statistics_resource_adaptor.hpp

Go to the documentation of this file.

/*
 * SPDX-FileCopyrightText: Copyright (c) 2020-2025, NVIDIA CORPORATION.
 * SPDX-License-Identifier: Apache-2.0
 */
#pragma once
 
#include <rmm/detail/export.hpp>
#include <rmm/mr/device/device_memory_resource.hpp>
#include <rmm/mr/device/per_device_resource.hpp>
#include <rmm/resource_ref.hpp>
 
#include <cstddef>
#include <mutex>
#include <shared_mutex>
#include <stack>
 
namespace RMM_NAMESPACE {
namespace mr {
template <typename Upstream>
class statistics_resource_adaptor final : public device_memory_resource {
 public:
  using read_lock_t =
    std::shared_lock<std::shared_mutex>;  
  using write_lock_t =
    std::unique_lock<std::shared_mutex>;  
  struct counter {
    int64_t value{0};  
    int64_t peak{0};   
    int64_t total{0};  
 
    counter& operator+=(int64_t val)
    {
      value += val;
      total += val;
      peak = std::max(value, peak);
      return *this;
    }
 
    counter& operator-=(int64_t val)
    {
      value -= val;
      return *this;
    }
 
    void add_counters_from_tracked_sub_block(const counter& val)
    {
      peak = std::max(value + val.peak, peak);
      value += val.value;
      total += val.total;
    }
  };
 
  statistics_resource_adaptor(device_async_resource_ref upstream) : upstream_{upstream} {}
 
  statistics_resource_adaptor(Upstream* upstream)
    : upstream_{to_device_async_resource_ref_checked(upstream)}
  {
  }
 
  statistics_resource_adaptor()                                              = delete;
  ~statistics_resource_adaptor() override                                    = default;
  statistics_resource_adaptor(statistics_resource_adaptor const&)            = delete;
  statistics_resource_adaptor& operator=(statistics_resource_adaptor const&) = delete;
  statistics_resource_adaptor(statistics_resource_adaptor&&) noexcept =
    default;  
  statistics_resource_adaptor& operator=(statistics_resource_adaptor&&) noexcept =
    default;  
 
  [[nodiscard]] rmm::device_async_resource_ref get_upstream_resource() const noexcept
  {
    return upstream_;
  }
 
  counter get_bytes_counter() const noexcept
  {
    read_lock_t lock(mtx_);
 
    return counter_stack_.top().first;
  }
 
  counter get_allocations_counter() const noexcept
  {
    read_lock_t lock(mtx_);
 
    return counter_stack_.top().second;
  }
 
  std::pair<counter, counter> push_counters()
  {
    write_lock_t lock(mtx_);
    auto ret = counter_stack_.top();
    counter_stack_.push({counter{}, counter{}});
    return ret;
  }
 
  std::pair<counter, counter> pop_counters()
  {
    write_lock_t lock(mtx_);
    if (counter_stack_.size() < 2) { throw std::out_of_range("cannot pop the last counter pair"); }
    auto ret = counter_stack_.top();
    counter_stack_.pop();
    // Update the new top pair of counters
    counter_stack_.top().first.add_counters_from_tracked_sub_block(ret.first);
    counter_stack_.top().second.add_counters_from_tracked_sub_block(ret.second);
    return ret;
  }
 
 private:
  void* do_allocate(std::size_t bytes, cuda_stream_view stream) override
  {
    void* ptr = get_upstream_resource().allocate(stream, bytes);
 
    // increment the stats
    {
      write_lock_t lock(mtx_);
 
      // Increment the allocation_count_ while we have the lock
      counter_stack_.top().first += bytes;
      counter_stack_.top().second += 1;
    }
 
    return ptr;
  }
 
  void do_deallocate(void* ptr, std::size_t bytes, cuda_stream_view stream) noexcept override
  {
    get_upstream_resource().deallocate(stream, ptr, bytes);
 
    {
      write_lock_t lock(mtx_);
 
      // Decrement the current allocated counts.
      counter_stack_.top().first -= bytes;
      counter_stack_.top().second -= 1;
    }
  }
 
  bool do_is_equal(device_memory_resource const& other) const noexcept override
  {
    if (this == &other) { return true; }
    auto cast = dynamic_cast<statistics_resource_adaptor<Upstream> const*>(&other);
    if (cast == nullptr) { return false; }
    return get_upstream_resource() == cast->get_upstream_resource();
  }
 
  // Stack of counter pairs <bytes, allocations>
  // Invariant: the stack always contains at least one entry
  std::stack<std::pair<counter, counter>> counter_stack_{{std::make_pair(counter{}, counter{})}};
  std::shared_mutex mutable mtx_;  // mutex for thread safe access to allocations_
  // the upstream resource used for satisfying allocation requests
  device_async_resource_ref upstream_;
};
  // end of group
}  // namespace mr
}  // namespace RMM_NAMESPACE