系统编程

开发平台：
Unix_Linux

qtconcurrentreducekernel.h：源码内容
							/****************************************************************************
**
** Copyright (C) 2008 Nokia Corporation and/or its subsidiary(-ies).
** Contact: Qt Software Information (qt-info@nokia.com)
**
** This file is part of the QtCore module of the Qt Toolkit.
**
** Commercial Usage
** Licensees holding valid Qt Commercial licenses may use this file in
** accordance with the Qt Commercial License Agreement provided with the
** Software or, alternatively, in accordance with the terms contained in
** a written agreement between you and Nokia.
**
**
** GNU General Public License Usage
** Alternatively, this file may be used under the terms of the GNU
** General Public License versions 2.0 or 3.0 as published by the Free
** Software Foundation and appearing in the file LICENSE.GPL included in
** the packaging of this file.  Please review the following information
** to ensure GNU General Public Licensing requirements will be met:
** http://www.fsf.org/licensing/licenses/info/GPLv2.html and
** http://www.gnu.org/copyleft/gpl.html.  In addition, as a special
** exception, Nokia gives you certain additional rights. These rights
** are described in the Nokia Qt GPL Exception version 1.3, included in
** the file GPL_EXCEPTION.txt in this package.
**
** Qt for Windows(R) Licensees
** As a special exception, Nokia, as the sole copyright holder for Qt
** Designer, grants users of the Qt/Eclipse Integration plug-in the
** right for the Qt/Eclipse Integration to link to functionality
** provided by Qt Designer and its related libraries.
**
** If you are unsure which license is appropriate for your use, please
** contact the sales department at qt-sales@nokia.com.
**
****************************************************************************/
#ifndef QTCONCURRENT_REDUCEKERNEL_H
#define QTCONCURRENT_REDUCEKERNEL_H
#include <QtCore/qglobal.h>
#ifndef QT_NO_CONCURRENT
#include <QtCore/qatomic.h>
#include <QtCore/qlist.h>
#include <QtCore/qmap.h>
#include <QtCore/qmutex.h>
#include <QtCore/qthread.h>
#include <QtCore/qvector.h>
QT_BEGIN_HEADER
QT_BEGIN_NAMESPACE
QT_MODULE(Core)
namespace QtConcurrent {
#ifndef qdoc
/*
    The ReduceQueueStartLimit and ReduceQueueThrottleLimit constants
    limit the reduce queue size for MapReduce. When the number of
    reduce blocks in the queue exceeds ReduceQueueStartLimit,
    MapReduce won't start any new threads, and when it exceeds
    ReduceQueueThrottleLimit running threads will be stopped.
*/
enum {
    ReduceQueueStartLimit = 20,
    ReduceQueueThrottleLimit = 30
};
// IntermediateResults holds a block of intermediate results from a
// map or filter functor. The begin/end offsets indicates the origin
// and range of the block.
template <typename T>
class IntermediateResults
{
public:
    int begin, end;
    QVector<T> vector;
};
#endif // qdoc
enum ReduceOption {
    UnorderedReduce = 0x1,
    OrderedReduce = 0x2,
    SequentialReduce = 0x4
    // ParallelReduce = 0x8
};
Q_DECLARE_FLAGS(ReduceOptions, ReduceOption)
Q_DECLARE_OPERATORS_FOR_FLAGS(ReduceOptions)
#ifndef qdoc
// supports both ordered and out-of-order reduction
template <typename ReduceFunctor, typename ReduceResultType, typename T>
class ReduceKernel
{
    typedef QMap<int, IntermediateResults<T> > ResultsMap;
    const ReduceOptions reduceOptions;
    QMutex mutex;
    int progress, resultsMapSize;
    ResultsMap resultsMap;
    bool canReduce(int begin) const
    {
        return (((reduceOptions & UnorderedReduce)
                 && progress == 0)
                || ((reduceOptions & OrderedReduce)
                    && progress == begin));
    }
    void reduceResult(ReduceFunctor &reduce,
                      ReduceResultType &r,
                      const IntermediateResults<T> &result)
    {
        for (int i = 0; i < result.vector.size(); ++i) {
            reduce(r, result.vector.at(i));
        }
    }
    void reduceResults(ReduceFunctor &reduce,
                       ReduceResultType &r,
                       ResultsMap &map)
    {
        typename ResultsMap::iterator it = map.begin();
        while (it != map.end()) {
            reduceResult(reduce, r, it.value());
            ++it;
        }
    }
public:
    ReduceKernel(ReduceOptions _reduceOptions)
        : reduceOptions(_reduceOptions), progress(0), resultsMapSize(0)
    { }
    void runReduce(ReduceFunctor &reduce,
                   ReduceResultType &r,
                   const IntermediateResults<T> &result)
    {
        QMutexLocker locker(&mutex);
        if (!canReduce(result.begin)) {
            ++resultsMapSize;
            resultsMap.insert(result.begin, result);
            return;
        }
        if (reduceOptions & UnorderedReduce) {
            // UnorderedReduce
            progress = -1;
            // reduce this result
            locker.unlock();
            reduceResult(reduce, r, result);
            locker.relock();
            // reduce all stored results as well
            while (!resultsMap.isEmpty()) {
                ResultsMap resultsMapCopy = resultsMap;
                resultsMap.clear();
                locker.unlock();
                reduceResults(reduce, r, resultsMapCopy);
                locker.relock();
                resultsMapSize -= resultsMapCopy.size();
            }
            progress = 0;
        } else {
            // reduce this result
            locker.unlock();
            reduceResult(reduce, r, result);
            locker.relock();
            // OrderedReduce
            progress += result.end - result.begin;
            // reduce as many other results as possible
            typename ResultsMap::iterator it = resultsMap.begin();
            while (it != resultsMap.end()) {
                if (it.value().begin != progress)
                    break;
                locker.unlock();
                reduceResult(reduce, r, it.value());
                locker.relock();
                --resultsMapSize;
                progress += it.value().end - it.value().begin;
                it = resultsMap.erase(it);
            }
        }
    }
    // final reduction
    void finish(ReduceFunctor &reduce, ReduceResultType &r)
    {
        reduceResults(reduce, r, resultsMap);
    }
    inline bool shouldThrottle()
    {
        return (resultsMapSize > (ReduceQueueThrottleLimit * QThread::idealThreadCount()));
    }
    inline bool shouldStartThread()
    {
        return (resultsMapSize <= (ReduceQueueStartLimit * QThread::idealThreadCount()));
    }
};
template <typename Sequence, typename Base, typename Functor1, typename Functor2>
struct SequenceHolder2 : public Base
{
    SequenceHolder2(const Sequence &_sequence,
                    Functor1 functor1,
                    Functor2 functor2,
                    ReduceOptions reduceOptions)
        : Base(_sequence.begin(), _sequence.end(), functor1, functor2, reduceOptions),
          sequence(_sequence)
    { }
    Sequence sequence;
    void finish()
    {
        Base::finish();
        // Clear the sequence to make sure all temporaries are destroyed
        // before finished is signaled.
        sequence = Sequence();
    }
};
#endif //qdoc
} // namespace QtConcurrent
QT_END_NAMESPACE
QT_END_HEADER
#endif // QT_NO_CONCURRENT
#endif

						