cub/block/block_reduce.cuh

File members: cub/block/block_reduce.cuh

/******************************************************************************
 * Copyright (c) 2011, Duane Merrill.  All rights reserved.
 * Copyright (c) 2011-2018, 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/specializations/block_reduce_raking.cuh>
#include <cub/block/specializations/block_reduce_raking_commutative_only.cuh>
#include <cub/block/specializations/block_reduce_warp_reductions.cuh>
#include <cub/thread/thread_operators.cuh>
#include <cub/util_ptx.cuh>
#include <cub/util_type.cuh>

#include <cuda/std/type_traits>

CUB_NAMESPACE_BEGIN

/******************************************************************************
 * Algorithmic variants
 ******************************************************************************/

enum BlockReduceAlgorithm
{

  BLOCK_REDUCE_RAKING_COMMUTATIVE_ONLY,

  BLOCK_REDUCE_RAKING,

  BLOCK_REDUCE_WARP_REDUCTIONS,
};

template <typename T,
          int BLOCK_DIM_X,
          BlockReduceAlgorithm ALGORITHM = BLOCK_REDUCE_WARP_REDUCTIONS,
          int BLOCK_DIM_Y                = 1,
          int BLOCK_DIM_Z                = 1,
          int LEGACY_PTX_ARCH            = 0>
class BlockReduce
{
private:
  enum
  {
    BLOCK_THREADS = BLOCK_DIM_X * BLOCK_DIM_Y * BLOCK_DIM_Z,
  };

  using WarpReductions        = BlockReduceWarpReductions<T, BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z>;
  using RakingCommutativeOnly = BlockReduceRakingCommutativeOnly<T, BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z>;
  using Raking                = BlockReduceRaking<T, BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z>;

  using InternalBlockReduce =
    ::cuda::std::_If<ALGORITHM == BLOCK_REDUCE_WARP_REDUCTIONS,
                     WarpReductions,
                     ::cuda::std::_If<ALGORITHM == BLOCK_REDUCE_RAKING_COMMUTATIVE_ONLY,
                                      RakingCommutativeOnly,
                                      Raking>>; // BlockReduceRaking

  using _TempStorage = typename InternalBlockReduce::TempStorage;

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

  _TempStorage& temp_storage;

  unsigned int linear_tid;

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

  _CCCL_DEVICE _CCCL_FORCEINLINE BlockReduce()
      : temp_storage(PrivateStorage())
      , linear_tid(RowMajorTid(BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z))
  {}

  _CCCL_DEVICE _CCCL_FORCEINLINE BlockReduce(TempStorage& temp_storage)
      : temp_storage(temp_storage.Alias())
      , linear_tid(RowMajorTid(BLOCK_DIM_X, BLOCK_DIM_Y, BLOCK_DIM_Z))
  {}

  template <typename ReductionOp>
  _CCCL_DEVICE _CCCL_FORCEINLINE T Reduce(T input, ReductionOp reduction_op)
  {
    return InternalBlockReduce(temp_storage).template Reduce<true>(input, BLOCK_THREADS, reduction_op);
  }

  template <int ITEMS_PER_THREAD, typename ReductionOp>
  _CCCL_DEVICE _CCCL_FORCEINLINE T Reduce(T (&inputs)[ITEMS_PER_THREAD], ReductionOp reduction_op)
  {
    // Reduce partials
    T partial = cub::ThreadReduce(inputs, reduction_op);
    return Reduce(partial, reduction_op);
  }

  template <typename ReductionOp>
  _CCCL_DEVICE _CCCL_FORCEINLINE T Reduce(T input, ReductionOp reduction_op, int num_valid)
  {
    // Determine if we skip bounds checking
    if (num_valid >= BLOCK_THREADS)
    {
      return InternalBlockReduce(temp_storage).template Reduce<true>(input, num_valid, reduction_op);
    }
    else
    {
      return InternalBlockReduce(temp_storage).template Reduce<false>(input, num_valid, reduction_op);
    }
  }

  _CCCL_DEVICE _CCCL_FORCEINLINE T Sum(T input)
  {
    return InternalBlockReduce(temp_storage).template Sum<true>(input, BLOCK_THREADS);
  }

  template <int ITEMS_PER_THREAD>
  _CCCL_DEVICE _CCCL_FORCEINLINE T Sum(T (&inputs)[ITEMS_PER_THREAD])
  {
    // Reduce partials
    T partial = cub::ThreadReduce(inputs, ::cuda::std::plus<>{});
    return Sum(partial);
  }

  _CCCL_DEVICE _CCCL_FORCEINLINE T Sum(T input, int num_valid)
  {
    // Determine if we skip bounds checking
    if (num_valid >= BLOCK_THREADS)
    {
      return InternalBlockReduce(temp_storage).template Sum<true>(input, num_valid);
    }
    else
    {
      return InternalBlockReduce(temp_storage).template Sum<false>(input, num_valid);
    }
  }

};

CUB_NAMESPACE_END