多线程并行归并排序算法实现

最近在做排序性能优化，从KDE dolphin文件管理源码上面看到一个十分高效的归并排序算法，这个算法能提高大量文件的文件夹排序性能。这个工程是基于QT, 所以多线程是用的QtConcurrent，如果要移植到其他平台就改一下多线程接口就行

/*
 * SPDX-FileCopyrightText: 2012 Peter Penz <peter.penz19@gmail.com>
 * SPDX-FileCopyrightText: 2012 Emmanuel Pescosta <emmanuelpescosta099@gmail.com>
 * SPDX-FileCopyrightText: 2013 Frank Reininghaus <frank78ac@googlemail.com>
 *
 * SPDX-License-Identifier: GPL-2.0-or-later
 */

#ifndef KFILEITEMMODELSORTALGORITHM_H
#define KFILEITEMMODELSORTALGORITHM_H

#include <QtConcurrentRun>

#include <algorithm>

/**
 * Sorts the items using the merge sort algorithm is used to assure a
 * worst-case of O(n * log(n)) and to keep the number of comparisons low.
 *
 * The implementation is based on qStableSortHelper() from qalgorithms.h
 * SPDX-FileCopyrightText: 2011 Nokia Corporation and/or its subsidiary(-ies).
 */

template <typename RandomAccessIterator, typename LessThan>
static void mergeSort(RandomAccessIterator begin,
                      RandomAccessIterator end,
                      const LessThan& lessThan)
{
    // The implementation is based on qStableSortHelper() from qalgorithms.h
    // SPDX-FileCopyrightText: 2011 Nokia Corporation and/or its subsidiary(-ies).

    const int span = end - begin;
    if (span < 2) {
        return;
    }

    const RandomAccessIterator middle = begin + span / 2;
    mergeSort(begin, middle, lessThan);
    mergeSort(middle, end, lessThan);
    merge(begin, middle, end, lessThan);
}

/**
 * Uses up to \a numberOfThreads threads to sort the items between
 * \a begin and \a end. Only item ranges longer than
 * \a parallelMergeSortingThreshold are split to be sorted by two different
 * threads.
 *
 * The comparison function \a lessThan must be reentrant.
 */

template <typename RandomAccessIterator, typename LessThan>
static void parallelMergeSort(RandomAccessIterator begin,
                              RandomAccessIterator end,
                              LessThan lessThan,
                              int numberOfThreads,
                              int parallelMergeSortingThreshold = 100)
{
    const int span = end - begin;

    if (numberOfThreads > 1 && span > parallelMergeSortingThreshold) {
        const int newNumberOfThreads = numberOfThreads / 2;
        const RandomAccessIterator middle = begin + span / 2;

        QFuture<void> future = QtConcurrent::run(parallelMergeSort<RandomAccessIterator, LessThan>, begin, middle, lessThan, newNumberOfThreads, parallelMergeSortingThreshold);
        parallelMergeSort(middle, end, lessThan, newNumberOfThreads, parallelMergeSortingThreshold);

        future.waitForFinished();

        merge(begin, middle, end, lessThan);
    } else {
        mergeSort(begin, end, lessThan);
    }
}

/**
 * Merges the sorted item ranges between \a begin and \a pivot and
 * between \a pivot and \a end into a single sorted range between
 * \a begin and \a end.
 *
 * The implementation is based on qMerge() from qalgorithms.h
 * SPDX-FileCopyrightText: 2011 Nokia Corporation and/or its subsidiary(-ies).
 */

template <typename RandomAccessIterator, typename LessThan>
static void merge(RandomAccessIterator begin,
                  RandomAccessIterator pivot,
                  RandomAccessIterator end,
                  const LessThan& lessThan)
{
    // The implementation is based on qMerge() from qalgorithms.h
    // SPDX-FileCopyrightText: 2011 Nokia Corporation and/or its subsidiary(-ies).

    const int len1 = pivot - begin;
    const int len2 = end - pivot;

    if (len1 == 0 || len2 == 0) {
        return;
    }

    if (len1 + len2 == 2) {
        if (lessThan(*(begin + 1), *(begin))) {
            qSwap(*begin, *(begin + 1));
        }
        return;
    }

    RandomAccessIterator firstCut;
    RandomAccessIterator secondCut;
    int len2Half;
    if (len1 > len2) {
        const int len1Half = len1 / 2;
        firstCut = begin + len1Half;
        secondCut = std::lower_bound<RandomAccessIterator,
            decltype(*firstCut), const LessThan&>(pivot, end, *firstCut, lessThan);
        len2Half = secondCut - pivot;
    } else {
        len2Half = len2 / 2;
        secondCut = pivot + len2Half;
        firstCut = std::upper_bound<RandomAccessIterator,
            decltype(*secondCut), const LessThan&>(begin, pivot, *secondCut, lessThan);
    }

    std::rotate(firstCut, pivot, secondCut);

    RandomAccessIterator newPivot = firstCut + len2Half;
    merge(begin, firstCut, newPivot, lessThan);
    merge(newPivot, secondCut, end, lessThan);
}

#endif

时间复杂度：最好、最坏和平均时间复杂度都是 O(nlogn)，排序性能不受待排序数据的混乱程度影响，比较稳定，这也是相对于快排的优势所在。

空间复杂度为：O(n)。合并子序列时需要用到辅助空间，长度为数列长度 n。