prefetch_resource_adaptor.hpp

Go to the documentation of this file.

/*
 * Copyright (c) 2024, 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/detail/export.hpp>
#include <rmm/mr/device/device_memory_resource.hpp>
#include <rmm/prefetch.hpp>
#include <rmm/resource_ref.hpp>
 
#include <cstddef>
#include <mutex>
#include <shared_mutex>
#include <stack>
 
namespace RMM_NAMESPACE {
namespace mr {
template <typename Upstream>
class prefetch_resource_adaptor final : public device_memory_resource {
 public:
  prefetch_resource_adaptor(device_async_resource_ref upstream) : upstream_{upstream} {}
 
  prefetch_resource_adaptor(Upstream* upstream)
    : upstream_{to_device_async_resource_ref_checked(upstream)}
  {
  }
 
  prefetch_resource_adaptor()                                            = delete;
  ~prefetch_resource_adaptor() override                                  = default;
  prefetch_resource_adaptor(prefetch_resource_adaptor const&)            = delete;
  prefetch_resource_adaptor& operator=(prefetch_resource_adaptor const&) = delete;
  prefetch_resource_adaptor(prefetch_resource_adaptor&&) noexcept =
    default;  
  prefetch_resource_adaptor& operator=(prefetch_resource_adaptor&&) noexcept =
    default;  
 
  [[nodiscard]] rmm::device_async_resource_ref get_upstream_resource() const noexcept
  {
    return upstream_;
  }
 
 private:
  void* do_allocate(std::size_t bytes, cuda_stream_view stream) override
  {
    void* ptr = get_upstream_resource().allocate_async(bytes, stream);
    rmm::prefetch(ptr, bytes, rmm::get_current_cuda_device(), stream);
    return ptr;
  }
 
  void do_deallocate(void* ptr, std::size_t bytes, cuda_stream_view stream) override
  {
    get_upstream_resource().deallocate_async(ptr, bytes, stream);
  }
 
  bool do_is_equal(device_memory_resource const& other) const noexcept override
  {
    if (this == &other) { return true; }
    auto cast = dynamic_cast<prefetch_resource_adaptor<Upstream> const*>(&other);
    if (cast == nullptr) { return false; }
    return get_upstream_resource() == cast->get_upstream_resource();
  }
 
  // the upstream resource used for satisfying allocation requests
  device_async_resource_ref upstream_;
};
  // end of group
}  // namespace mr
}  // namespace RMM_NAMESPACE