cub/warp/warp_load.cuh

File members: cub/warp/warp_load.cuh

/******************************************************************************
 * Copyright (c) 2011-2021, NVIDIA CORPORATION.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright
 *       notice, this list of conditions and the following disclaimer in the
 *       documentation and/or other materials provided with the distribution.
 *     * Neither the name of the NVIDIA CORPORATION nor the
 *       names of its contributors may be used to endorse or promote products
 *       derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE FOR ANY
 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 ******************************************************************************/

#pragma once

#include <cub/config.cuh>

#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 <cub/block/block_load.cuh>
#include <cub/iterator/cache_modified_input_iterator.cuh>
#include <cub/util_ptx.cuh>
#include <cub/util_type.cuh>
#include <cub/warp/warp_exchange.cuh>

CUB_NAMESPACE_BEGIN

enum WarpLoadAlgorithm
{
  WARP_LOAD_DIRECT,

  WARP_LOAD_STRIPED,

  WARP_LOAD_VECTORIZE,

  WARP_LOAD_TRANSPOSE
};

template <typename InputT,
          int ITEMS_PER_THREAD,
          WarpLoadAlgorithm ALGORITHM = WARP_LOAD_DIRECT,
          int LOGICAL_WARP_THREADS    = CUB_PTX_WARP_THREADS,
          int LEGACY_PTX_ARCH         = 0>
class WarpLoad
{
  static constexpr bool IS_ARCH_WARP = LOGICAL_WARP_THREADS == CUB_WARP_THREADS(0);

  static_assert(PowerOfTwo<LOGICAL_WARP_THREADS>::VALUE, "LOGICAL_WARP_THREADS must be a power of two");

private:
  /*****************************************************************************
   * Algorithmic variants
   ****************************************************************************/

  template <WarpLoadAlgorithm _POLICY, int DUMMY>
  struct LoadInternal;

  template <int DUMMY>
  struct LoadInternal<WARP_LOAD_DIRECT, DUMMY>
  {
    using TempStorage = NullType;

    int linear_tid;

    _CCCL_DEVICE _CCCL_FORCEINLINE LoadInternal(TempStorage& /*temp_storage*/, int linear_tid)
        : linear_tid(linear_tid)
    {}

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD])
    {
      LoadDirectBlocked(linear_tid, block_itr, items);
    }

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      LoadDirectBlocked(linear_tid, block_itr, items, valid_items);
    }

    template <typename InputIteratorT, typename DefaultT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items, DefaultT oob_default)
    {
      LoadDirectBlocked(linear_tid, block_itr, items, valid_items, oob_default);
    }
  };

  template <int DUMMY>
  struct LoadInternal<WARP_LOAD_STRIPED, DUMMY>
  {
    using TempStorage = NullType;

    int linear_tid;

    _CCCL_DEVICE _CCCL_FORCEINLINE LoadInternal(TempStorage& /*temp_storage*/, int linear_tid)
        : linear_tid(linear_tid)
    {}

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD])
    {
      LoadDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items);
    }

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      LoadDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items, valid_items);
    }

    template <typename InputIteratorT, typename DefaultT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items, DefaultT oob_default)
    {
      LoadDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items, valid_items, oob_default);
    }
  };

  template <int DUMMY>
  struct LoadInternal<WARP_LOAD_VECTORIZE, DUMMY>
  {
    using TempStorage = NullType;

    int linear_tid;

    _CCCL_DEVICE _CCCL_FORCEINLINE LoadInternal(TempStorage& /*temp_storage*/, int linear_tid)
        : linear_tid(linear_tid)
    {}

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Load(InputT* block_ptr, InputT (&items)[ITEMS_PER_THREAD])
    {
      InternalLoadDirectBlockedVectorized<LOAD_DEFAULT>(linear_tid, block_ptr, items);
    }

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Load(const InputT* block_ptr, InputT (&items)[ITEMS_PER_THREAD])
    {
      InternalLoadDirectBlockedVectorized<LOAD_DEFAULT>(linear_tid, block_ptr, items);
    }

    template <CacheLoadModifier MODIFIER, typename ValueType, typename OffsetT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(CacheModifiedInputIterator<MODIFIER, ValueType, OffsetT> block_itr, InputT (&items)[ITEMS_PER_THREAD])
    {
      InternalLoadDirectBlockedVectorized<MODIFIER>(linear_tid, block_itr.ptr, items);
    }

    template <typename _InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Load(_InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD])
    {
      LoadDirectBlocked(linear_tid, block_itr, items);
    }

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      LoadDirectBlocked(linear_tid, block_itr, items, valid_items);
    }

    template <typename InputIteratorT, typename DefaultT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items, DefaultT oob_default)
    {
      LoadDirectBlocked(linear_tid, block_itr, items, valid_items, oob_default);
    }
  };

  template <int DUMMY>
  struct LoadInternal<WARP_LOAD_TRANSPOSE, DUMMY>
  {
    using WarpExchangeT = WarpExchange<InputT, ITEMS_PER_THREAD, LOGICAL_WARP_THREADS>;

    struct _TempStorage : WarpExchangeT::TempStorage
    {};

    struct TempStorage : Uninitialized<_TempStorage>
    {};

    _TempStorage& temp_storage;

    int linear_tid;

    _CCCL_DEVICE _CCCL_FORCEINLINE LoadInternal(TempStorage& temp_storage, int linear_tid)
        : temp_storage(temp_storage.Alias())
        , linear_tid(linear_tid)
    {}

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD])
    {
      LoadDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items);
      WarpExchangeT(temp_storage).StripedToBlocked(items, items);
    }

    template <typename InputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      LoadDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items, valid_items);
      WarpExchangeT(temp_storage).StripedToBlocked(items, items);
    }

    template <typename InputIteratorT, typename DefaultT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void
    Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items, DefaultT oob_default)
    {
      LoadDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items, valid_items, oob_default);
      WarpExchangeT(temp_storage).StripedToBlocked(items, items);
    }
  };

  /*****************************************************************************
   * Type definitions
   ****************************************************************************/

  using InternalLoad = LoadInternal<ALGORITHM, 0>;

  using _TempStorage = typename InternalLoad::TempStorage;

  /*****************************************************************************
   * Utility methods
   ****************************************************************************/

  _CCCL_DEVICE _CCCL_FORCEINLINE _TempStorage& PrivateStorage()
  {
    __shared__ _TempStorage private_storage;
    return private_storage;
  }

  /*****************************************************************************
   * Thread fields
   ****************************************************************************/

  _TempStorage& temp_storage;

  int linear_tid;

public:
  struct TempStorage : Uninitialized<_TempStorage>
  {};

  _CCCL_DEVICE _CCCL_FORCEINLINE WarpLoad()
      : temp_storage(PrivateStorage())
      , linear_tid(IS_ARCH_WARP ? LaneId() : (LaneId() % LOGICAL_WARP_THREADS))
  {}

  _CCCL_DEVICE _CCCL_FORCEINLINE WarpLoad(TempStorage& temp_storage)
      : temp_storage(temp_storage.Alias())
      , linear_tid(IS_ARCH_WARP ? LaneId() : (LaneId() % LOGICAL_WARP_THREADS))
  {}

  template <typename InputIteratorT>
  _CCCL_DEVICE _CCCL_FORCEINLINE void Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD])
  {
    InternalLoad(temp_storage, linear_tid).Load(block_itr, items);
  }

  template <typename InputIteratorT>
  _CCCL_DEVICE _CCCL_FORCEINLINE void Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items)
  {
    InternalLoad(temp_storage, linear_tid).Load(block_itr, items, valid_items);
  }

  template <typename InputIteratorT, typename DefaultT>
  _CCCL_DEVICE _CCCL_FORCEINLINE void
  Load(InputIteratorT block_itr, InputT (&items)[ITEMS_PER_THREAD], int valid_items, DefaultT oob_default)
  {
    InternalLoad(temp_storage, linear_tid).Load(block_itr, items, valid_items, oob_default);
  }

};

CUB_NAMESPACE_END