new type-erased memory resources (NVIDIA#2824)

pciolkosz · Dec 17, 2024 · 82cff38 · 82cff38
1 parent 7d057c4
commit 82cff38
Show file tree

Hide file tree

Showing 23 changed files with 1,293 additions and 450 deletions.
diff --git a/cudax/include/cuda/experimental/__container/uninitialized_async_buffer.cuh b/cudax/include/cuda/experimental/__container/uninitialized_async_buffer.cuh
@@ -22,7 +22,6 @@
 #endif // no system header
 
 #include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource_ref.h>
 #include <cuda/std/__memory/align.h>
 #include <cuda/std/__new/launder.h>
 #include <cuda/std/__type_traits/type_set.h>
@@ -118,7 +117,7 @@ private:
   _CCCL_NODISCARD_FRIEND _CCCL_HIDE_FROM_ABI auto
   __cudax_launch_transform(::cuda::stream_ref, uninitialized_async_buffer& __self) noexcept
     _CCCL_TRAILING_REQUIRES(_CUDA_VSTD::span<_Tp>)(
-      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<mr::device_accessible, _Properties...>)
+      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<device_accessible, _Properties...>)
   {
     // TODO add auto synchronization
     return {__self.__get_data(), __self.size()};
@@ -130,7 +129,7 @@ private:
   _CCCL_NODISCARD_FRIEND _CCCL_HIDE_FROM_ABI auto
   __cudax_launch_transform(::cuda::stream_ref, const uninitialized_async_buffer& __self) noexcept
     _CCCL_TRAILING_REQUIRES(_CUDA_VSTD::span<const _Tp>)(
-      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<mr::device_accessible, _Properties...>)
+      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<device_accessible, _Properties...>)
   {
     // TODO add auto synchronization
     return {__self.__get_data(), __self.size()};
@@ -287,15 +286,15 @@ public:
   _CCCL_HIDE_FROM_ABI uninitialized_async_buffer __replace_allocation(const size_t __count)
   {
     // Create a new buffer with a reference to the stored memory resource and swap allocation information
-    uninitialized_async_buffer __ret{_CUDA_VMR::async_resource_ref<_Properties...>{__mr_}, __stream_, __count};
+    uninitialized_async_buffer __ret{async_resource_ref<_Properties...>{__mr_}, __stream_, __count};
     _CUDA_VSTD::swap(__count_, __ret.__count_);
     _CUDA_VSTD::swap(__buf_, __ret.__buf_);
     return __ret;
   }
 };
 
 template <class _Tp>
-using uninitialized_async_device_buffer = uninitialized_async_buffer<_Tp, mr::device_accessible>;
+using uninitialized_async_device_buffer = uninitialized_async_buffer<_Tp, device_accessible>;
 
 } // namespace cuda::experimental
 

diff --git a/cudax/include/cuda/experimental/__container/uninitialized_buffer.cuh b/cudax/include/cuda/experimental/__container/uninitialized_buffer.cuh
@@ -22,7 +22,6 @@
 #endif // no system header
 
 #include <cuda/__memory_resource/properties.h>
-#include <cuda/__memory_resource/resource_ref.h>
 #include <cuda/std/__memory/align.h>
 #include <cuda/std/__new/launder.h>
 #include <cuda/std/__type_traits/type_set.h>
@@ -108,7 +107,7 @@ private:
   _CCCL_NODISCARD_FRIEND _CCCL_HIDE_FROM_ABI auto
   __cudax_launch_transform(::cuda::stream_ref, uninitialized_buffer& __self) noexcept
     _CCCL_TRAILING_REQUIRES(_CUDA_VSTD::span<_Tp>)(
-      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<mr::device_accessible, _Properties...>)
+      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<device_accessible, _Properties...>)
   {
     return {__self.__get_data(), __self.size()};
   }
@@ -119,7 +118,7 @@ private:
   _CCCL_NODISCARD_FRIEND _CCCL_HIDE_FROM_ABI auto
   __cudax_launch_transform(::cuda::stream_ref, const uninitialized_buffer& __self) noexcept
     _CCCL_TRAILING_REQUIRES(_CUDA_VSTD::span<const _Tp>)(
-      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<mr::device_accessible, _Properties...>)
+      _CUDA_VSTD::same_as<_Tp, _Tp2>&& _CUDA_VSTD::__is_included_in_v<device_accessible, _Properties...>)
   {
     return {__self.__get_data(), __self.size()};
   }
@@ -252,15 +251,15 @@ public:
   _CCCL_HIDE_FROM_ABI uninitialized_buffer __replace_allocation(const size_t __count)
   {
     // Create a new buffer with a reference to the stored memory resource and swap allocation information
-    uninitialized_buffer __ret{_CUDA_VMR::resource_ref<_Properties...>{__mr_}, __count};
+    uninitialized_buffer __ret{resource_ref<_Properties...>{__mr_}, __count};
     _CUDA_VSTD::swap(__count_, __ret.__count_);
     _CUDA_VSTD::swap(__buf_, __ret.__buf_);
     return __ret;
   }
 };
 
 template <class _Tp>
-using uninitialized_device_buffer = uninitialized_buffer<_Tp, mr::device_accessible>;
+using uninitialized_device_buffer = uninitialized_buffer<_Tp, device_accessible>;
 
 } // namespace cuda::experimental