cub/warp/warp_store.cuh

File members: cub/warp/warp_store.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_store.cuh>
#include <cub/util_ptx.cuh>
#include <cub/util_type.cuh>
#include <cub/warp/warp_exchange.cuh>

CUB_NAMESPACE_BEGIN

enum WarpStoreAlgorithm
{
  WARP_STORE_DIRECT,

  WARP_STORE_STRIPED,

  WARP_STORE_VECTORIZE,

  WARP_STORE_TRANSPOSE
};

template <typename T,
          int ITEMS_PER_THREAD,
          WarpStoreAlgorithm ALGORITHM = WARP_STORE_DIRECT,
          int LOGICAL_WARP_THREADS     = CUB_PTX_WARP_THREADS,
          int LEGACY_PTX_ARCH          = 0>
class WarpStore
{
  static_assert(PowerOfTwo<LOGICAL_WARP_THREADS>::VALUE, "LOGICAL_WARP_THREADS must be a power of two");

  static constexpr bool IS_ARCH_WARP = LOGICAL_WARP_THREADS == CUB_WARP_THREADS(0);

private:
  template <WarpStoreAlgorithm _POLICY, int DUMMY>
  struct StoreInternal;

  template <int DUMMY>
  struct StoreInternal<WARP_STORE_DIRECT, DUMMY>
  {
    using TempStorage = NullType;

    int linear_tid;

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

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD])
    {
      StoreDirectBlocked(linear_tid, block_itr, items);
    }

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      StoreDirectBlocked(linear_tid, block_itr, items, valid_items);
    }
  };

  template <int DUMMY>
  struct StoreInternal<WARP_STORE_STRIPED, DUMMY>
  {
    using TempStorage = NullType;

    int linear_tid;

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

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD])
    {
      StoreDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items);
    }

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      StoreDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items, valid_items);
    }
  };

  template <int DUMMY>
  struct StoreInternal<WARP_STORE_VECTORIZE, DUMMY>
  {
    using TempStorage = NullType;

    int linear_tid;

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

    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(T* block_ptr, T (&items)[ITEMS_PER_THREAD])
    {
      StoreDirectBlockedVectorized(linear_tid, block_ptr, items);
    }

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD])
    {
      StoreDirectBlocked(linear_tid, block_itr, items);
    }

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      StoreDirectBlocked(linear_tid, block_itr, items, valid_items);
    }
  };

  template <int DUMMY>
  struct StoreInternal<WARP_STORE_TRANSPOSE, DUMMY>
  {
    using WarpExchangeT = WarpExchange<T, ITEMS_PER_THREAD, LOGICAL_WARP_THREADS>;

    struct _TempStorage : WarpExchangeT::TempStorage
    {};

    struct TempStorage : Uninitialized<_TempStorage>
    {};

    _TempStorage& temp_storage;

    int linear_tid;

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

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD])
    {
      WarpExchangeT(temp_storage).BlockedToStriped(items, items);
      StoreDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items);
    }

    template <typename OutputIteratorT>
    _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD], int valid_items)
    {
      WarpExchangeT(temp_storage).BlockedToStriped(items, items);
      StoreDirectStriped<LOGICAL_WARP_THREADS>(linear_tid, block_itr, items, valid_items);
    }
  };

  using InternalStore = StoreInternal<ALGORITHM, 0>;

  using _TempStorage = typename InternalStore::TempStorage;

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

  _TempStorage& temp_storage;

  int linear_tid;

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

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

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

  template <typename OutputIteratorT>
  _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD])
  {
    InternalStore(temp_storage, linear_tid).Store(block_itr, items);
  }

  template <typename OutputIteratorT>
  _CCCL_DEVICE _CCCL_FORCEINLINE void Store(OutputIteratorT block_itr, T (&items)[ITEMS_PER_THREAD], int valid_items)
  {
    InternalStore(temp_storage, linear_tid).Store(block_itr, items, valid_items);
  }

};

CUB_NAMESPACE_END