project_6/cccl_upstream/thrust/thrust/mr/pool_options.h

// SPDX-FileCopyrightText: Copyright (c) 2018, NVIDIA Corporation. All rights reserved.
// SPDX-License-Identifier: Apache-2.0

/*! \file
 *  \brief A type used by the pooling resource adaptors to fine-tune their
 *  behavior.
 */

#pragma once

#include <thrust/detail/config.h>

#if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
#  pragma GCC system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_CLANG)
#  pragma clang system_header
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_MSVC)
#  pragma system_header
#endif // no system header

#include <thrust/detail/config/memory_resource.h>

#include <cuda/__cmath/pow2.h>
#include <cuda/__memory/is_valid_alignment.h>
#include <cuda/std/cstddef>

THRUST_NAMESPACE_BEGIN
namespace mr
{
/*! \addtogroup memory_resources Memory Resources
 *  \ingroup memory_management
 *  \{
 */

/*! A type used for configuring pooling resource adaptors, to fine-tune their behavior and parameters.
 */
struct pool_options
{
  /*! The minimal number of blocks, i.e. pieces of memory handed off to the user from a pool of a given size, in a
   * single chunk allocated from upstream.
   */
  std::size_t min_blocks_per_chunk;
  /*! The minimal number of bytes in a single chunk allocated from upstream.
   */
  std::size_t min_bytes_per_chunk;
  /*! The maximal number of blocks, i.e. pieces of memory handed off to the user from a pool of a given size, in a
   * single chunk allocated from upstream.
   */
  std::size_t max_blocks_per_chunk;
  /*! The maximal number of bytes in a single chunk allocated from upstream.
   */
  std::size_t max_bytes_per_chunk;

  /*! The size of blocks in the smallest pool covered by the pool resource. All allocation requests below this size will
   *      be rounded up to this size.
   */
  std::size_t smallest_block_size;
  /*! The size of blocks in the largest pool covered by the pool resource. All allocation requests above this size will
   *      be considered oversized, allocated directly from upstream (and not from a pool), and cached only of \p
   * cache_oversized is true.
   */
  std::size_t largest_block_size;

  /*! The alignment of all blocks in internal pools of the pool resource. All allocation requests above this alignment
   *      will be considered oversized, allocated directly from upstream (and not from a pool), and cached only of
   *      \p cache_oversized is true.
   */
  std::size_t alignment;

  /*! Decides whether oversized and overaligned blocks are cached for later use, or immediately return it to the
   * upstream resource.
   */
  bool cache_oversized;

  /*! The size factor at which a cached allocation is considered too ridiculously oversized to use to fulfill an
   * allocation request. For instance: the user requests an allocation of size 1024 bytes. A block of size 32 * 1024
   * bytes is cached. If \p cached_size_cutoff_factor is 32 or less, this block will be considered too big for that
   * allocation request.
   */
  std::size_t cached_size_cutoff_factor;
  /*! The alignment factor at which a cached allocation is considered too ridiculously overaligned to use to fulfill an
   *      allocation request. For instance: the user requests an allocation aligned to 32 bytes. A block aligned to 1024
   * bytes is cached. If \p cached_size_cutoff_factor is 32 or less, this block will be considered too overaligned for
   * that allocation request.
   */
  std::size_t cached_alignment_cutoff_factor;

  /*! Checks if the options are self-consistent.
   *
   *  /returns true if the options are self-consistent, false otherwise.
   */
  bool validate() const
  {
    if (smallest_block_size != 0 && !::cuda::is_power_of_two(smallest_block_size))
    {
      return false;
    }
    if (largest_block_size != 0 && !::cuda::is_power_of_two(largest_block_size))
    {
      return false;
    }

    // Forcing the alignment to be always valid would be a breaking change, because if the user didn't set the
    // alignment, this method would return false. So, we still accept 0.
    if (!::cuda::__is_valid_alignment(alignment) && alignment != 0)
    {
      return false;
    }

    if (max_bytes_per_chunk == 0 || max_blocks_per_chunk == 0)
    {
      return false;
    }
    if (smallest_block_size == 0 || largest_block_size == 0)
    {
      return false;
    }

    if (min_blocks_per_chunk > max_blocks_per_chunk)
    {
      return false;
    }
    if (min_bytes_per_chunk > max_bytes_per_chunk)
    {
      return false;
    }

    if (smallest_block_size > largest_block_size)
    {
      return false;
    }

    if (min_blocks_per_chunk * smallest_block_size > max_bytes_per_chunk)
    {
      return false;
    }
    if (min_blocks_per_chunk * largest_block_size > max_bytes_per_chunk)
    {
      return false;
    }

    if (max_blocks_per_chunk * largest_block_size < min_bytes_per_chunk)
    {
      return false;
    }
    if (max_blocks_per_chunk * smallest_block_size < min_bytes_per_chunk)
    {
      return false;
    }

    if (alignment > smallest_block_size)
    {
      return false;
    }

    return true;
  }
};

/*! \} // memory_resources
 */
} // namespace mr
THRUST_NAMESPACE_END