/home/runner/work/cccl/cccl/cub/cub/thread/thread_operators.cuh

File members: /home/runner/work/cccl/cccl/cub/cub/thread/thread_operators.cuh

/******************************************************************************
 * Copyright (c) 2011, Duane Merrill.  All rights reserved.
 * Copyright (c) 2011-2022, 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.
 *
 ******************************************************************************/

/******************************************************************************
 * Simple functor operators
 ******************************************************************************/

#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/util_cpp_dialect.cuh>
#include <cub/util_type.cuh>

_CCCL_SUPPRESS_DEPRECATED_PUSH
#include <cuda/std/functional>
_CCCL_SUPPRESS_DEPRECATED_POP
#include <cuda/std/type_traits>
#include <cuda/std/utility>

CUB_NAMESPACE_BEGIN

template <typename EqualityOp>
struct InequalityWrapper
{
  EqualityOp op;

  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE InequalityWrapper(EqualityOp op)
      : op(op)
  {}

  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE bool operator()(T&& t, U&& u)
  {
    return !op(::cuda::std::forward<T>(t), ::cuda::std::forward<U>(u));
  }
};

#if _CCCL_STD_VER > 2011
using Equality   = ::cuda::std::equal_to<>;
using Inequality = ::cuda::std::not_equal_to<>;
using Sum        = ::cuda::std::plus<>;
using Difference = ::cuda::std::minus<>;
using Division   = ::cuda::std::divides<>;
#else
struct Equality
{
  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE bool operator()(T&& t, U&& u) const
  {
    return ::cuda::std::forward<T>(t) == ::cuda::std::forward<U>(u);
  }
};

struct Inequality
{
  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE bool operator()(T&& t, U&& u) const
  {
    return ::cuda::std::forward<T>(t) != ::cuda::std::forward<U>(u);
  }
};

struct Sum
{
  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE auto operator()(T&& t, U&& u) const
    -> decltype(::cuda::std::forward<T>(t) + ::cuda::std::forward<U>(u))
  {
    return ::cuda::std::forward<T>(t) + ::cuda::std::forward<U>(u);
  }
};

struct Difference
{
  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE auto operator()(T&& t, U&& u) const
    -> decltype(::cuda::std::forward<T>(t) - ::cuda::std::forward<U>(u))
  {
    return ::cuda::std::forward<T>(t) - ::cuda::std::forward<U>(u);
  }
};

struct Division
{
  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE auto operator()(T&& t, U&& u) const
    -> decltype(::cuda::std::forward<T>(t) / ::cuda::std::forward<U>(u))
  {
    return ::cuda::std::forward<T>(t) / ::cuda::std::forward<U>(u);
  }
};
#endif

struct Max
{
  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE typename ::cuda::std::common_type<T, U>::type operator()(T&& t, U&& u) const
  {
    return CUB_MAX(t, u);
  }
};

struct ArgMax
{
  template <typename T, typename OffsetT>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE KeyValuePair<OffsetT, T>
  operator()(const KeyValuePair<OffsetT, T>& a, const KeyValuePair<OffsetT, T>& b) const
  {
    // Mooch BUG (device reduce argmax gk110 3.2 million random fp32)
    // return ((b.value > a.value) ||
    //         ((a.value == b.value) && (b.key < a.key)))
    //      ? b : a;

    if ((b.value > a.value) || ((a.value == b.value) && (b.key < a.key)))
    {
      return b;
    }

    return a;
  }
};

struct Min
{
  template <typename T, typename U>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE typename ::cuda::std::common_type<T, U>::type operator()(T&& t, U&& u) const
  {
    return CUB_MIN(t, u);
  }
};

struct ArgMin
{
  template <typename T, typename OffsetT>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE KeyValuePair<OffsetT, T>
  operator()(const KeyValuePair<OffsetT, T>& a, const KeyValuePair<OffsetT, T>& b) const
  {
    // Mooch BUG (device reduce argmax gk110 3.2 million random fp32)
    // return ((b.value < a.value) ||
    //         ((a.value == b.value) && (b.key < a.key)))
    //      ? b : a;

    if ((b.value < a.value) || ((a.value == b.value) && (b.key < a.key)))
    {
      return b;
    }

    return a;
  }
};

namespace detail
{
template <class OpT>
struct basic_binary_op_t
{
  static constexpr bool value = false;
};

template <>
struct basic_binary_op_t<Sum>
{
  static constexpr bool value = true;
};

template <>
struct basic_binary_op_t<Min>
{
  static constexpr bool value = true;
};

template <>
struct basic_binary_op_t<Max>
{
  static constexpr bool value = true;
};
} // namespace detail

template <typename B>
struct CastOp
{
  template <typename A>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE B operator()(A&& a) const
  {
    return (B) a;
  }
};

template <typename ScanOp>
class SwizzleScanOp
{
private:
  ScanOp scan_op;

public:
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE SwizzleScanOp(ScanOp scan_op)
      : scan_op(scan_op)
  {}

  template <typename T>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE T operator()(const T& a, const T& b)
  {
    T _a(a);
    T _b(b);

    return scan_op(_b, _a);
  }
};

template <typename ReductionOpT>
struct ReduceBySegmentOp
{
  ReductionOpT op;

  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE ReduceBySegmentOp() {}

  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE ReduceBySegmentOp(ReductionOpT op)
      : op(op)
  {}

  template <typename KeyValuePairT>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE KeyValuePairT operator()(const KeyValuePairT& first, const KeyValuePairT& second)
  {
    KeyValuePairT retval;
    retval.key = first.key + second.key;
#ifdef _NVHPC_CUDA // WAR bug on nvc++
    if (second.key)
    {
      retval.value = second.value;
    }
    else
    {
      // If second.value isn't copied into a temporary here, nvc++ will
      // crash while compiling the TestScanByKeyWithLargeTypes test in
      // thrust/testing/scan_by_key.cu:
      auto v2      = second.value;
      retval.value = op(first.value, v2);
    }
#else // not nvc++:
    // if (second.key) {
    //   The second partial reduction spans a segment reset, so it's value
    //   aggregate becomes the running aggregate
    // else {
    //   The second partial reduction does not span a reset, so accumulate both
    //   into the running aggregate
    // }
    retval.value = (second.key) ? second.value : op(first.value, second.value);
#endif
    return retval;
  }
};

template <typename ReductionOpT>
struct ReduceByKeyOp
{
  ReductionOpT op;

  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE ReduceByKeyOp() {}

  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE ReduceByKeyOp(ReductionOpT op)
      : op(op)
  {}

  template <typename KeyValuePairT>
  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE KeyValuePairT operator()(const KeyValuePairT& first, const KeyValuePairT& second)
  {
    KeyValuePairT retval = second;

    if (first.key == second.key)
    {
      retval.value = op(first.value, retval.value);
    }

    return retval;
  }
};

template <typename BinaryOpT>
struct BinaryFlip
{
  BinaryOpT binary_op;

  _CCCL_HOST_DEVICE explicit BinaryFlip(BinaryOpT binary_op)
      : binary_op(binary_op)
  {}

  template <typename T, typename U>
  _CCCL_DEVICE auto operator()(T&& t, U&& u)
    -> decltype(binary_op(::cuda::std::forward<U>(u), ::cuda::std::forward<T>(t)))
  {
    return binary_op(::cuda::std::forward<U>(u), ::cuda::std::forward<T>(t));
  }
};

template <typename BinaryOpT>
_CCCL_HOST_DEVICE BinaryFlip<BinaryOpT> MakeBinaryFlip(BinaryOpT binary_op)
{
  return BinaryFlip<BinaryOpT>(binary_op);
}

CUB_NAMESPACE_END