator
   {
     return __lhs.__ptr_ == __rhs.__ptr_;
   }
-#  if _CCCL_STD_VER <= 2017
+#    if _CCCL_STD_VER <= 2017
   //! @brief Inequality comparison between two heterogeneous_iterator
   //! @param __lhs A heterogeneous_iterator.
   //! @param __rhs Another heterogeneous_iterator.
@@ -350,15 +352,15 @@ class heterogeneous_iterator
   {
     return __lhs.__ptr_ != __rhs.__ptr_;
   }
-#  endif // _CCCL_STD_VER <= 2017
+#    endif // _CCCL_STD_VER <= 2017
 
-#  if _LIBCUDACXX_HAS_SPACESHIP_OPERATOR()
+#    if _LIBCUDACXX_HAS_SPACESHIP_OPERATOR()
   [[nodiscard]] _CCCL_API friend constexpr ::cuda::std::strong_ordering
   operator<=>(const heterogeneous_iterator& __lhs, const heterogeneous_iterator& __rhs) noexcept
   {
     return __lhs.__ptr_ <=> __rhs.__ptr_;
   }
-#  else // ^^^ _LIBCUDACXX_HAS_SPACESHIP_OPERATOR() ^^^ /  vvv !_LIBCUDACXX_HAS_SPACESHIP_OPERATOR() vvv
+#    else // ^^^ _LIBCUDACXX_HAS_SPACESHIP_OPERATOR() ^^^ /  vvv !_LIBCUDACXX_HAS_SPACESHIP_OPERATOR() vvv
   //! @brief Less than relation between two heterogeneous_iterator
   //! @param __lhs A heterogeneous_iterator.
   //! @param __rhs Another heterogeneous_iterator.
@@ -399,8 +401,8 @@ class heterogeneous_iterator
   {
     return __lhs.__ptr_ >= __rhs.__ptr_;
   }
-#  endif // !_LIBCUDACXX_HAS_SPACESHIP_OPERATOR()
-#endif // _CCCL_DOXYGEN_INVOKED
+#    endif // !_LIBCUDACXX_HAS_SPACESHIP_OPERATOR()
+#  endif // _CCCL_DOXYGEN_INVOKED
 
   _CCCL_API constexpr pointer __unwrap() const noexcept
   {
@@ -431,6 +433,8 @@ struct pointer_traits<::cuda::heterogeneous_iterator<_Tp, _Properties...>>
 
 _CCCL_END_NAMESPACE_CUDA_STD
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //__CUDAX__CONTAINERS_HETEROGENEOUS_ITERATOR_CUH
diff --git a/libcudacxx/include/cuda/__container/uninitialized_async_buffer.h b/libcudacxx/include/cuda/__container/uninitialized_async_buffer.h
index d897bb7d57c3..c9bcb8ab1e0d 100644
--- a/libcudacxx/include/cuda/__container/uninitialized_async_buffer.h
+++ b/libcudacxx/include/cuda/__container/uninitialized_async_buffer.h
@@ -21,19 +21,21 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__memory_resource/any_resource.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__stream/stream_ref.h>
-#include <cuda/std/__memory/addressof.h>
-#include <cuda/std/__memory/align.h>
-#include <cuda/std/__new/launder.h>
-#include <cuda/std/__type_traits/type_set.h>
-#include <cuda/std/__utility/exchange.h>
-#include <cuda/std/__utility/move.h>
-#include <cuda/std/__utility/swap.h>
-#include <cuda/std/span>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__memory_resource/any_resource.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__stream/stream_ref.h>
+#  include <cuda/std/__memory/addressof.h>
+#  include <cuda/std/__memory/align.h>
+#  include <cuda/std/__new/launder.h>
+#  include <cuda/std/__type_traits/type_set.h>
+#  include <cuda/std/__utility/exchange.h>
+#  include <cuda/std/__utility/move.h>
+#  include <cuda/std/__utility/swap.h>
+#  include <cuda/std/span>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 //! @file
 //! The \c __uninitialized_async_buffer class provides a typed buffer allocated
@@ -138,7 +140,7 @@ class __uninitialized_async_buffer
     return {__self.__get_data(), __self.size()};
   }
 
-#ifndef _CCCL_DOXYGEN_INVOKED
+#  ifndef _CCCL_DOXYGEN_INVOKED
   // This is needed to ensure that we do not do a deep copy in
   // __replace_allocation
   struct __fake_resource_ref
@@ -179,7 +181,7 @@ class __uninitialized_async_buffer
     _CCCL_REQUIRES(::cuda::std::__is_included_in_v<_Property, _Properties...>)
     _CCCL_HIDE_FROM_ABI friend constexpr void get_property(const __fake_resource_ref&, _Property) noexcept {}
   };
-#endif // _CCCL_DOXYGEN_INVOKED
+#  endif // _CCCL_DOXYGEN_INVOKED
 
 public:
   using value_type      = _Tp;
@@ -411,6 +413,8 @@ template <class _Tp>
 using uninitialized_async_device_buffer = __uninitialized_async_buffer<_Tp, ::cuda::mr::device_accessible>;
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //__CUDAX__CONTAINERS_UNINITIALIZED_ASYNC_BUFFER_H
diff --git a/libcudacxx/include/cuda/__device/arch_traits.h b/libcudacxx/include/cuda/__device/arch_traits.h
index f406c3782ca1..ef410120a281 100644
--- a/libcudacxx/include/cuda/__device/arch_traits.h
+++ b/libcudacxx/include/cuda/__device/arch_traits.h
@@ -21,17 +21,15 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_HAS_CTK()
-
-#  include <cuda/__device/arch_id.h>
-#  include <cuda/__device/compute_capability.h>
-#  include <cuda/__fwd/devices.h>
-#  include <cuda/std/__exception/cuda_error.h>
-#  include <cuda/std/__type_traits/always_false.h>
-#  include <cuda/std/cstdint>
-#  include <cuda/std/limits>
+#include <cuda/__device/arch_id.h>
+#include <cuda/__device/compute_capability.h>
+#include <cuda/__fwd/devices.h>
+#include <cuda/std/__exception/cuda_error.h>
+#include <cuda/std/__type_traits/always_false.h>
+#include <cuda/std/cstdint>
+#include <cuda/std/limits>
 
-#  include <cuda/std/__cccl/prologue.h>
+#include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -507,7 +505,11 @@ template <>
     case arch_id::sm_121a:
       return ::cuda::arch_traits<arch_id::sm_121a>();
     default:
+#if _CCCL_HAS_CTK()
       ::cuda::__throw_cuda_error(::cudaErrorInvalidValue, "Traits requested for an unknown architecture");
+#else // ^^^ _CCCL_HAS_CTK() ^^^ / vvv !_CCCL_HAS_CTK() vvv
+      ::cuda::__throw_cuda_error(/*cudaErrorInvalidValue*/ 1, "Traits requested for an unknown architecture");
+#endif // ^^^ !_CCCL_HAS_CTK() ^^^
       break;
   }
 }
@@ -522,7 +524,7 @@ template <>
 
 _CCCL_END_NAMESPACE_CUDA
 
-#  if _CCCL_CUDA_COMPILATION()
+#if _CCCL_CUDA_COMPILATION()
 
 _CCCL_BEGIN_NAMESPACE_CUDA_DEVICE
 
@@ -535,19 +537,17 @@ _CCCL_BEGIN_NAMESPACE_CUDA_DEVICE
 template <class _Dummy = void>
 [[nodiscard]] _CCCL_DEVICE_API inline _CCCL_TARGET_CONSTEXPR ::cuda::arch_traits_t current_arch_traits() noexcept
 {
-#    if _CCCL_DEVICE_COMPILATION()
+#  if _CCCL_DEVICE_COMPILATION()
   return ::cuda::arch_traits_for(::cuda::device::current_arch_id<_Dummy>());
-#    else // ^^^ _CCCL_DEVICE_COMPILATION() ^^^ / vvv !_CCCL_DEVICE_COMPILATION() vvv
+#  else // ^^^ _CCCL_DEVICE_COMPILATION() ^^^ / vvv !_CCCL_DEVICE_COMPILATION() vvv
   return {};
-#    endif // ^^^ !_CCCL_DEVICE_COMPILATION() ^^^
+#  endif // ^^^ !_CCCL_DEVICE_COMPILATION() ^^^
 }
 
 _CCCL_END_NAMESPACE_CUDA_DEVICE
 
-#  endif // _CCCL_CUDA_COMPILATION
-
-#  include <cuda/std/__cccl/epilogue.h>
+#endif // _CCCL_CUDA_COMPILATION
 
-#endif // _CCCL_HAS_CTK()
+#include <cuda/std/__cccl/epilogue.h>
 
 #endif // _CUDA___DEVICE_ARCH_TRAITS_H
diff --git a/libcudacxx/include/cuda/__event/timed_event.h b/libcudacxx/include/cuda/__event/timed_event.h
index 603c140fb7c1..a313f5bacffa 100644
--- a/libcudacxx/include/cuda/__event/timed_event.h
+++ b/libcudacxx/include/cuda/__event/timed_event.h
@@ -11,9 +11,6 @@
 #ifndef _CUDA___EVENT_TIMED_EVENT_H
 #define _CUDA___EVENT_TIMED_EVENT_H
 
-#include <cuda_runtime_api.h>
-// cuda_runtime_api needs to come first
-
 #include <cuda/std/detail/__config>
 
 #if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
diff --git a/libcudacxx/include/cuda/__functional/for_each_canceled.h b/libcudacxx/include/cuda/__functional/for_each_canceled.h
index 54a97ad0290b..b230911e5e8c 100644
--- a/libcudacxx/include/cuda/__functional/for_each_canceled.h
+++ b/libcudacxx/include/cuda/__functional/for_each_canceled.h
@@ -21,14 +21,14 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/std/__functional/invoke.h>
-#include <cuda/std/__utility/move.h>
-#include <cuda/std/__utility/unreachable.h>
-#include <cuda/std/cstdint>
+#if _CCCL_CUDA_COMPILATION()
 
-#include <nv/target>
+#  include <cuda/std/__functional/invoke.h>
+#  include <cuda/std/__utility/move.h>
+#  include <cuda/std/__utility/unreachable.h>
+#  include <cuda/std/cstdint>
 
-#if _CCCL_CUDA_COMPILATION()
+#  include <nv/target>
 
 #  include <cuda/std/__cccl/prologue.h>
 
diff --git a/libcudacxx/include/cuda/__fwd/devices.h b/libcudacxx/include/cuda/__fwd/devices.h
index 5e77301b2e6a..3158b66ec2e1 100644
--- a/libcudacxx/include/cuda/__fwd/devices.h
+++ b/libcudacxx/include/cuda/__fwd/devices.h
@@ -27,10 +27,13 @@
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
+#if _CCCL_HAS_CTK()
 class __physical_device;
 class device_ref;
 template <::cudaDeviceAttr _Attr>
 struct __dev_attr;
+#endif // _CCCL_HAS_CTK()
+
 struct arch_traits_t;
 class compute_capability;
 enum class arch_id : int;
diff --git a/libcudacxx/include/cuda/__hierarchy/block_level.h b/libcudacxx/include/cuda/__hierarchy/block_level.h
index 37257ce9a1eb..dc746527bf6a 100644
--- a/libcudacxx/include/cuda/__hierarchy/block_level.h
+++ b/libcudacxx/include/cuda/__hierarchy/block_level.h
@@ -21,14 +21,16 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/hierarchy_query_result.h>
-#include <cuda/__hierarchy/native_hierarchy_level_base.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__mdspan/extents.h>
-#include <cuda/std/__type_traits/is_integer.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/hierarchy_query_result.h>
+#  include <cuda/__hierarchy/native_hierarchy_level_base.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__mdspan/extents.h>
+#  include <cuda/std/__type_traits/is_integer.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -43,6 +45,8 @@ struct block_level : __native_hierarchy_level_base<block_level>
   using __base_type = __native_hierarchy_level_base<block_level>;
   using __base_type::count_as;
   using __base_type::extents_as;
+
+#  if _CCCL_CUDA_COMPILATION()
   using __base_type::index_as;
   using __base_type::rank_as;
 
@@ -132,12 +136,15 @@ struct block_level : __native_hierarchy_level_base<block_level>
     const auto __idx  = index_as<_Tp>(__level, __hier_unpacked);
     return static_cast<_Tp>((__idx.z * __dims.y + __idx.y) * __dims.x + __idx.x);
   }
+#  endif // _CCCL_CUDA_COMPILATION()
 };
 
 _CCCL_GLOBAL_CONSTANT block_level block;
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_BLOCK_LEVEL_H
diff --git a/libcudacxx/include/cuda/__hierarchy/cluster_level.h b/libcudacxx/include/cuda/__hierarchy/cluster_level.h
index 70ef8fdce89c..72a40127fbb4 100644
--- a/libcudacxx/include/cuda/__hierarchy/cluster_level.h
+++ b/libcudacxx/include/cuda/__hierarchy/cluster_level.h
@@ -21,14 +21,16 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/hierarchy_query_result.h>
-#include <cuda/__hierarchy/native_hierarchy_level_base.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__mdspan/extents.h>
-#include <cuda/std/__type_traits/is_integer.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/hierarchy_query_result.h>
+#  include <cuda/__hierarchy/native_hierarchy_level_base.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__mdspan/extents.h>
+#  include <cuda/std/__type_traits/is_integer.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -42,6 +44,8 @@ struct cluster_level : __native_hierarchy_level_base<cluster_level>
 
   using __base_type = __native_hierarchy_level_base<cluster_level>;
   using __base_type::extents_as;
+
+#  if _CCCL_CUDA_COMPILATION()
   using __base_type::index_as;
 
   // interactions with grid level
@@ -63,12 +67,15 @@ struct cluster_level : __native_hierarchy_level_base<cluster_level>
     NV_IF_TARGET(NV_PROVIDES_SM_90, (__idx = ::__clusterIdx();))
     return {static_cast<_Tp>(__idx.x), static_cast<_Tp>(__idx.y), static_cast<_Tp>(__idx.z)};
   }
+#  endif // _CCCL_CUDA_COMPILATION()
 };
 
 _CCCL_GLOBAL_CONSTANT cluster_level cluster;
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_CLUSTER_LEVEL_H
diff --git a/libcudacxx/include/cuda/__hierarchy/dimensions.h b/libcudacxx/include/cuda/__hierarchy/dimensions.h
index 395c4c6e996e..97151555b330 100644
--- a/libcudacxx/include/cuda/__hierarchy/dimensions.h
+++ b/libcudacxx/include/cuda/__hierarchy/dimensions.h
@@ -21,10 +21,12 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/std/__mdspan/extents.h>
-#include <cuda/std/functional>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__mdspan/extents.h>
+#  include <cuda/std/functional>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -166,6 +168,8 @@ template <class _TyTrunc, class _Index, class _Dims>
 } // namespace __detail
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_DIMENSIONS_H
diff --git a/libcudacxx/include/cuda/__hierarchy/get_launch_dimensions.h b/libcudacxx/include/cuda/__hierarchy/get_launch_dimensions.h
index ee3d07ee8ae7..173c07413c15 100644
--- a/libcudacxx/include/cuda/__hierarchy/get_launch_dimensions.h
+++ b/libcudacxx/include/cuda/__hierarchy/get_launch_dimensions.h
@@ -21,14 +21,16 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__hierarchy/block_level.h>
-#include <cuda/__hierarchy/cluster_level.h>
-#include <cuda/__hierarchy/grid_level.h>
-#include <cuda/__hierarchy/thread_level.h>
-#include <cuda/__hierarchy/traits.h>
-#include <cuda/std/tuple>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__hierarchy/block_level.h>
+#  include <cuda/__hierarchy/cluster_level.h>
+#  include <cuda/__hierarchy/grid_level.h>
+#  include <cuda/__hierarchy/thread_level.h>
+#  include <cuda/__hierarchy/traits.h>
+#  include <cuda/std/tuple>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -85,6 +87,8 @@ constexpr auto _CCCL_HOST get_launch_dimensions(const hierarchy_dimensions<_Leve
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_GET_LAUNCH_DIMENSIONS_H
diff --git a/libcudacxx/include/cuda/__hierarchy/grid_level.h b/libcudacxx/include/cuda/__hierarchy/grid_level.h
index 9cd6a111709d..3ab829798028 100644
--- a/libcudacxx/include/cuda/__hierarchy/grid_level.h
+++ b/libcudacxx/include/cuda/__hierarchy/grid_level.h
@@ -21,10 +21,12 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/native_hierarchy_level_base.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/native_hierarchy_level_base.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -39,6 +41,8 @@ _CCCL_GLOBAL_CONSTANT grid_level grid;
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_GRID_LEVEL_H
diff --git a/libcudacxx/include/cuda/__hierarchy/hierarchy_dimensions.h b/libcudacxx/include/cuda/__hierarchy/hierarchy_dimensions.h
index 816c5553f26c..6bf0f9894936 100644
--- a/libcudacxx/include/cuda/__hierarchy/hierarchy_dimensions.h
+++ b/libcudacxx/include/cuda/__hierarchy/hierarchy_dimensions.h
@@ -21,22 +21,24 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/level_dimensions.h>
-#include <cuda/__hierarchy/traits.h>
-#include <cuda/std/__type_traits/fold.h>
-#include <cuda/std/__type_traits/is_same.h>
-#include <cuda/std/__type_traits/remove_cvref.h>
-#include <cuda/std/__type_traits/type_list.h>
-#include <cuda/std/__utility/declval.h>
-#include <cuda/std/__utility/integer_sequence.h>
-#include <cuda/std/array>
-#include <cuda/std/span>
-#include <cuda/std/tuple>
-
-#include <nv/target>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/level_dimensions.h>
+#  include <cuda/__hierarchy/traits.h>
+#  include <cuda/std/__type_traits/fold.h>
+#  include <cuda/std/__type_traits/is_same.h>
+#  include <cuda/std/__type_traits/remove_cvref.h>
+#  include <cuda/std/__type_traits/type_list.h>
+#  include <cuda/std/__utility/declval.h>
+#  include <cuda/std/__utility/integer_sequence.h>
+#  include <cuda/std/array>
+#  include <cuda/std/span>
+#  include <cuda/std/tuple>
+
+#  include <nv/target>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -256,7 +258,7 @@ struct __hierarchy_extents_helper
 };
 
 template <class _Tp, size_t... _Extents>
-[[nodiscard]] _CCCL_DEVICE constexpr auto __static_index_hint(const dimensions<_Tp, _Extents...>& __dims, ::dim3 __index)
+[[nodiscard]] _CCCL_DEVICE constexpr auto __static_index_hint(const dimensions<_Tp, _Extents...>&, ::dim3 __index)
 {
   using _HintedIndexT = dimensions<_Tp, (_Extents == 1 ? 0 : ::cuda::std::dynamic_extent)...>;
   return _HintedIndexT(__index.x, __index.y, __index.z);
@@ -378,10 +380,10 @@ struct hierarchy_dimensions
       : levels(__ls)
   {}
 
-#if !defined(_CCCL_NO_THREE_WAY_COMPARISON) && !_CCCL_COMPILER(MSVC, <, 19, 39) && !_CCCL_COMPILER(GCC, <, 12)
+#  if !defined(_CCCL_NO_THREE_WAY_COMPARISON) && !_CCCL_COMPILER(MSVC, <, 19, 39) && !_CCCL_COMPILER(GCC, <, 12)
   [[nodiscard]] _CCCL_HIDE_FROM_ABI constexpr bool operator==(const hierarchy_dimensions&) const noexcept = default;
-#else // ^^^ !_CCCL_NO_THREE_WAY_COMPARISON ^^^ / vvv
-      // _CCCL_NO_THREE_WAY_COMPARISON vvv
+#  else // ^^^ !_CCCL_NO_THREE_WAY_COMPARISON ^^^ / vvv
+        // _CCCL_NO_THREE_WAY_COMPARISON vvv
   [[nodiscard]] _CCCL_API friend constexpr bool
   operator==(const hierarchy_dimensions& __left, const hierarchy_dimensions& __right) noexcept
   {
@@ -393,7 +395,7 @@ struct hierarchy_dimensions
   {
     return __left.levels != __right.levels;
   }
-#endif // _CCCL_NO_THREE_WAY_COMPARISON
+#  endif // _CCCL_NO_THREE_WAY_COMPARISON
 
 private:
   // This being static is a bit of a hack to make extents_type working without
@@ -831,12 +833,12 @@ struct hierarchy_dimensions
     }
   }
 
-#ifndef _CCCL_DOXYGEN_INVOKED // Do not document
+#  ifndef _CCCL_DOXYGEN_INVOKED // Do not document
   constexpr hierarchy_dimensions combine([[maybe_unused]] __empty_hierarchy __empty) const
   {
     return *this;
   }
-#endif // _CCCL_DOXYGEN_INVOKED
+#  endif // _CCCL_DOXYGEN_INVOKED
 };
 
 // TODO consider having LUnit optional argument for template argument deduction
@@ -916,6 +918,8 @@ constexpr auto hierarchy_add_level(const hierarchy_dimensions<_Unit, _Levels...>
 }
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_HIERARCHY_DIMENSIONS_H
diff --git a/libcudacxx/include/cuda/__hierarchy/hierarchy_level_base.h b/libcudacxx/include/cuda/__hierarchy/hierarchy_level_base.h
index 6914f4882dee..b473e7826da6 100644
--- a/libcudacxx/include/cuda/__hierarchy/hierarchy_level_base.h
+++ b/libcudacxx/include/cuda/__hierarchy/hierarchy_level_base.h
@@ -21,19 +21,21 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/hierarchy_query_result.h>
-#include <cuda/__hierarchy/traits.h>
-#include <cuda/std/__algorithm/max.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__cstddef/types.h>
-#include <cuda/std/__functional/operations.h>
-#include <cuda/std/__mdspan/extents.h>
-#include <cuda/std/__type_traits/is_integer.h>
-#include <cuda/std/__utility/integer_sequence.h>
-#include <cuda/std/array>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/hierarchy_query_result.h>
+#  include <cuda/__hierarchy/traits.h>
+#  include <cuda/std/__algorithm/max.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__cstddef/types.h>
+#  include <cuda/std/__functional/operations.h>
+#  include <cuda/std/__mdspan/extents.h>
+#  include <cuda/std/__type_traits/is_integer.h>
+#  include <cuda/std/__utility/integer_sequence.h>
+#  include <cuda/std/array>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -171,7 +173,7 @@ struct hierarchy_level_base
       __level, ::cuda::__unpack_hierarchy_if_needed(__hier));
   }
 
-#if _CCCL_CUDA_COMPILATION()
+#  if _CCCL_CUDA_COMPILATION()
   _CCCL_TEMPLATE(class _InLevel, class _Hierarchy)
   _CCCL_REQUIRES(__is_hierarchy_level_v<_InLevel> _CCCL_AND __is_or_has_hierarchy_member_v<_Hierarchy>)
   [[nodiscard]] _CCCL_DEVICE_API static constexpr auto index(const _InLevel& __level, const _Hierarchy& __hier) noexcept
@@ -188,7 +190,7 @@ struct hierarchy_level_base
     return _Level::template rank_as<__default_1d_query_type<_InLevel>>(
       __level, ::cuda::__unpack_hierarchy_if_needed(__hier));
   }
-#endif // _CCCL_CUDA_COMPILATION()
+#  endif // _CCCL_CUDA_COMPILATION()
 
   _CCCL_TEMPLATE(class _Tp, class _InLevel, class _Hierarchy)
   _CCCL_REQUIRES(::cuda::std::__cccl_is_integer_v<_Tp> _CCCL_AND __is_hierarchy_level_v<_InLevel> _CCCL_AND
@@ -238,7 +240,7 @@ struct hierarchy_level_base
     return __count_as_impl<_Tp>(__level, ::cuda::__unpack_hierarchy_if_needed(__hier));
   }
 
-#if _CCCL_CUDA_COMPILATION()
+#  if _CCCL_CUDA_COMPILATION()
   _CCCL_TEMPLATE(class _Tp, class _InLevel, class _Hierarchy)
   _CCCL_REQUIRES(::cuda::std::__cccl_is_integer_v<_Tp> _CCCL_AND __is_hierarchy_level_v<_InLevel> _CCCL_AND
                    __is_or_has_hierarchy_member_v<_Hierarchy>)
@@ -310,7 +312,7 @@ struct hierarchy_level_base
     }
     return __ret;
   }
-#endif // _CCCL_CUDA_COMPILATION()
+#  endif // _CCCL_CUDA_COMPILATION()
 
 private:
   template <class>
@@ -361,6 +363,8 @@ struct hierarchy_level_base
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_HIERARCHY_LEVEL_BASE_H
diff --git a/libcudacxx/include/cuda/__hierarchy/hierarchy_levels.h b/libcudacxx/include/cuda/__hierarchy/hierarchy_levels.h
index fe32c5e69bc8..2569f0403be9 100644
--- a/libcudacxx/include/cuda/__hierarchy/hierarchy_levels.h
+++ b/libcudacxx/include/cuda/__hierarchy/hierarchy_levels.h
@@ -21,13 +21,15 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/dimensions.h>
-#include <cuda/std/__type_traits/type_list.h>
+#if _CCCL_HAS_CTK()
 
-#include <nv/target>
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/dimensions.h>
+#  include <cuda/std/__type_traits/type_list.h>
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <nv/target>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -119,6 +121,7 @@ namespace __detail
 template <typename _Unit, typename _Level>
 struct __dims_helper;
 
+#  if _CCCL_CUDA_COMPILATION()
 template <typename _Level>
 struct __dims_helper<_Level, _Level>
 {
@@ -188,6 +191,7 @@ struct __dims_helper<cluster_level, grid_level>
     NV_IF_ELSE_TARGET(NV_PROVIDES_SM_90, (return __clusterIdx();), (return ::dim3(0, 0, 0);));
   }
 };
+#  endif // _CCCL_CUDA_COMPILATION()
 
 // Seems like a compiler bug, where NODISCARD is marked as ignored due to void
 // return type, while its not possible to ever have void return type here
@@ -414,6 +418,8 @@ _CCCL_DEVICE auto index(const _Unit&, const _Level&)
 } // namespace hierarchy
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_HIERARCHY_LEVELS_H
diff --git a/libcudacxx/include/cuda/__hierarchy/hierarchy_query_result.h b/libcudacxx/include/cuda/__hierarchy/hierarchy_query_result.h
index 0192198ba036..dd73dce727d0 100644
--- a/libcudacxx/include/cuda/__hierarchy/hierarchy_query_result.h
+++ b/libcudacxx/include/cuda/__hierarchy/hierarchy_query_result.h
@@ -21,13 +21,15 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__cstddef/types.h>
-#include <cuda/std/__type_traits/is_convertible.h>
-#include <cuda/std/__type_traits/is_nothrow_convertible.h>
-#include <cuda/std/__type_traits/is_same.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__cstddef/types.h>
+#  include <cuda/std/__type_traits/is_convertible.h>
+#  include <cuda/std/__type_traits/is_nothrow_convertible.h>
+#  include <cuda/std/__type_traits/is_same.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -144,6 +146,8 @@ struct hierarchy_query_result
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_HIERARCHY_QUERY_RESULT_H
diff --git a/libcudacxx/include/cuda/__hierarchy/level_dimensions.h b/libcudacxx/include/cuda/__hierarchy/level_dimensions.h
index ec4fcd38d9fd..42a76be3a689 100644
--- a/libcudacxx/include/cuda/__hierarchy/level_dimensions.h
+++ b/libcudacxx/include/cuda/__hierarchy/level_dimensions.h
@@ -21,15 +21,17 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/block_level.h>
-#include <cuda/__hierarchy/cluster_level.h>
-#include <cuda/__hierarchy/grid_level.h>
-#include <cuda/__hierarchy/hierarchy_levels.h>
-#include <cuda/std/span>
-#include <cuda/std/type_traits>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/block_level.h>
+#  include <cuda/__hierarchy/cluster_level.h>
+#  include <cuda/__hierarchy/grid_level.h>
+#  include <cuda/__hierarchy/hierarchy_levels.h>
+#  include <cuda/std/span>
+#  include <cuda/std/type_traits>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -140,10 +142,10 @@ struct level_dimensions
   _CCCL_API constexpr level_dimensions()
       : dims(){};
 
-#if !defined(_CCCL_NO_THREE_WAY_COMPARISON) && !_CCCL_COMPILER(MSVC, <, 19, 39) && !_CCCL_COMPILER(GCC, <, 12)
+#  if !defined(_CCCL_NO_THREE_WAY_COMPARISON) && !_CCCL_COMPILER(MSVC, <, 19, 39) && !_CCCL_COMPILER(GCC, <, 12)
   [[nodiscard]] _CCCL_HIDE_FROM_ABI constexpr bool operator==(const level_dimensions&) const noexcept = default;
-#else // ^^^ !_CCCL_NO_THREE_WAY_COMPARISON ^^^ / vvv
-      // _CCCL_NO_THREE_WAY_COMPARISON vvv
+#  else // ^^^ !_CCCL_NO_THREE_WAY_COMPARISON ^^^ / vvv
+        // _CCCL_NO_THREE_WAY_COMPARISON vvv
   [[nodiscard]] _CCCL_API friend constexpr bool
   operator==(const level_dimensions& __left, const level_dimensions& __right) noexcept
   {
@@ -155,7 +157,7 @@ struct level_dimensions
   {
     return __left.dims != __right.dims;
   }
-#endif // _CCCL_NO_THREE_WAY_COMPARISON
+#  endif // _CCCL_NO_THREE_WAY_COMPARISON
 };
 
 /**
@@ -234,6 +236,8 @@ _CCCL_API constexpr auto block_dims(_Dims __dims) noexcept
 }
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_LEVEL_DIMENSIONS_H
diff --git a/libcudacxx/include/cuda/__hierarchy/native_hierarchy_level_base.h b/libcudacxx/include/cuda/__hierarchy/native_hierarchy_level_base.h
index 2d97796a7a41..fed901e297f4 100644
--- a/libcudacxx/include/cuda/__hierarchy/native_hierarchy_level_base.h
+++ b/libcudacxx/include/cuda/__hierarchy/native_hierarchy_level_base.h
@@ -21,33 +21,40 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/hierarchy_level_base.h>
-#include <cuda/__hierarchy/hierarchy_query_result.h>
-#include <cuda/__hierarchy/traits.h>
-#include <cuda/std/__algorithm/max.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__cstddef/types.h>
-#include <cuda/std/__mdspan/extents.h>
-#include <cuda/std/__type_traits/common_type.h>
-#include <cuda/std/__type_traits/is_integer.h>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/hierarchy_level_base.h>
+#  include <cuda/__hierarchy/hierarchy_query_result.h>
+#  include <cuda/__hierarchy/traits.h>
+#  include <cuda/std/__algorithm/max.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__cstddef/types.h>
+#  include <cuda/std/__mdspan/extents.h>
+#  include <cuda/std/__type_traits/common_type.h>
+#  include <cuda/std/__type_traits/is_integer.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
 // cudafe++ makes the queries (that are device only) return void when compiling for host, which causes host compilers
 // to warn about applying [[nodiscard]] to a function that returns void.
 _CCCL_DIAG_PUSH
-#if _CCCL_CUDA_COMPILER(NVCC)
+#  if _CCCL_CUDA_COMPILER(NVCC)
 _CCCL_DIAG_SUPPRESS_GCC("-Wattributes")
 _CCCL_DIAG_SUPPRESS_CLANG("-Wignored-attributes")
 _CCCL_DIAG_SUPPRESS_NVHPC(nodiscard_doesnt_apply)
-#endif // _CCCL_CUDA_COMPILER(NVCC)
+#  endif // _CCCL_CUDA_COMPILER(NVCC)
 
 template <class _Level>
 struct __native_hierarchy_level_base : hierarchy_level_base<_Level>
 {
+  template <class _InLevel>
+  using __default_md_query_type = unsigned;
+  template <class _InLevel>
+  using __default_1d_query_type = ::cuda::std::size_t;
+
   using __base_type = hierarchy_level_base<_Level>;
   using __base_type::count;
   using __base_type::count_as;
@@ -55,18 +62,13 @@ struct __native_hierarchy_level_base : hierarchy_level_base<_Level>
   using __base_type::dims_as;
   using __base_type::extents;
   using __base_type::extents_as;
+  using __base_type::static_dims;
+
+#  if _CCCL_CUDA_COMPILATION()
   using __base_type::index;
   using __base_type::index_as;
   using __base_type::rank;
   using __base_type::rank_as;
-  using __base_type::static_dims;
-
-  template <class _InLevel>
-  using __default_md_query_type = unsigned;
-  template <class _InLevel>
-  using __default_1d_query_type = ::cuda::std::size_t;
-
-#if _CCCL_CUDA_COMPILATION()
 
   _CCCL_TEMPLATE(class _InLevel)
   _CCCL_REQUIRES(__is_native_hierarchy_level_v<_InLevel>)
@@ -187,7 +189,7 @@ struct __native_hierarchy_level_base : hierarchy_level_base<_Level>
     return __ret;
   }
 
-#endif // _CCCL_CUDA_COMPILATION()
+#  endif // _CCCL_CUDA_COMPILATION()
 };
 
 _CCCL_DIAG_POP
@@ -198,6 +200,8 @@ struct __native_hierarchy_level_base<grid_level> : hierarchy_level_base<grid_lev
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_NATIVE_HIERARCHY_LEVEL_BASE_H
diff --git a/libcudacxx/include/cuda/__hierarchy/thread_level.h b/libcudacxx/include/cuda/__hierarchy/thread_level.h
index f8beda576d1f..ad117180262b 100644
--- a/libcudacxx/include/cuda/__hierarchy/thread_level.h
+++ b/libcudacxx/include/cuda/__hierarchy/thread_level.h
@@ -21,14 +21,16 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/hierarchy_query_result.h>
-#include <cuda/__hierarchy/native_hierarchy_level_base.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__mdspan/extents.h>
-#include <cuda/std/__type_traits/is_integer.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/hierarchy_query_result.h>
+#  include <cuda/__hierarchy/native_hierarchy_level_base.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__mdspan/extents.h>
+#  include <cuda/std/__type_traits/is_integer.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -42,6 +44,8 @@ struct thread_level : __native_hierarchy_level_base<thread_level>
 
   using __base_type = __native_hierarchy_level_base<thread_level>;
   using __base_type::extents_as;
+
+#  if _CCCL_CUDA_COMPILATION()
   using __base_type::index_as;
   using __base_type::rank_as;
 
@@ -85,12 +89,15 @@ struct thread_level : __native_hierarchy_level_base<thread_level>
   {
     return static_cast<_Tp>(::cuda::ptx::get_sreg_laneid());
   }
+#  endif // _CCCL_CUDA_COMPILATION()
 };
 
 _CCCL_GLOBAL_CONSTANT thread_level gpu_thread;
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_THREAD_LEVEL_H
diff --git a/libcudacxx/include/cuda/__hierarchy/traits.h b/libcudacxx/include/cuda/__hierarchy/traits.h
index 77e44b8cfd3a..9818d7fc21cf 100644
--- a/libcudacxx/include/cuda/__hierarchy/traits.h
+++ b/libcudacxx/include/cuda/__hierarchy/traits.h
@@ -21,15 +21,17 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/std/__tuple_dir/get.h>
-#include <cuda/std/__type_traits/is_same.h>
-#include <cuda/std/__type_traits/remove_cvref.h>
-#include <cuda/std/__type_traits/type_list.h>
-#include <cuda/std/__type_traits/void_t.h>
-#include <cuda/std/cstdint>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/std/__tuple_dir/get.h>
+#  include <cuda/std/__type_traits/is_same.h>
+#  include <cuda/std/__type_traits/remove_cvref.h>
+#  include <cuda/std/__type_traits/type_list.h>
+#  include <cuda/std/__type_traits/void_t.h>
+#  include <cuda/std/cstdint>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -124,6 +126,8 @@ inline constexpr bool __is_or_has_hierarchy_member_v = __has_hierarchy_member<_T
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_TRAITS_H
diff --git a/libcudacxx/include/cuda/__hierarchy/warp_level.h b/libcudacxx/include/cuda/__hierarchy/warp_level.h
index 6e1d91703916..bfbfff2af66f 100644
--- a/libcudacxx/include/cuda/__hierarchy/warp_level.h
+++ b/libcudacxx/include/cuda/__hierarchy/warp_level.h
@@ -21,17 +21,19 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__fwd/hierarchy.h>
-#include <cuda/__hierarchy/block_level.h>
-#include <cuda/__hierarchy/hierarchy_query_result.h>
-#include <cuda/__hierarchy/native_hierarchy_level_base.h>
-#include <cuda/__hierarchy/thread_level.h>
-#include <cuda/__ptx/instructions/get_sreg.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__mdspan/extents.h>
-#include <cuda/std/__type_traits/is_integer.h>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__fwd/hierarchy.h>
+#  include <cuda/__hierarchy/block_level.h>
+#  include <cuda/__hierarchy/hierarchy_query_result.h>
+#  include <cuda/__hierarchy/native_hierarchy_level_base.h>
+#  include <cuda/__hierarchy/thread_level.h>
+#  include <cuda/__ptx/instructions/get_sreg.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__mdspan/extents.h>
+#  include <cuda/std/__type_traits/is_integer.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -41,6 +43,8 @@ struct warp_level : __native_hierarchy_level_base<warp_level>
 
   using __base_type = __native_hierarchy_level_base<warp_level>;
   using __base_type::extents_as;
+
+#  if _CCCL_CUDA_COMPILATION()
   using __base_type::index_as;
 
   _CCCL_TEMPLATE(class _Tp)
@@ -56,12 +60,15 @@ struct warp_level : __native_hierarchy_level_base<warp_level>
   {
     return {static_cast<_Tp>(gpu_thread.rank(block) / 32), 0, 0};
   }
+#  endif // _CCCL_CUDA_COMPILATION()
 };
 
 _CCCL_GLOBAL_CONSTANT warp_level warp;
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___HIERARCHY_WARP_LEVEL_H
diff --git a/libcudacxx/include/cuda/__memcpy_async/dispatch_memcpy_async.h b/libcudacxx/include/cuda/__memcpy_async/dispatch_memcpy_async.h
index 8fdfae61c6e9..5a281a1f1c20 100644
--- a/libcudacxx/include/cuda/__memcpy_async/dispatch_memcpy_async.h
+++ b/libcudacxx/include/cuda/__memcpy_async/dispatch_memcpy_async.h
@@ -55,8 +55,8 @@ template <::cuda::std::size_t _Align, typename _Group>
   char* __dest_char,
   char const* __src_char,
   ::cuda::std::size_t __size,
-  ::cuda::std::uint32_t __allowed_completions,
-  ::cuda::std::uint64_t* __bar_handle)
+  [[maybe_unused]] ::cuda::std::uint32_t __allowed_completions,
+  [[maybe_unused]] ::cuda::std::uint64_t* __bar_handle)
 {
   ::cuda::__cp_async_fallback_mechanism<_Align>(__group, __dest_char, __src_char, __size);
   return __completion_mechanism::__sync;
@@ -68,8 +68,8 @@ template <::cuda::std::size_t _Align, typename _Group>
   char* __dest_char,
   char const* __src_char,
   ::cuda::std::size_t __size,
-  ::cuda::std::uint32_t __allowed_completions,
-  ::cuda::std::uint64_t* __bar_handle)
+  [[maybe_unused]] ::cuda::std::uint32_t __allowed_completions,
+  [[maybe_unused]] ::cuda::std::uint64_t* __bar_handle)
 {
 #if __cccl_ptx_isa >= 800
   NV_IF_TARGET(
@@ -114,8 +114,8 @@ template <::cuda::std::size_t _Align, typename _Group>
   char* __dest_char,
   char const* __src_char,
   ::cuda::std::size_t __size,
-  ::cuda::std::uint32_t __allowed_completions,
-  ::cuda::std::uint64_t* __bar_handle)
+  [[maybe_unused]] ::cuda::std::uint32_t __allowed_completions,
+  [[maybe_unused]] ::cuda::std::uint64_t* __bar_handle)
 {
   NV_IF_ELSE_TARGET(
     NV_IS_DEVICE,
@@ -149,8 +149,8 @@ template <::cuda::std::size_t _Align, typename _Group>
   _Group const& __group,
   char* __dest_char,
   char const* __src_char,
-  ::cuda::std::size_t __size,
-  ::cuda::std::uint32_t __allowed_completions)
+  [[maybe_unused]] ::cuda::std::size_t __size,
+  [[maybe_unused]] ::cuda::std::uint32_t __allowed_completions)
 {
   _CCCL_ASSERT(!(__allowed_completions & uint32_t(__completion_mechanism::__mbarrier_complete_tx)),
                "Cannot allow mbarrier_complete_tx completion mechanism when not passing a barrier. ");
diff --git a/libcudacxx/include/cuda/__memcpy_async/is_local_smem_barrier.h b/libcudacxx/include/cuda/__memcpy_async/is_local_smem_barrier.h
index 7c9f69bf8d9d..994d557f4353 100644
--- a/libcudacxx/include/cuda/__memcpy_async/is_local_smem_barrier.h
+++ b/libcudacxx/include/cuda/__memcpy_async/is_local_smem_barrier.h
@@ -38,7 +38,7 @@ template <thread_scope _Sco,
           typename _CompF,
           bool _Is_mbarrier = (_Sco == thread_scope_block)
                            && ::cuda::std::is_same_v<_CompF, ::cuda::std::__empty_completion>>
-_CCCL_API inline bool __is_local_smem_barrier(barrier<_Sco, _CompF>& __barrier)
+_CCCL_API inline bool __is_local_smem_barrier([[maybe_unused]] barrier<_Sco, _CompF>& __barrier)
 {
   NV_IF_ELSE_TARGET(
     NV_IS_DEVICE,
diff --git a/libcudacxx/include/cuda/__memcpy_async/memcpy_completion.h b/libcudacxx/include/cuda/__memcpy_async/memcpy_completion.h
index a9441d0d40ee..cd0dcccfc9d6 100644
--- a/libcudacxx/include/cuda/__memcpy_async/memcpy_completion.h
+++ b/libcudacxx/include/cuda/__memcpy_async/memcpy_completion.h
@@ -107,8 +107,8 @@ struct __memcpy_completion_impl
   }
 
   template <typename _Group, thread_scope _Sco, typename _CompF>
-  [[nodiscard]] _CCCL_API inline static async_contract_fulfillment __defer_non_smem_barrier(
-    __completion_mechanism __cm, _Group const& __group, ::cuda::std::size_t __size, barrier<_Sco, _CompF>& __barrier)
+  [[nodiscard]] _CCCL_API inline static async_contract_fulfillment
+  __defer_non_smem_barrier(__completion_mechanism __cm, _Group const&, ::cuda::std::size_t, barrier<_Sco, _CompF>&)
   {
     // Overload for non-smem barriers.
     switch (__cm)
diff --git a/libcudacxx/include/cuda/__memcpy_async/try_get_barrier_handle.h b/libcudacxx/include/cuda/__memcpy_async/try_get_barrier_handle.h
index f73962782dc3..5817dfae535b 100644
--- a/libcudacxx/include/cuda/__memcpy_async/try_get_barrier_handle.h
+++ b/libcudacxx/include/cuda/__memcpy_async/try_get_barrier_handle.h
@@ -36,7 +36,7 @@ _CCCL_BEGIN_NAMESPACE_CUDA
 
 //! @brief __try_get_barrier_handle returns barrier handle of block-scoped barriers and a nullptr otherwise.
 template <thread_scope _Sco, typename _CompF>
-_CCCL_API inline ::cuda::std::uint64_t* __try_get_barrier_handle(barrier<_Sco, _CompF>& __barrier)
+_CCCL_API inline ::cuda::std::uint64_t* __try_get_barrier_handle(barrier<_Sco, _CompF>&)
 {
   return nullptr;
 }
diff --git a/libcudacxx/include/cuda/__memory_pool/device_memory_pool.h b/libcudacxx/include/cuda/__memory_pool/device_memory_pool.h
index 9cc447088df0..1c8838b0e2b5 100644
--- a/libcudacxx/include/cuda/__memory_pool/device_memory_pool.h
+++ b/libcudacxx/include/cuda/__memory_pool/device_memory_pool.h
@@ -21,18 +21,15 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_CUDA_COMPILER(CLANG)
-#  include <cuda_runtime.h>
-#  include <cuda_runtime_api.h>
-#endif // _CCCL_CUDA_COMPILER(CLANG)
+#if _CCCL_HAS_CTK()
 
-#include <cuda/__memory_pool/memory_pool_base.h>
-#include <cuda/__memory_resource/get_property.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__runtime/api_wrapper.h>
-#include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/__memory_pool/memory_pool_base.h>
+#  include <cuda/__memory_resource/get_property.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__runtime/api_wrapper.h>
+#  include <cuda/std/__concepts/concept_macros.h>
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__cccl/prologue.h>
 
 //! @file
 //! The \c device_memory_pool class provides an asynchronous memory resource
@@ -161,6 +158,8 @@ static_assert(::cuda::mr::resource_with<device_memory_pool, ::cuda::mr::device_a
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_DEVICE_MEMORY_POOL_H
diff --git a/libcudacxx/include/cuda/__memory_pool/memory_pool_base.h b/libcudacxx/include/cuda/__memory_pool/memory_pool_base.h
index d5de79377806..01db9b8fe3be 100644
--- a/libcudacxx/include/cuda/__memory_pool/memory_pool_base.h
+++ b/libcudacxx/include/cuda/__memory_pool/memory_pool_base.h
@@ -21,23 +21,20 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_CUDA_COMPILER(CLANG)
-#  include <cuda_runtime.h>
-#  include <cuda_runtime_api.h>
-#endif // _CCCL_CUDA_COMPILER(CLANG)
-
-#include <cuda/__device/attributes.h>
-#include <cuda/__device/device_ref.h>
-#include <cuda/__memory_resource/any_resource.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__runtime/api_wrapper.h>
-#include <cuda/__stream/internal_streams.h>
-#include <cuda/__stream/stream.h>
-#include <cuda/__stream/stream_ref.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/cstddef>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__device/attributes.h>
+#  include <cuda/__device/device_ref.h>
+#  include <cuda/__memory_resource/any_resource.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__runtime/api_wrapper.h>
+#  include <cuda/__stream/internal_streams.h>
+#  include <cuda/__stream/stream.h>
+#  include <cuda/__stream/stream_ref.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/cstddef>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -199,9 +196,9 @@ inline void __verify_device_supports_export_handle_type(
     return;
   }
   if (__location.type != ::CU_MEM_LOCATION_TYPE_DEVICE
-#if _CCCL_CTK_AT_LEAST(12, 6)
+#  if _CCCL_CTK_AT_LEAST(12, 6)
       && __location.type != ::CU_MEM_LOCATION_TYPE_HOST_NUMA
-#endif
+#  endif
   )
   {
     ::cuda::__throw_cuda_error(::cudaErrorNotSupported,
@@ -222,17 +219,17 @@ __get_default_memory_pool(const CUmemLocation __location, [[maybe_unused]] const
   auto __device = __location.type == ::CU_MEM_LOCATION_TYPE_DEVICE ? __location.id : 0;
   ::cuda::__verify_device_supports_stream_ordered_allocations(__device);
 
-#if _CCCL_CTK_AT_LEAST(13, 0)
+#  if _CCCL_CTK_AT_LEAST(13, 0)
   ::cudaMemPool_t __pool = ::cuda::__driver::__getDefaultMemPool(__location, __allocation_type);
   if (::cuda::memory_pool_attributes::release_threshold(__pool) == 0)
   {
     ::cuda::memory_pool_attributes::release_threshold.set(__pool, ::cuda::std::numeric_limits<size_t>::max());
   }
-#else // ^^^ _CCCL_CTK_AT_LEAST(13, 0) ^^^ / vvv _CCCL_CTK_BELOW(13, 0) vvv
+#  else // ^^^ _CCCL_CTK_AT_LEAST(13, 0) ^^^ / vvv _CCCL_CTK_BELOW(13, 0) vvv
   _CCCL_ASSERT(__location.type == ::CU_MEM_LOCATION_TYPE_DEVICE,
                "Before CUDA 13 only device memory pools have a default");
   ::cudaMemPool_t __pool = ::cuda::__driver::__deviceGetDefaultMemPool(__device);
-#endif // ^^^ _CCCL_CTK_BELOW(13, 0) ^^^
+#  endif // ^^^ _CCCL_CTK_BELOW(13, 0) ^^^
   return __pool;
 }
 
@@ -294,27 +291,27 @@ struct memory_pool_properties
   __pool_properties.handleTypes = ::CUmemAllocationHandleType(__properties.allocation_handle_type);
   __pool_properties.location    = __location;
 
-#if _CCCL_CTK_AT_LEAST(12, 2)
+#  if _CCCL_CTK_AT_LEAST(12, 2)
   if (__properties.max_pool_size != 0)
   {
-#  if _CCCL_CTK_AT_LEAST(13, 0)
+#    if _CCCL_CTK_AT_LEAST(13, 0)
     if (__allocation_type == ::CU_MEM_ALLOCATION_TYPE_MANAGED)
     {
       ::cuda::std::__throw_invalid_argument("Max pool size is not supported for managed memory pools");
     }
-#  endif // _CCCL_CTK_AT_LEAST(13, 0)
+#    endif // _CCCL_CTK_AT_LEAST(13, 0)
     if (__properties.initial_pool_size > __properties.max_pool_size)
     {
       ::cuda::std::__throw_invalid_argument("Initial pool size must be less than the max pool size");
     }
   }
   __pool_properties.maxSize = __properties.max_pool_size;
-#else
+#  else
   if (__properties.max_pool_size != 0)
   {
     ::cuda::std::__throw_invalid_argument("Max pool size is not supported on this CUDA version");
   }
-#endif // _CCCL_CTK_AT_LEAST(12, 2)
+#  endif // _CCCL_CTK_AT_LEAST(12, 2)
 
   if (__properties.initial_pool_size > __properties.release_threshold)
   {
@@ -627,18 +624,20 @@ class __memory_pool_base
     return __pool_ == __rhs.__pool_;
   }
 
-#if _CCCL_STD_VER <= 2017
+#  if _CCCL_STD_VER <= 2017
   //! @brief Inequality comparison with another __memory_pool_base.
   //! @returns true if underlying \c cudaMemPool_t are not equal.
   [[nodiscard]] _CCCL_HOST_API bool operator!=(__memory_pool_base const& __rhs) const noexcept
   {
     return __pool_ != __rhs.__pool_;
   }
-#endif // _CCCL_STD_VER <= 2017
+#  endif // _CCCL_STD_VER <= 2017
 };
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___MEMORY_RESOURCE_MEMORY_POOL_BASE_H
diff --git a/libcudacxx/include/cuda/__memory_pool/pinned_memory_pool.h b/libcudacxx/include/cuda/__memory_pool/pinned_memory_pool.h
index 89318fc59209..38b9702e6563 100644
--- a/libcudacxx/include/cuda/__memory_pool/pinned_memory_pool.h
+++ b/libcudacxx/include/cuda/__memory_pool/pinned_memory_pool.h
@@ -21,25 +21,22 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_CUDA_COMPILER(CLANG)
-#  include <cuda_runtime.h>
-#  include <cuda_runtime_api.h>
-#endif // _CCCL_CUDA_COMPILER(CLANG)
+#if _CCCL_HAS_CTK()
 
-#include <cuda/__device/all_devices.h>
-#include <cuda/__memory_pool/memory_pool_base.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__exception/throw_error.h>
+#  include <cuda/__device/all_devices.h>
+#  include <cuda/__memory_pool/memory_pool_base.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__exception/throw_error.h>
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__cccl/prologue.h>
 
 //! @file
 //! The \c pinned_memory_resource class provides a memory resource that
 //! allocates pinned memory.
 _CCCL_BEGIN_NAMESPACE_CUDA
 
-#if _CCCL_CTK_AT_LEAST(12, 6)
+#  if _CCCL_CTK_AT_LEAST(12, 6)
 
 static ::cudaMemPool_t __get_default_host_pinned_pool();
 
@@ -110,7 +107,7 @@ struct pinned_memory_pool : pinned_memory_pool_ref
 {
   using reference_type = pinned_memory_pool_ref;
 
-#  if _CCCL_CTK_AT_LEAST(13, 0)
+#    if _CCCL_CTK_AT_LEAST(13, 0)
   //! @brief Constructs a \c pinned_memory_pool with optional properties.
   //! Properties include the initial pool size and the release threshold. If the
   //! pool size grows beyond the release threshold, unused memory held by the
@@ -129,7 +126,7 @@ struct pinned_memory_pool : pinned_memory_pool_ref
   {
     enable_access_from(cuda::devices);
   }
-#  endif // _CCCL_CTK_AT_LEAST(13, 0)
+#    endif // _CCCL_CTK_AT_LEAST(13, 0)
 
   //! @brief Constructs a \c pinned_memory_pool with the specified NUMA node id
   //! and optional properties. Properties include the initial pool size and the
@@ -190,7 +187,7 @@ static_assert(::cuda::mr::resource_with<pinned_memory_pool, ::cuda::mr::host_acc
 
 [[nodiscard]] static ::cudaMemPool_t __get_default_host_pinned_pool()
 {
-#  if _CCCL_CTK_AT_LEAST(13, 0)
+#    if _CCCL_CTK_AT_LEAST(13, 0)
   static ::cudaMemPool_t __default_pool = []() {
     ::cudaMemPool_t __pool = ::cuda::__get_default_memory_pool(
       ::CUmemLocation{::CU_MEM_LOCATION_TYPE_HOST, 0}, ::CU_MEM_ALLOCATION_TYPE_PINNED);
@@ -200,19 +197,21 @@ static_assert(::cuda::mr::resource_with<pinned_memory_pool, ::cuda::mr::host_acc
     return __pool;
   }();
 
-#  else // ^^^ _CCCL_CTK_AT_LEAST(13, 0) ^^^ / vvv _CCCL_CTK_BELOW(13, 0) vvv
+#    else // ^^^ _CCCL_CTK_AT_LEAST(13, 0) ^^^ / vvv _CCCL_CTK_BELOW(13, 0) vvv
   static ::cudaMemPool_t __default_pool = []() {
     cuda::pinned_memory_pool __pool(0);
     return __pool.release();
   }();
-#  endif // ^^^ _CCCL_CTK_BELOW(13, 0) ^^^
+#    endif // ^^^ _CCCL_CTK_BELOW(13, 0) ^^^
   return __default_pool;
 }
 
-#endif // _CCCL_CTK_AT_LEAST(12, 6)
+#  endif // _CCCL_CTK_AT_LEAST(12, 6)
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_PINNED_MEMORY_POOL_H
diff --git a/libcudacxx/include/cuda/__memory_resource/any_resource.h b/libcudacxx/include/cuda/__memory_resource/any_resource.h
index 97f4e370034c..949000c91a21 100644
--- a/libcudacxx/include/cuda/__memory_resource/any_resource.h
+++ b/libcudacxx/include/cuda/__memory_resource/any_resource.h
@@ -21,19 +21,21 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__memory_resource/get_property.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource.h>
-#include <cuda/__utility/basic_any.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__utility/delegate_constructors.h>
-#include <cuda/std/__utility/forward.h>
-#include <cuda/std/optional>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__memory_resource/get_property.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__memory_resource/resource.h>
+#  include <cuda/__utility/basic_any.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__utility/delegate_constructors.h>
+#  include <cuda/std/__utility/forward.h>
+#  include <cuda/std/optional>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA_MR
-#ifndef _CCCL_DOXYGEN_INVOKED // Do not document this
+#  ifndef _CCCL_DOXYGEN_INVOKED // Do not document this
 
 template <class _Property>
 using __property_result_t _CCCL_NODEBUG_ALIAS = ::cuda::std::__type_call1< //
@@ -376,7 +378,7 @@ synchronous_resource_ref<_Properties...> __as_resource_ref(resource_ref<_Propert
   return __mr;
 }
 
-#else // ^^^ !_CCCL_DOXYGEN_INVOKED ^^^ / vvv _CCCL_DOXYGEN_INVOKED vvv
+#  else // ^^^ !_CCCL_DOXYGEN_INVOKED ^^^ / vvv _CCCL_DOXYGEN_INVOKED vvv
 
 enum class _ResourceKind
 {
@@ -821,7 +823,7 @@ using synchronous_resource_ref = basic_resource_ref<_ResourceKind::_Synchronous,
 template <class... _Properties>
 using resource_ref = basic_resource_ref<_ResourceKind::_Asynchronous, _Properties...>;
 
-#endif // _CCCL_DOXYGEN_INVOKED
+#  endif // _CCCL_DOXYGEN_INVOKED
 
 //! @rst
 //! .. _cudax-memory-resource-make-any-resource:
@@ -877,6 +879,8 @@ auto make_any_resource(_Args&&... __args) -> any_resource<_Properties...>
 
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___MEMORY_RESOURCE_ANY_RESOURCE_H
diff --git a/libcudacxx/include/cuda/__memory_resource/get_memory_resource.h b/libcudacxx/include/cuda/__memory_resource/get_memory_resource.h
index f9a2bd5c1ea4..3c984f86df3b 100644
--- a/libcudacxx/include/cuda/__memory_resource/get_memory_resource.h
+++ b/libcudacxx/include/cuda/__memory_resource/get_memory_resource.h
@@ -21,15 +21,17 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource.h>
-#include <cuda/__stream/stream_ref.h>
-#include <cuda/std/__concepts/equality_comparable.h>
-#include <cuda/std/__execution/env.h>
-#include <cuda/std/__type_traits/is_same.h>
-#include <cuda/std/__type_traits/remove_cvref.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__memory_resource/resource.h>
+#  include <cuda/__stream/stream_ref.h>
+#  include <cuda/std/__concepts/equality_comparable.h>
+#  include <cuda/std/__execution/env.h>
+#  include <cuda/std/__type_traits/is_same.h>
+#  include <cuda/std/__type_traits/remove_cvref.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA_MR
 
@@ -77,6 +79,8 @@ _CCCL_GLOBAL_CONSTANT auto get_memory_resource = get_memory_resource_t{};
 
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA__MEMORY_RESOURCE_GET_MEMORY_RESOURCE_H
diff --git a/libcudacxx/include/cuda/__memory_resource/get_property.h b/libcudacxx/include/cuda/__memory_resource/get_property.h
index 4de47677cd9a..9623d4f62107 100644
--- a/libcudacxx/include/cuda/__memory_resource/get_property.h
+++ b/libcudacxx/include/cuda/__memory_resource/get_property.h
@@ -21,12 +21,14 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/std/__concepts/same_as.h>
-#include <cuda/std/__type_traits/remove_const_ref.h>
-#include <cuda/std/__type_traits/void_t.h>
-#include <cuda/std/__utility/declval.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__concepts/same_as.h>
+#  include <cuda/std/__type_traits/remove_const_ref.h>
+#  include <cuda/std/__type_traits/void_t.h>
+#  include <cuda/std/__utility/declval.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -148,6 +150,8 @@ using forward_property = __forward_property::__fn<_Derived, _Upstream>;
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_GET_PROPERTY_H
diff --git a/libcudacxx/include/cuda/__memory_resource/legacy_managed_memory_resource.h b/libcudacxx/include/cuda/__memory_resource/legacy_managed_memory_resource.h
index 8db6cffbf549..fe3c7af16863 100644
--- a/libcudacxx/include/cuda/__memory_resource/legacy_managed_memory_resource.h
+++ b/libcudacxx/include/cuda/__memory_resource/legacy_managed_memory_resource.h
@@ -21,20 +21,18 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_CUDA_COMPILER(CLANG)
-#  include <cuda_runtime_api.h>
-#endif // _CCCL_CUDA_COMPILER(CLANG)
+#if _CCCL_HAS_CTK()
 
-#include <cuda/__memory_resource/any_resource.h>
-#include <cuda/__memory_resource/get_property.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource.h>
-#include <cuda/__runtime/api_wrapper.h>
-#include <cuda/__stream/internal_streams.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__exception/throw_error.h>
+#  include <cuda/__memory_resource/any_resource.h>
+#  include <cuda/__memory_resource/get_property.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__memory_resource/resource.h>
+#  include <cuda/__runtime/api_wrapper.h>
+#  include <cuda/__stream/internal_streams.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__exception/throw_error.h>
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__cccl/prologue.h>
 
 //! @file
 //! The \c managed_memory_resource class provides a memory resource that allocates managed memory.
@@ -107,7 +105,7 @@ class legacy_managed_memory_resource
   {
     return __flags_ == __other.__flags_;
   }
-#if _CCCL_STD_VER <= 2017
+#  if _CCCL_STD_VER <= 2017
   //! @brief Inequality comparison with another \c managed_memory_resource.
   //! @param __other The other \c managed_memory_resource.
   //! @return Whether both \c managed_memory_resource were constructed with different flags.
@@ -115,7 +113,7 @@ class legacy_managed_memory_resource
   {
     return __flags_ != __other.__flags_;
   }
-#endif // _CCCL_STD_VER <= 2017
+#  endif // _CCCL_STD_VER <= 2017
 
   //! @brief Enables the \c device_accessible property
   _CCCL_HOST_API friend constexpr void
@@ -143,6 +141,8 @@ static_assert(::cuda::mr::synchronous_resource_with<legacy_managed_memory_resour
 
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_LEGACY_MANAGED_MEMORY_RESOURCE_H
diff --git a/libcudacxx/include/cuda/__memory_resource/legacy_pinned_memory_resource.h b/libcudacxx/include/cuda/__memory_resource/legacy_pinned_memory_resource.h
index a43ccb248001..911dafb428e5 100644
--- a/libcudacxx/include/cuda/__memory_resource/legacy_pinned_memory_resource.h
+++ b/libcudacxx/include/cuda/__memory_resource/legacy_pinned_memory_resource.h
@@ -21,20 +21,17 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_CUDA_COMPILER(CLANG)
-#  include <cuda_runtime.h>
-#  include <cuda_runtime_api.h>
-#endif // _CCCL_CUDA_COMPILER(CLANG)
+#if _CCCL_HAS_CTK()
 
-#include <cuda/__device/device_ref.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource.h>
-#include <cuda/__runtime/api_wrapper.h>
-#include <cuda/__runtime/ensure_current_context.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__exception/throw_error.h>
+#  include <cuda/__device/device_ref.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__memory_resource/resource.h>
+#  include <cuda/__runtime/api_wrapper.h>
+#  include <cuda/__runtime/ensure_current_context.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__exception/throw_error.h>
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__cccl/prologue.h>
 
 //! @file
 //! The \c legacy_pinned_memory_resource class provides a memory resource that allocates pinned memory.
@@ -99,7 +96,7 @@ class legacy_pinned_memory_resource
   {
     return true;
   }
-#if _CCCL_STD_VER <= 2017
+#  if _CCCL_STD_VER <= 2017
   //! @brief Equality comparison with another \c legacy_pinned_memory_resource.
   //! @param __other The other \c legacy_pinned_memory_resource.
   //! @return Whether both \c legacy_pinned_memory_resource were constructed with different flags.
@@ -107,7 +104,7 @@ class legacy_pinned_memory_resource
   {
     return false;
   }
-#endif // _CCCL_STD_VER <= 2017
+#  endif // _CCCL_STD_VER <= 2017
 
   //! @brief Enables the \c device_accessible property
   _CCCL_HOST_API friend constexpr void
@@ -136,6 +133,8 @@ static_assert(::cuda::mr::synchronous_resource_with<legacy_pinned_memory_resourc
 
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_LEGACY_PINNED_MEMORY_RESOURCE_H
diff --git a/libcudacxx/include/cuda/__memory_resource/properties.h b/libcudacxx/include/cuda/__memory_resource/properties.h
index c6f8ade529b7..6eac8634755c 100644
--- a/libcudacxx/include/cuda/__memory_resource/properties.h
+++ b/libcudacxx/include/cuda/__memory_resource/properties.h
@@ -21,11 +21,13 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/std/__type_traits/decay.h>
-#include <cuda/std/__type_traits/type_set.h>
-#include <cuda/std/cstddef>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__type_traits/decay.h>
+#  include <cuda/std/__type_traits/type_set.h>
+#  include <cuda/std/cstddef>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA_MR
 
@@ -125,6 +127,8 @@ struct __memory_accessability_from_properties
 
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_PROPERTIES_H
diff --git a/libcudacxx/include/cuda/__memory_resource/resource.h b/libcudacxx/include/cuda/__memory_resource/resource.h
index 3e169651a95b..c9ffeacfc451 100644
--- a/libcudacxx/include/cuda/__memory_resource/resource.h
+++ b/libcudacxx/include/cuda/__memory_resource/resource.h
@@ -21,19 +21,21 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__memory_resource/get_property.h>
-#include <cuda/__stream/stream_ref.h>
-#include <cuda/__utility/__basic_any/semiregular.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/std/__concepts/convertible_to.h>
-#include <cuda/std/__concepts/equality_comparable.h>
-#include <cuda/std/__concepts/same_as.h>
-#include <cuda/std/__tuple_dir/sfinae_helpers.h>
-#include <cuda/std/__type_traits/decay.h>
-#include <cuda/std/__type_traits/fold.h>
-#include <cuda/std/__type_traits/is_same.h>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__memory_resource/get_property.h>
+#  include <cuda/__stream/stream_ref.h>
+#  include <cuda/__utility/__basic_any/semiregular.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/std/__concepts/convertible_to.h>
+#  include <cuda/std/__concepts/equality_comparable.h>
+#  include <cuda/std/__concepts/same_as.h>
+#  include <cuda/std/__tuple_dir/sfinae_helpers.h>
+#  include <cuda/std/__type_traits/decay.h>
+#  include <cuda/std/__type_traits/fold.h>
+#  include <cuda/std/__type_traits/is_same.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA_MR
 
@@ -130,6 +132,8 @@ _CCCL_CONCEPT __non_polymorphic_resources = _CCCL_REQUIRES_EXPR((_Resource, _Oth
 
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_RESOURCE_H
diff --git a/libcudacxx/include/cuda/__memory_resource/shared_resource.h b/libcudacxx/include/cuda/__memory_resource/shared_resource.h
index fe01c2e798f1..3bfc7d590eb8 100644
--- a/libcudacxx/include/cuda/__memory_resource/shared_resource.h
+++ b/libcudacxx/include/cuda/__memory_resource/shared_resource.h
@@ -21,19 +21,22 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource.h>
-#include <cuda/std/__new_>
-#include <cuda/std/__type_traits/is_swappable.h>
-#include <cuda/std/__utility/exchange.h>
-#include <cuda/std/__utility/forward.h>
-#include <cuda/std/__utility/in_place.h>
-#include <cuda/std/__utility/move.h>
-#include <cuda/std/atomic>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__memory_resource/resource.h>
+#  include <cuda/std/__new_>
+#  include <cuda/std/__type_traits/is_swappable.h>
+#  include <cuda/std/__utility/exchange.h>
+#  include <cuda/std/__utility/forward.h>
+#  include <cuda/std/__utility/in_place.h>
+#  include <cuda/std/__utility/move.h>
+#  include <cuda/std/atomic>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA_MR
+
 //! @rst
 //! .. _cudax-memory-resource-shared-resource:
 //!
@@ -254,8 +257,11 @@ auto make_shared_resource(_Args&&... __args) -> shared_resource<_Resource>
                 "_Resource does not satisfy the cuda::mr::synchronous_resource concept");
   return shared_resource<_Resource>{::cuda::std::in_place_type<_Resource>, ::cuda::std::forward<_Args>(__args)...};
 }
+
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___MEMORY_RESOURCE_SHARED_RESOURCE_H
diff --git a/libcudacxx/include/cuda/__memory_resource/synchronous_resource_adapter.h b/libcudacxx/include/cuda/__memory_resource/synchronous_resource_adapter.h
index 634ad7f7b336..65eecbcbe0f1 100644
--- a/libcudacxx/include/cuda/__memory_resource/synchronous_resource_adapter.h
+++ b/libcudacxx/include/cuda/__memory_resource/synchronous_resource_adapter.h
@@ -21,13 +21,15 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__memory_resource/get_property.h>
-#include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource.h>
-#include <cuda/std/__concepts/concept_macros.h>
-#include <cuda/stream>
+#if _CCCL_HAS_CTK()
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/__memory_resource/get_property.h>
+#  include <cuda/__memory_resource/properties.h>
+#  include <cuda/__memory_resource/resource.h>
+#  include <cuda/std/__concepts/concept_macros.h>
+#  include <cuda/stream>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA_MR
 
@@ -103,12 +105,12 @@ struct synchronous_resource_adapter
     return __resource == __rhs.__resource;
   }
 
-#if _CCCL_STD_VER <= 2017
+#  if _CCCL_STD_VER <= 2017
   [[nodiscard]] _CCCL_HOST_API bool operator!=(const synchronous_resource_adapter& __rhs) const noexcept
   {
     return __resource != __rhs.__resource;
   }
-#endif // _CCCL_STD_VER <= 2017
+#  endif // _CCCL_STD_VER <= 2017
 
   _CCCL_HOST_API _Resource& upstream_resource() noexcept
   {
@@ -138,6 +140,8 @@ _CCCL_HOST_API decltype(auto) __adapt_if_synchronous(_Resource&& __resource) noe
 }
 _CCCL_END_NAMESPACE_CUDA_MR
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif //_CUDA___MEMORY_RESOURCE_SYNCHRONOUS_RESOURCE_ADAPTER_H
diff --git a/libcudacxx/include/cuda/__stream/internal_streams.h b/libcudacxx/include/cuda/__stream/internal_streams.h
index e4f74c9b9b29..6cce1db7c2c3 100644
--- a/libcudacxx/include/cuda/__stream/internal_streams.h
+++ b/libcudacxx/include/cuda/__stream/internal_streams.h
@@ -21,11 +21,11 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__stream/stream.h>
+#if _CCCL_HAS_CTK()
 
-#include <cuda_runtime_api.h>
+#  include <cuda/__stream/stream.h>
 
-#include <cuda/std/__cccl/prologue.h>
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 
@@ -44,6 +44,8 @@ inline ::cuda::stream_ref __cccl_allocation_stream()
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA___STREAM_INTERNAL_STREAMS_H
diff --git a/libcudacxx/include/cuda/__stream/launch_transform.h b/libcudacxx/include/cuda/__stream/launch_transform.h
index d0dbfb7851fa..cfe876466503 100644
--- a/libcudacxx/include/cuda/__stream/launch_transform.h
+++ b/libcudacxx/include/cuda/__stream/launch_transform.h
@@ -21,20 +21,22 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/__stream/stream_ref.h>
-#include <cuda/__type_traits/is_instantiable_with.h>
-#include <cuda/std/__memory/addressof.h>
-#include <cuda/std/__memory/construct_at.h>
-#include <cuda/std/__new/launder.h>
-#include <cuda/std/__optional/optional.h>
-#include <cuda/std/__type_traits/decay.h>
-#include <cuda/std/__type_traits/is_callable.h>
-#include <cuda/std/__type_traits/is_reference.h>
-#include <cuda/std/__utility/declval.h>
-#include <cuda/std/__utility/forward.h>
-#include <cuda/std/__utility/move.h>
-
-#include <cuda/std/__cccl/prologue.h>
+#if _CCCL_HAS_CTK()
+
+#  include <cuda/__stream/stream_ref.h>
+#  include <cuda/__type_traits/is_instantiable_with.h>
+#  include <cuda/std/__memory/addressof.h>
+#  include <cuda/std/__memory/construct_at.h>
+#  include <cuda/std/__new/launder.h>
+#  include <cuda/std/__optional/optional.h>
+#  include <cuda/std/__type_traits/decay.h>
+#  include <cuda/std/__type_traits/is_callable.h>
+#  include <cuda/std/__type_traits/is_reference.h>
+#  include <cuda/std/__utility/declval.h>
+#  include <cuda/std/__utility/forward.h>
+#  include <cuda/std/__utility/move.h>
+
+#  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA
 namespace __detail
@@ -188,6 +190,8 @@ using transformed_device_argument_t _CCCL_NODEBUG_ALIAS =
 
 _CCCL_END_NAMESPACE_CUDA
 
-#include <cuda/std/__cccl/epilogue.h>
+#  include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CCCL_HAS_CTK()
 
 #endif // _CUDA__STREAM_LAUNCH_TRANSFORM_H
diff --git a/libcudacxx/include/cuda/pipeline b/libcudacxx/include/cuda/pipeline
index 110fe60e70a9..e100d6aa3b5c 100644
--- a/libcudacxx/include/cuda/pipeline
+++ b/libcudacxx/include/cuda/pipeline
@@ -473,15 +473,15 @@ _CCCL_API inline pipeline<thread_scope_thread> make_pipeline()
 }
 
 template <uint8_t _Prior>
-_CCCL_API inline void pipeline_consumer_wait_prior(pipeline<thread_scope_thread>& __pipeline)
+_CCCL_API inline void pipeline_consumer_wait_prior([[maybe_unused]] pipeline<thread_scope_thread>& __pipeline)
 {
   NV_IF_TARGET(NV_PROVIDES_SM_80, ::cuda::device::__pipeline_consumer_wait<_Prior>(__pipeline);
                __pipeline.__tail = __pipeline.__head - _Prior;)
 }
 
 template <thread_scope _Scope>
-_CCCL_API inline void
-pipeline_producer_commit([[maybe_unused]] pipeline<thread_scope_thread>& __pipeline, barrier<_Scope>& __barrier)
+_CCCL_API inline void pipeline_producer_commit([[maybe_unused]] pipeline<thread_scope_thread>& __pipeline,
+                                               [[maybe_unused]] barrier<_Scope>& __barrier)
 {
   NV_IF_TARGET(NV_PROVIDES_SM_80,
                ((void) __memcpy_completion_impl::__defer(