cuda_async_view_memory_resource.hpp

Go to the documentation of this file.

/*
 * Copyright (c) 2021-2025, 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_device.hpp>
#include <rmm/cuda_stream_view.hpp>
#include <rmm/detail/error.hpp>
#include <rmm/detail/export.hpp>
#include <rmm/detail/thrust_namespace.h>
#include <rmm/mr/device/device_memory_resource.hpp>
 
#include <cuda_runtime_api.h>
 
#include <cstddef>
 
namespace RMM_NAMESPACE {
namespace mr {
class cuda_async_view_memory_resource final : public device_memory_resource {
 public:
  cuda_async_view_memory_resource(cudaMemPool_t valid_pool_handle)
    : cuda_pool_handle_{[valid_pool_handle]() {
        RMM_EXPECTS(nullptr != valid_pool_handle, "Unexpected null pool handle.");
        return valid_pool_handle;
      }()}
  {
    // Check if cudaMallocAsync Memory pool supported
    auto const device = rmm::get_current_cuda_device();
    int cuda_pool_supported{};
    auto result =
      cudaDeviceGetAttribute(&cuda_pool_supported, cudaDevAttrMemoryPoolsSupported, device.value());
    RMM_EXPECTS(result == cudaSuccess && cuda_pool_supported,
                "cudaMallocAsync not supported with this CUDA driver/runtime version");
  }
 
  [[nodiscard]] cudaMemPool_t pool_handle() const noexcept { return cuda_pool_handle_; }
 
  cuda_async_view_memory_resource() = default;
  cuda_async_view_memory_resource(cuda_async_view_memory_resource const&) =
    default;  
  cuda_async_view_memory_resource(cuda_async_view_memory_resource&&) =
    default;  
  cuda_async_view_memory_resource& operator=(cuda_async_view_memory_resource const&) =
    default;  
  cuda_async_view_memory_resource& operator=(cuda_async_view_memory_resource&&) =
    default;  
 
 private:
  cudaMemPool_t cuda_pool_handle_{};
 
  void* do_allocate(std::size_t bytes, rmm::cuda_stream_view stream) override
  {
    void* ptr{nullptr};
    if (bytes > 0) {
      RMM_CUDA_TRY_ALLOC(cudaMallocFromPoolAsync(&ptr, bytes, pool_handle(), stream.value()),
                         bytes);
    }
    return ptr;
  }
 
  void do_deallocate(void* ptr,
                     [[maybe_unused]] std::size_t bytes,
                     rmm::cuda_stream_view stream) override
  {
    if (ptr != nullptr) { RMM_ASSERT_CUDA_SUCCESS(cudaFreeAsync(ptr, stream.value())); }
  }
 
  [[nodiscard]] bool do_is_equal(device_memory_resource const& other) const noexcept override
  {
    return dynamic_cast<cuda_async_view_memory_resource const*>(&other) != nullptr;
  }
};
  // end of group
}  // namespace mr
}  // namespace RMM_NAMESPACE