pinned_memory_resource.hpp

Go to the documentation of this file.

/*
 * Copyright (c) 2020-2025, NVIDIA CORPORATION.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
#pragma once
 
#include <rmm/aligned.hpp>
#include <rmm/cuda_stream_view.hpp>
#include <rmm/detail/aligned.hpp>
#include <rmm/detail/error.hpp>
#include <rmm/detail/export.hpp>
#include <rmm/mr/host/host_memory_resource.hpp>
 
#include <cstddef>
 
namespace RMM_NAMESPACE {
namespace mr {
class pinned_memory_resource final : public host_memory_resource {
 public:
  pinned_memory_resource()                              = default;
  ~pinned_memory_resource() override                    = default;
  pinned_memory_resource(pinned_memory_resource const&) = default;  
  pinned_memory_resource(pinned_memory_resource&&)      = default;  
  pinned_memory_resource& operator=(pinned_memory_resource const&) =
    default;  
  pinned_memory_resource& operator=(pinned_memory_resource&&) =
    default;  
 
  [[nodiscard]] void* allocate_async(std::size_t bytes, std::size_t alignment, cuda_stream_view)
  {
    return do_allocate(bytes, alignment);
  }
 
  [[nodiscard]] void* allocate_async(std::size_t bytes, cuda_stream_view)
  {
    return do_allocate(bytes);
  }
 
  void deallocate_async(void* ptr,
                        std::size_t bytes,
                        std::size_t alignment,
                        cuda_stream_view) noexcept
  {
    do_deallocate(ptr, rmm::align_up(bytes, alignment));
  }
 
  friend void get_property(pinned_memory_resource const&, cuda::mr::device_accessible) noexcept {}
 
 private:
  void* do_allocate(std::size_t bytes, std::size_t alignment = alignof(std::max_align_t)) override
  {
    // don't allocate anything if the user requested zero bytes
    if (0 == bytes) { return nullptr; }
 
    // If the requested alignment isn't supported, use default
    alignment =
      (rmm::is_supported_alignment(alignment)) ? alignment : rmm::RMM_DEFAULT_HOST_ALIGNMENT;
 
    return rmm::detail::aligned_host_allocate(bytes, alignment, [](std::size_t size) {
      void* ptr{nullptr};
      RMM_CUDA_TRY_ALLOC(cudaMallocHost(&ptr, size), size);
      return ptr;
    });
  }
 
  void do_deallocate(void* ptr,
                     std::size_t bytes,
                     std::size_t alignment = alignof(std::max_align_t)) noexcept override
  {
    if (nullptr == ptr) { return; }
    rmm::detail::aligned_host_deallocate(
      ptr, bytes, alignment, [](void* ptr) { RMM_ASSERT_CUDA_SUCCESS(cudaFreeHost(ptr)); });
  }
 
#if CCCL_MAJOR_VERSION > 3 || (CCCL_MAJOR_VERSION == 3 && CCCL_MINOR_VERSION >= 1)
 
 public:
  // Explicitly inherit the allocate and deallocate functions from the host_memory_resource class.
  // Due to inheritance and name hiding rules, we need to declare these with "using" when we
  // override allocate and deallocate for CCCL 3.1.0+ compatibility.
  using host_memory_resource::allocate;
  using host_memory_resource::deallocate;
 
  void* allocate(cuda_stream_view stream, std::size_t bytes, std::size_t alignment)
  {
    return this->allocate_async(bytes, alignment, stream);
  }
 
  void deallocate(cuda_stream_view stream,
                  void* ptr,
                  std::size_t bytes,
                  std::size_t alignment) noexcept
  {
    return this->deallocate_async(ptr, bytes, alignment, stream);
  }
 
#endif
};
 
// static property checks
static_assert(rmm::detail::polyfill::async_resource_with<pinned_memory_resource,
                                                         cuda::mr::host_accessible,
                                                         cuda::mr::device_accessible>);
  // end of group
}  // namespace mr
}  // namespace RMM_NAMESPACE