FM / detail

/usr/include/boost/compute/algorithm/detail/

/usr/include/boost/compute/algorithm/detail

Name	Size	Mode	Actions
balanced_path.hpp	6267	0644	edit dl rm
binary_find.hpp	4626	0644	edit dl rm
compact.hpp	2188	0644	edit dl rm
copy_on_device.hpp	7382	0644	edit dl rm
copy_to_device.hpp	6731	0644	edit dl rm
copy_to_host.hpp	6658	0644	edit dl rm
count_if_with_ballot.hpp	2605	0644	edit dl rm
count_if_with_reduce.hpp	2648	0644	edit dl rm
count_if_with_threads.hpp	4241	0644	edit dl rm
find_extrema.hpp	2441	0644	edit dl rm
find_extrema_on_cpu.hpp	5333	0644	edit dl rm
find_extrema_with_atomics.hpp	4251	0644	edit dl rm
find_extrema_with_reduce.hpp	18672	0644	edit dl rm
find_if_with_atomics.hpp	8509	0644	edit dl rm
inplace_reduce.hpp	4839	0644	edit dl rm
insertion_sort.hpp	6030	0644	edit dl rm
merge_path.hpp	3874	0644	edit dl rm
merge_sort_on_cpu.hpp	14654	0644	edit dl rm
merge_sort_on_gpu.hpp	22093	0644	edit dl rm
merge_with_merge_path.hpp	7339	0644	edit dl rm
radix_sort.hpp	15502	0644	edit dl rm
random_fill.hpp	1880	0644	edit dl rm
reduce_by_key.hpp	5243	0644	edit dl rm
reduce_by_key_with_scan.hpp	23735	0644	edit dl rm
reduce_on_cpu.hpp	3949	0644	edit dl rm
reduce_on_gpu.hpp	10391	0644	edit dl rm
scan.hpp	1545	0644	edit dl rm
scan_on_cpu.hpp	7056	0644	edit dl rm
scan_on_gpu.hpp	11320	0644	edit dl rm
search_all.hpp	2579	0644	edit dl rm
serial_accumulate.hpp	1990	0644	edit dl rm
serial_count_if.hpp	2243	0644	edit dl rm
serial_find_extrema.hpp	3151	0644	edit dl rm
serial_merge.hpp	3504	0644	edit dl rm
serial_reduce.hpp	2145	0644	edit dl rm
serial_reduce_by_key.hpp	4240	0644	edit dl rm
serial_scan.hpp	3184	0644	edit dl rm

Edit: /usr/include/boost/compute/algorithm/detail/reduce_by_key.hpp (5243B)

//---------------------------------------------------------------------------//
// Copyright (c) 2015 Jakub Szuppe <j.szuppe@gmail.com>
//
// Distributed under the Boost Software License, Version 1.0
// See accompanying file LICENSE_1_0.txt or copy at
// http://www.boost.org/LICENSE_1_0.txt
//
// See http://boostorg.github.com/compute for more information.
//---------------------------------------------------------------------------//

#ifndef BOOST_COMPUTE_ALGORITHM_DETAIL_REDUCE_BY_KEY_HPP
#define BOOST_COMPUTE_ALGORITHM_DETAIL_REDUCE_BY_KEY_HPP

#include <algorithm>
#include <iterator>

#include <boost/compute/command_queue.hpp>
#include <boost/compute/functional.hpp>
#include <boost/compute/container/vector.hpp>
#include <boost/compute/detail/iterator_range_size.hpp>
#include <boost/compute/algorithm/detail/serial_reduce_by_key.hpp>
#include <boost/compute/algorithm/detail/reduce_by_key_with_scan.hpp>
#include <boost/compute/type_traits.hpp>

namespace boost {
namespace compute {
namespace detail {

template<class InputKeyIterator, class InputValueIterator,
         class OutputKeyIterator, class OutputValueIterator,
         class BinaryFunction, class BinaryPredicate>
size_t reduce_by_key_on_gpu(InputKeyIterator keys_first,
                            InputKeyIterator keys_last,
                            InputValueIterator values_first,
                            OutputKeyIterator keys_result,
                            OutputValueIterator values_result,
                            BinaryFunction function,
                            BinaryPredicate predicate,
                            command_queue &queue)
{
    return detail::reduce_by_key_with_scan(keys_first, keys_last, values_first,
                                           keys_result, values_result, function,
                                           predicate, queue);
}

template<class InputKeyIterator, class InputValueIterator,
         class OutputKeyIterator, class OutputValueIterator>
bool reduce_by_key_on_gpu_requirements_met(InputKeyIterator keys_first,
                                           InputValueIterator values_first,
                                           OutputKeyIterator keys_result,
                                           OutputValueIterator values_result,
                                           const size_t count,
                                           command_queue &queue)
{
    const device &device = queue.get_device();
    return (count > 256)
               && !(device.type() & device::cpu)
               && reduce_by_key_with_scan_requirements_met(keys_first, values_first,
                                                           keys_result,values_result,
                                                           count, queue);
    return true;
}

template<class InputKeyIterator, class InputValueIterator,
         class OutputKeyIterator, class OutputValueIterator,
         class BinaryFunction, class BinaryPredicate>
inline std::pair<OutputKeyIterator, OutputValueIterator>
dispatch_reduce_by_key(InputKeyIterator keys_first,
                       InputKeyIterator keys_last,
                       InputValueIterator values_first,
                       OutputKeyIterator keys_result,
                       OutputValueIterator values_result,
                       BinaryFunction function,
                       BinaryPredicate predicate,
                       command_queue &queue)
{
    typedef typename
        std::iterator_traits<OutputKeyIterator>::difference_type key_difference_type;
    typedef typename
        std::iterator_traits<OutputValueIterator>::difference_type value_difference_type;

const size_t count = detail::iterator_range_size(keys_first, keys_last);
    if (count < 2) {
        boost::compute::copy_n(keys_first, count, keys_result, queue);
        boost::compute::copy_n(values_first, count, values_result, queue);
        return
            std::make_pair<OutputKeyIterator, OutputValueIterator>(
                keys_result + static_cast<key_difference_type>(count),
                values_result + static_cast<value_difference_type>(count)
            );
    }

size_t result_size = 0;
    if(reduce_by_key_on_gpu_requirements_met(keys_first, values_first, keys_result,
                                             values_result, count, queue)){
        result_size =
            detail::reduce_by_key_on_gpu(keys_first, keys_last, values_first,
                                         keys_result, values_result, function,
                                         predicate, queue);
    }
    else {
        result_size =
              detail::serial_reduce_by_key(keys_first, keys_last, values_first,
                                           keys_result, values_result, function,
                                           predicate, queue);
    }

return
        std::make_pair<OutputKeyIterator, OutputValueIterator>(
            keys_result + static_cast<key_difference_type>(result_size),
            values_result + static_cast<value_difference_type>(result_size)
        );
}

} // end detail namespace
} // end compute namespace
} // end boost namespace

#endif // BOOST_COMPUTE_ALGORITHM_DETAIL_REDUCE_BY_KEY_HPP

//---------------------------------------------------------------------------// // Copyright (c) 2015 Jakub Szuppe // // Distributed under the Boost Software License, Version 1.0 // See accompanying file LICENSE_1_0.txt or copy at // http://www.boost.org/LICENSE_1_0.txt // // See http://boostorg.github.com/compute for more information. //---------------------------------------------------------------------------// #ifndef BOOST_COMPUTE_ALGORITHM_DETAIL_REDUCE_BY_KEY_HPP #define BOOST_COMPUTE_ALGORITHM_DETAIL_REDUCE_BY_KEY_HPP #include #include #include #include #include #include #include #include #include namespace boost { namespace compute { namespace detail { template size_t reduce_by_key_on_gpu(InputKeyIterator keys_first, InputKeyIterator keys_last, InputValueIterator values_first, OutputKeyIterator keys_result, OutputValueIterator values_result, BinaryFunction function, BinaryPredicate predicate, command_queue &queue) { return detail::reduce_by_key_with_scan(keys_first, keys_last, values_first, keys_result, values_result, function, predicate, queue); } template bool reduce_by_key_on_gpu_requirements_met(InputKeyIterator keys_first, InputValueIterator values_first, OutputKeyIterator keys_result, OutputValueIterator values_result, const size_t count, command_queue &queue) { const device &device = queue.get_device(); return (count > 256) && !(device.type() & device::cpu) && reduce_by_key_with_scan_requirements_met(keys_first, values_first, keys_result,values_result, count, queue); return true; } template inline std::pair dispatch_reduce_by_key(InputKeyIterator keys_first, InputKeyIterator keys_last, InputValueIterator values_first, OutputKeyIterator keys_result, OutputValueIterator values_result, BinaryFunction function, BinaryPredicate predicate, command_queue &queue) { typedef typename std::iterator_traits::difference_type key_difference_type; typedef typename std::iterator_traits::difference_type value_difference_type; const size_t count = detail::iterator_range_size(keys_first, keys_last); if (count < 2) { boost::compute::copy_n(keys_first, count, keys_result, queue); boost::compute::copy_n(values_first, count, values_result, queue); return std::make_pair( keys_result + static_cast(count), values_result + static_cast(count) ); } size_t result_size = 0; if(reduce_by_key_on_gpu_requirements_met(keys_first, values_first, keys_result, values_result, count, queue)){ result_size = detail::reduce_by_key_on_gpu(keys_first, keys_last, values_first, keys_result, values_result, function, predicate, queue); } else { result_size = detail::serial_reduce_by_key(keys_first, keys_last, values_first, keys_result, values_result, function, predicate, queue); } return std::make_pair( keys_result + static_cast(result_size), values_result + static_cast(result_size) ); } } // end detail namespace } // end compute namespace } // end boost namespace #endif // BOOST_COMPUTE_ALGORITHM_DETAIL_REDUCE_BY_KEY_HPP