Allow all public headers to be included with host compilers only

libcudacxx/cmake/LibcudacxxPublicHeaderTestingHost.cmake

@@ -6,25 +6,16 @@
 
 # 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(
   GLOB public_headers_host_only
   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)

libcudacxx/include/cuda/__container/buffer.h

@@ -21,35 +21,37 @@
 #  pragma system_header
 #endif // no system header
 
-#if _CCCL_CUDA_COMPILATION()
-#  include <cub/device/device_transform.cuh>
-#endif // _CCCL_CUDA_COMPILATION()
-
-#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_CUDA_COMPILATION()
+#    include <cub/device/device_transform.cuh>
+#  endif // _CCCL_CUDA_COMPILATION()
+
+#  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
 
   //! @}
 
@@ -683,13 +685,13 @@ __fill_n(cuda::stream_ref __stream, _Tp* __first, ::cuda::std::size_t __count, c
     }
     else
     {
-#if _CCCL_CUDA_COMPILATION()
+#  if _CCCL_CUDA_COMPILATION()
       ::cuda::__ensure_current_context __guard(__stream);
       ::cub::DeviceTransform::Fill(__first, __count, __value, __stream.get());
-#else // ^^^ _CCCL_CUDA_COMPILATION() ^^^ / vvv !_CCCL_CUDA_COMPILATION() vvv
+#  else // ^^^ _CCCL_CUDA_COMPILATION() ^^^ / vvv !_CCCL_CUDA_COMPILATION() vvv
       static_assert(sizeof(_Tp) <= 4,
                     "CUDA compiler is required to initialize an async_buffer with elements larger than 4 bytes");
-#endif // ^^^ !_CCCL_CUDA_COMPILATION() ^^^
+#  endif // ^^^ !_CCCL_CUDA_COMPILATION() ^^^
     }
   }
 }
@@ -886,6 +888,8 @@ 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

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

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