Move host stdlib wrappers to <cuda/std/__host_stdlib> directory (#7411)

Author: davebayer

cub/cub/detail/ptx-json-parser.cuh

@@ -5,7 +5,7 @@
 
 #include <cub/config.cuh>
 
-#include <cuda/std/__cccl/algorithm_wrapper.h>
+#include <cuda/std/__host_stdlib/algorithm>
 
 #include <format>
 #include <string_view>

cub/cub/util_allocator.cuh

@@ -22,12 +22,12 @@
 #include <cub/util_debug.cuh>
 #include <cub/util_namespace.cuh>
 
+#include <cuda/std/__host_stdlib/math.h>
+
 #include <map>
 #include <mutex>
 #include <set>
 
-#include <math.h>
-
 CUB_NAMESPACE_BEGIN
 
 /******************************************************************************

cub/test/cmake/check_source_files.cmake

@@ -64,9 +64,9 @@ endif()
 # stdpar library.
 #
 # The headers following headers should be used instead:
-# <algorithm> -> <cuda/std/__cccl/algorithm_wrapper.h>
-# <memory>    -> <cuda/std/__cccl/memory_wrapper.h>
-# <numeric>   -> <cuda/std/__cccl/numeric_wrapper.h>
+# <algorithm> -> <cuda/std/__host_stdlib/algorithm>
+# <memory>    -> <cuda/std/__host_stdlib/memory>
+# <numeric>   -> <cuda/std/__host_stdlib/numeric>
 #
 set(
   stdpar_header_exclusions
@@ -143,21 +143,21 @@ foreach (src ${cub_srcs})
 
     if (NOT algorithm_count EQUAL 0)
       message(
-        "'${src}' includes the <algorithm> header. Replace with <cuda/std/__cccl/algorithm_wrapper.h>."
+        "'${src}' includes the <algorithm> header. Replace with <cuda/std/__host_stdlib/algorithm>."
       )
       set(found_errors 1)
     endif()
 
     if (NOT memory_count EQUAL 0)
       message(
-        "'${src}' includes the <memory> header. Replace with <cuda/std/__cccl/memory_wrapper.h>."
+        "'${src}' includes the <memory> header. Replace with <cuda/std/__host_stdlib/memory>."
       )
       set(found_errors 1)
     endif()
 
     if (NOT numeric_count EQUAL 0)
       message(
-        "'${src}' includes the <numeric> header. Replace with <cuda/std/__cccl/numeric_wrapper.h>."
+        "'${src}' includes the <numeric> header. Replace with <cuda/std/__host_stdlib/numeric>."
       )
       set(found_errors 1)
     endif()

cudax/include/cuda/experimental/__cuco/__hyperloglog/hyperloglog_impl.cuh

@@ -32,6 +32,7 @@
 #include <cuda/std/__bit/countr.h>
 #include <cuda/std/__bit/integral.h>
 #include <cuda/std/__cstddef/types.h>
+#include <cuda/std/__host_stdlib/stdexcept>
 #include <cuda/std/__iterator/concepts.h>
 #include <cuda/std/__memory/addressof.h>
 #include <cuda/std/__memory/pointer_traits.h>
@@ -44,8 +45,6 @@
 #include <cuda/experimental/__cuco/hash_functions.cuh>
 #include <cuda/experimental/memory_resource.cuh>
 
-#include <stdexcept>
-
 #include <cooperative_groups.h>
 
 #include <cooperative_groups/reduce.h>

cudax/include/cuda/experimental/__cuco/hyperloglog.cuh

@@ -26,6 +26,7 @@
 #include <cuda/std/__bit/countr.h>
 #include <cuda/std/__cccl/assert.h>
 #include <cuda/std/__cstddef/types.h>
+#include <cuda/std/__host_stdlib/stdexcept>
 #include <cuda/std/__utility/forward.h>
 
 #include <cuda/experimental/__cuco/hash_functions.cuh>
@@ -34,8 +35,6 @@
 #include <cuda/experimental/memory_resource.cuh>
 #include <cuda/experimental/stream.cuh>
 
-#include <stdexcept>
-
 #include <cuda/std/__cccl/prologue.h>
 
 namespace cuda::experimental::cuco

cudax/include/cuda/experimental/__cufile/exception.cuh

@@ -21,10 +21,10 @@
 
 #include <cuda/std/__exception/cuda_error.h>
 #include <cuda/std/__exception/terminate.h>
+#include <cuda/std/__host_stdlib/stdexcept>
 #include <cuda/std/source_location>
 
 #include <cstdio>
-#include <stdexcept>
 
 #include <cufile.h>
 

libcudacxx/include/cuda/__device/physical_device.h

@@ -26,8 +26,8 @@
 #  include <cuda/__device/device_ref.h>
 #  include <cuda/__driver/driver_api.h>
 #  include <cuda/__fwd/devices.h>
-#  include <cuda/std/__cccl/memory_wrapper.h>
 #  include <cuda/std/__cstddef/types.h>
+#  include <cuda/std/__host_stdlib/memory>
 #  include <cuda/std/span>
 #  include <cuda/std/string_view>
 

libcudacxx/include/cuda/__driver/driver_api.h

@@ -25,6 +25,7 @@
 
 #  include <cuda/std/__cstddef/types.h>
 #  include <cuda/std/__exception/cuda_error.h>
+#  include <cuda/std/__host_stdlib/stdexcept>
 #  include <cuda/std/__internal/namespaces.h>
 #  include <cuda/std/__limits/numeric_limits.h>
 #  include <cuda/std/__type_traits/always_false.h>
@@ -35,8 +36,6 @@
 #    include <dlfcn.h>
 #  endif
 
-#  include <stdexcept>
-
 #  include <cuda.h>
 
 #  include <cuda/std/__cccl/prologue.h>

libcudacxx/include/cuda/__mdspan/dlpack_to_mdspan.h

@@ -29,16 +29,13 @@
 #  include <cuda/__memory/is_aligned.h>
 #  include <cuda/std/__cstddef/types.h>
 #  include <cuda/std/__exception/exception_macros.h>
+#  include <cuda/std/__host_stdlib/stdexcept>
 #  include <cuda/std/__type_traits/is_same.h>
 #  include <cuda/std/__utility/cmp.h>
 #  include <cuda/std/array>
 #  include <cuda/std/cstdint>
 #  include <cuda/std/mdspan>
 
-#  if !_CCCL_COMPILER(NVRTC)
-#    include <stdexcept>
-#  endif // !_CCCL_COMPILER(NVRTC)
-
 #  include <dlpack/dlpack.h>
 //
 #  include <cuda/std/__cccl/prologue.h>

libcudacxx/include/cuda/__mdspan/mdspan_to_dlpack.h

@@ -30,6 +30,7 @@
 #  include <cuda/std/__cstddef/types.h>
 #  include <cuda/std/__exception/exception_macros.h>
 #  include <cuda/std/__fwd/complex.h>
+#  include <cuda/std/__host_stdlib/stdexcept>
 #  include <cuda/std/__limits/numeric_limits.h>
 #  include <cuda/std/__type_traits/always_false.h>
 #  include <cuda/std/__type_traits/is_pointer.h>
@@ -41,8 +42,6 @@
 #  include <cuda/std/cstdint>
 #  include <cuda/std/mdspan>
 
-#  include <stdexcept>
-
 #  include <cuda/std/__cccl/prologue.h>
 
 _CCCL_BEGIN_NAMESPACE_CUDA