block_reduce.hpp

// Copyright (c) 2017-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.

#ifndef ROCPRIM_BLOCK_BLOCK_REDUCE_HPP_
#define ROCPRIM_BLOCK_BLOCK_REDUCE_HPP_

#include <type_traits>

#include "../config.hpp"
#include "../detail/various.hpp"

#include "../intrinsics.hpp"
#include "../functional.hpp"

#include "detail/block_reduce_warp_reduce.hpp"
#include "detail/block_reduce_raking_reduce.hpp"



BEGIN_ROCPRIM_NAMESPACE

enum class block_reduce_algorithm
{
    using_warp_reduce,
    raking_reduce,
    raking_reduce_commutative_only,
    default_algorithm = using_warp_reduce
};

namespace detail
{

// Selector for block_reduce algorithm which gives block reduce implementation
// type based on passed block_reduce_algorithm enum
template<block_reduce_algorithm Algorithm>
struct select_block_reduce_impl;

template<>
struct select_block_reduce_impl<block_reduce_algorithm::using_warp_reduce>
{
    template<class T, unsigned int BlockSizeX, unsigned int BlockSizeY, unsigned int BlockSizeZ>
    using type = block_reduce_warp_reduce<T, BlockSizeX, BlockSizeY, BlockSizeZ>;
};

template<>
struct select_block_reduce_impl<block_reduce_algorithm::raking_reduce>
{
    template<class T, unsigned int BlockSizeX, unsigned int BlockSizeY, unsigned int BlockSizeZ>
    using type = block_reduce_raking_reduce<T, BlockSizeX, BlockSizeY, BlockSizeZ>;
};

template<>
struct select_block_reduce_impl<block_reduce_algorithm::raking_reduce_commutative_only>
{
    template<class T, unsigned int BlockSizeX, unsigned int BlockSizeY, unsigned int BlockSizeZ>
    using type = block_reduce_raking_reduce<T, BlockSizeX, BlockSizeY, BlockSizeZ, true>;
};


} // end namespace detail

template<
    class T,
    unsigned int BlockSizeX,
    block_reduce_algorithm Algorithm = block_reduce_algorithm::default_algorithm,
    unsigned int BlockSizeY = 1,
    unsigned int BlockSizeZ = 1
>
class block_reduce
#ifndef DOXYGEN_SHOULD_SKIP_THIS
    : private detail::select_block_reduce_impl<Algorithm>::template type<T, BlockSizeX, BlockSizeY, BlockSizeZ>
#endif
{
    using base_type = typename detail::select_block_reduce_impl<Algorithm>::template type<T, BlockSizeX, BlockSizeY, BlockSizeZ>;
public:
    using storage_type = typename base_type::storage_type;

    template<class BinaryFunction = ::rocprim::plus<T>>
    ROCPRIM_DEVICE ROCPRIM_INLINE
    void reduce(T input,
                T& output,
                storage_type& storage,
                BinaryFunction reduce_op = BinaryFunction())
    {
        base_type::reduce(input, output, storage, reduce_op);
    }

    template<class BinaryFunction = ::rocprim::plus<T>>
    ROCPRIM_DEVICE ROCPRIM_FORCE_INLINE
    void reduce(T input,
                T& output,
                BinaryFunction reduce_op = BinaryFunction())
    {
        base_type::reduce(input, output, reduce_op);
    }

    template<
        unsigned int ItemsPerThread,
        class BinaryFunction = ::rocprim::plus<T>
    >
    ROCPRIM_DEVICE ROCPRIM_INLINE
    void reduce(T (&input)[ItemsPerThread],
                T& output,
                storage_type& storage,
                BinaryFunction reduce_op = BinaryFunction())
    {
        base_type::reduce(input, output, storage, reduce_op);
    }

    template<
        unsigned int ItemsPerThread,
        class BinaryFunction = ::rocprim::plus<T>
    >
    ROCPRIM_DEVICE ROCPRIM_FORCE_INLINE
    void reduce(T (&input)[ItemsPerThread],
                T& output,
                BinaryFunction reduce_op = BinaryFunction())
    {
        base_type::reduce(input, output, reduce_op);
    }

    template<class BinaryFunction = ::rocprim::plus<T>>
    ROCPRIM_DEVICE ROCPRIM_INLINE
    void reduce(T input,
                T& output,
                unsigned int valid_items,
                storage_type& storage,
                BinaryFunction reduce_op = BinaryFunction())
    {
        base_type::reduce(input, output, valid_items, storage, reduce_op);
    }

    template<class BinaryFunction = ::rocprim::plus<T>>
    ROCPRIM_DEVICE ROCPRIM_FORCE_INLINE
    void reduce(T input,
                T& output,
                unsigned int valid_items,
                BinaryFunction reduce_op = BinaryFunction())
    {
        base_type::reduce(input, output, valid_items, reduce_op);
    }
};

END_ROCPRIM_NAMESPACE

// end of group blockmodule

#endif // ROCPRIM_BLOCK_BLOCK_REDUCE_HPP_