device_memory_resource.hpp

Go to the documentation of this file.

/*
 * Copyright (c) 2019-2025, 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/cuda_stream_view.hpp>
#include <rmm/detail/aligned.hpp>
#include <rmm/detail/cccl_adaptors.hpp>
#include <rmm/detail/cuda_memory_resource.hpp>
#include <rmm/detail/error.hpp>
#include <rmm/detail/export.hpp>
#include <rmm/detail/nvtx/ranges.hpp>
 
#include <cstddef>
 
namespace RMM_NAMESPACE {
namespace mr {
class device_memory_resource {
 public:
  device_memory_resource()                              = default;
  virtual ~device_memory_resource()                     = default;
  device_memory_resource(device_memory_resource const&) = default;  
  device_memory_resource(device_memory_resource&&) noexcept =
    default;  
  device_memory_resource& operator=(device_memory_resource const&) =
    default;  
  device_memory_resource& operator=(device_memory_resource&&) noexcept =
    default;  
 
  void* allocate(std::size_t bytes, cuda_stream_view stream = cuda_stream_view{})
  {
    RMM_FUNC_RANGE();
    return do_allocate(bytes, stream);
  }
 
  void deallocate(void* ptr, std::size_t bytes, cuda_stream_view stream = cuda_stream_view{})
  {
    RMM_FUNC_RANGE();
    do_deallocate(ptr, bytes, stream);
  }
 
  [[nodiscard]] bool is_equal(device_memory_resource const& other) const noexcept
  {
    return do_is_equal(other);
  }
 
  void* allocate(std::size_t bytes, std::size_t alignment)
  {
    RMM_FUNC_RANGE();
    return do_allocate(rmm::align_up(bytes, alignment), cuda_stream_view{});
  }
 
  void deallocate(void* ptr, std::size_t bytes, std::size_t alignment)
  {
    RMM_FUNC_RANGE();
    do_deallocate(ptr, rmm::align_up(bytes, alignment), cuda_stream_view{});
  }
 
  void* allocate_async(std::size_t bytes, std::size_t alignment, cuda_stream_view stream)
  {
    RMM_FUNC_RANGE();
    return do_allocate(rmm::align_up(bytes, alignment), stream);
  }
 
  void* allocate_async(std::size_t bytes, cuda_stream_view stream)
  {
    RMM_FUNC_RANGE();
    return do_allocate(bytes, stream);
  }
 
  void deallocate_async(void* ptr,
                        std::size_t bytes,
                        std::size_t alignment,
                        cuda_stream_view stream)
  {
    RMM_FUNC_RANGE();
    do_deallocate(ptr, rmm::align_up(bytes, alignment), stream);
  }
 
  void deallocate_async(void* ptr, std::size_t bytes, cuda_stream_view stream)
  {
    RMM_FUNC_RANGE();
    do_deallocate(ptr, bytes, stream);
  }
 
#if CCCL_MAJOR_VERSION > 3 || (CCCL_MAJOR_VERSION == 3 && CCCL_MINOR_VERSION >= 1)
  // CCCL >= 3.1 needs a different set of methods to satisfy the memory resource concepts
 
  void* allocate_sync(std::size_t bytes, std::size_t alignment = rmm::CUDA_ALLOCATION_ALIGNMENT)
  {
    RMM_EXPECTS(alignment <= 256 && rmm::is_supported_alignment(alignment),
                "Alignment must be less than or equal to 256 and a power of two");
    return do_allocate(rmm::align_up(bytes, alignment), cuda_stream_view{});
  }
 
  void deallocate_sync(void* ptr,
                       std::size_t bytes,
                       std::size_t alignment = rmm::CUDA_ALLOCATION_ALIGNMENT) noexcept
  {
    do_deallocate(ptr, rmm::align_up(bytes, alignment), cuda_stream_view{});
  }
 
  void* allocate(cuda_stream_view stream,
                 std::size_t bytes,
                 std::size_t alignment = rmm::CUDA_ALLOCATION_ALIGNMENT)
  {
    RMM_EXPECTS(alignment <= 256 && rmm::is_supported_alignment(alignment),
                "Alignment must be less than or equal to 256 and a power of two");
    return do_allocate(rmm::align_up(bytes, alignment), stream);
  }
 
  void deallocate(cuda_stream_view stream,
                  void* ptr,
                  std::size_t bytes,
                  std::size_t alignment = rmm::CUDA_ALLOCATION_ALIGNMENT) noexcept
  {
    do_deallocate(ptr, rmm::align_up(bytes, alignment), stream);
  }
#endif  // CCCL >= 3.1
 
  [[nodiscard]] bool operator==(device_memory_resource const& other) const noexcept
  {
    return do_is_equal(other);
  }
 
  [[nodiscard]] bool operator!=(device_memory_resource const& other) const noexcept
  {
    return !do_is_equal(other);
  }
 
  friend void get_property(device_memory_resource const&, cuda::mr::device_accessible) noexcept {}
 
 private:
  virtual void* do_allocate(std::size_t bytes, cuda_stream_view stream) = 0;
 
  virtual void do_deallocate(void* ptr, std::size_t bytes, cuda_stream_view stream) = 0;
 
  [[nodiscard]] virtual bool do_is_equal(device_memory_resource const& other) const noexcept
  {
    return this == &other;
  }
};
 
// static property checks
static_assert(rmm::detail::polyfill::resource<device_memory_resource>);
static_assert(rmm::detail::polyfill::async_resource<device_memory_resource>);
static_assert(
  rmm::detail::polyfill::resource_with<device_memory_resource, cuda::mr::device_accessible>);
static_assert(
  rmm::detail::polyfill::async_resource_with<device_memory_resource, cuda::mr::device_accessible>);
  // end of group
}  // namespace mr
}  // namespace RMM_NAMESPACE