TheRock/rocm-libraries/projects/rocthrust/thrust/system/cuda/detail/async/reduce.h

/******************************************************************************
 * Copyright (c) 2016, 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.
 *
 ******************************************************************************/

// TODO: Optimize for thrust::plus

// TODO: Move into system::cuda

#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/cpp_version_check.h>

#if _CCCL_STD_VER >= 2017

#  if _CCCL_HAS_CUDA_COMPILER

#    include <thrust/system/cuda/config.h>

#    include <thrust/distance.h>
#    include <thrust/iterator/iterator_traits.h>
#    include <thrust/system/cuda/detail/async/customization.h>
#    include <thrust/system/cuda/detail/reduce.h>
#    include <thrust/system/cuda/future.h>

#    include <cuda/std/type_traits>

#    include <type_traits>

_CCCL_SUPPRESS_DEPRECATED_PUSH
THRUST_NAMESPACE_BEGIN

namespace system
{
namespace cuda
{
namespace detail
{

template <typename DerivedPolicy, typename ForwardIt, typename Size, typename T, typename BinaryOp>
unique_eager_future<::cuda::std::remove_cvref_t<T>>
async_reduce_n(execution_policy<DerivedPolicy>& policy, ForwardIt first, Size n, T init, BinaryOp op)
{
  using U = ::cuda::std::remove_cvref_t<T>;

  auto const device_alloc = get_async_device_allocator(policy);

  using pointer = typename thrust::detail::allocator_traits<decltype(device_alloc)>::template rebind_traits<U>::pointer;

  unique_eager_future_promise_pair<U, pointer> fp;

  // Determine temporary device storage requirements.

  size_t tmp_size = 0;
  thrust::cuda_cub::throw_on_error(
    cub::DeviceReduce::Reduce(nullptr, tmp_size, first, static_cast<U*>(nullptr), n, op, init, nullptr // Null stream,
                                                                                                       // just for
                                                                                                       // sizing.
                              ),
    "after reduction sizing");

  // Allocate temporary storage.

  auto content = uninitialized_allocate_unique_n<std::uint8_t>(device_alloc, sizeof(U) + tmp_size);

  // The array was dynamically allocated, so we assume that it's suitably
  // aligned for any type of data. `malloc`/`cudaMalloc`/`new`/`std::allocator`
  // make this guarantee.
  auto const content_ptr = content.get();
  U* const ret_ptr       = thrust::detail::aligned_reinterpret_cast<U*>(raw_pointer_cast(content_ptr));
  void* const tmp_ptr    = static_cast<void*>(raw_pointer_cast(content_ptr + sizeof(U)));

  // Set up stream with dependencies.

  cudaStream_t const user_raw_stream = thrust::cuda_cub::stream(policy);

  if (thrust::cuda_cub::default_stream() != user_raw_stream)
  {
    fp = make_dependent_future<U, pointer>(
      [](decltype(content) const& c) {
        return pointer(thrust::detail::aligned_reinterpret_cast<U*>(raw_pointer_cast(c.get())));
      },
      std::tuple_cat(std::make_tuple(std::move(content), unique_stream(nonowning, user_raw_stream)),
                     extract_dependencies(std::move(thrust::detail::derived_cast(policy)))));
  }
  else
  {
    fp = make_dependent_future<U, pointer>(
      [](decltype(content) const& c) {
        return pointer(thrust::detail::aligned_reinterpret_cast<U*>(raw_pointer_cast(c.get())));
      },
      std::tuple_cat(std::make_tuple(std::move(content)),
                     extract_dependencies(std::move(thrust::detail::derived_cast(policy)))));
  }

  // Run reduction.

  thrust::cuda_cub::throw_on_error(
    cub::DeviceReduce::Reduce(tmp_ptr, tmp_size, first, ret_ptr, n, op, init, fp.future.stream().native_handle()),
    "after reduction launch");

  return std::move(fp.future);
}

} // namespace detail
} // namespace cuda
} // namespace system

namespace cuda_cub
{

// ADL entry point.
template <typename DerivedPolicy, typename ForwardIt, typename Sentinel, typename T, typename BinaryOp>
auto async_reduce(execution_policy<DerivedPolicy>& policy, ForwardIt first, Sentinel last, T init, BinaryOp op)
  THRUST_RETURNS(thrust::system::cuda::detail::async_reduce_n(policy, first, thrust::distance(first, last), init, op))

} // namespace cuda_cub

///////////////////////////////////////////////////////////////////////////////

namespace system
{
namespace cuda
{
namespace detail
{

template <typename DerivedPolicy, typename ForwardIt, typename Size, typename OutputIt, typename T, typename BinaryOp>
unique_eager_event async_reduce_into_n(
  execution_policy<DerivedPolicy>& policy, ForwardIt first, Size n, OutputIt output, T init, BinaryOp op)
{
  using U = ::cuda::std::remove_cvref_t<T>;

  auto const device_alloc = get_async_device_allocator(policy);

  unique_eager_event e;

  // Determine temporary device storage requirements.

  size_t tmp_size = 0;
  thrust::cuda_cub::throw_on_error(
    cub::DeviceReduce::Reduce(nullptr, tmp_size, first, static_cast<U*>(nullptr), n, op, init, nullptr // Null stream,
                                                                                                       // just for
                                                                                                       // sizing.
                              ),
    "after reduction sizing");

  // Allocate temporary storage.

  auto content = uninitialized_allocate_unique_n<std::uint8_t>(device_alloc, tmp_size);

  // The array was dynamically allocated, so we assume that it's suitably
  // aligned for any type of data. `malloc`/`cudaMalloc`/`new`/`std::allocator`
  // make this guarantee.
  auto const content_ptr = content.get();

  void* const tmp_ptr = static_cast<void*>(raw_pointer_cast(content_ptr));

  // Set up stream with dependencies.

  cudaStream_t const user_raw_stream = thrust::cuda_cub::stream(policy);

  if (thrust::cuda_cub::default_stream() != user_raw_stream)
  {
    e = make_dependent_event(
      std::tuple_cat(std::make_tuple(std::move(content), unique_stream(nonowning, user_raw_stream)),
                     extract_dependencies(std::move(thrust::detail::derived_cast(policy)))));
  }
  else
  {
    e = make_dependent_event(std::tuple_cat(
      std::make_tuple(std::move(content)), extract_dependencies(std::move(thrust::detail::derived_cast(policy)))));
  }

  // Run reduction.

  thrust::cuda_cub::throw_on_error(
    cub::DeviceReduce::Reduce(tmp_ptr, tmp_size, first, output, n, op, init, e.stream().native_handle()),
    "after reduction launch");

  return e;
}

} // namespace detail
} // namespace cuda
} // namespace system

namespace cuda_cub
{

// ADL entry point.
template <typename DerivedPolicy, typename ForwardIt, typename Sentinel, typename OutputIt, typename T, typename BinaryOp>
auto async_reduce_into(
  execution_policy<DerivedPolicy>& policy, ForwardIt first, Sentinel last, OutputIt output, T init, BinaryOp op)
  THRUST_RETURNS(
    thrust::system::cuda::detail::async_reduce_into_n(policy, first, thrust::distance(first, last), output, init, op))

} // namespace cuda_cub

_CCCL_SUPPRESS_DEPRECATED_POP
THRUST_NAMESPACE_END

#  endif // _CCCL_CUDA_COMPILER

#endif