summaryrefslogtreecommitdiffstats
path: root/src/concurrent/qtconcurrentiteratekernel.h
blob: 86150f1d973fde9a00e7d304936133ce0065fd68 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
// Copyright (C) 2016 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only

#ifndef QTCONCURRENT_ITERATEKERNEL_H
#define QTCONCURRENT_ITERATEKERNEL_H

#include <QtConcurrent/qtconcurrent_global.h>

#if !defined(QT_NO_CONCURRENT) || defined(Q_CLANG_QDOC)

#include <QtCore/qatomic.h>
#include <QtConcurrent/qtconcurrentmedian.h>
#include <QtConcurrent/qtconcurrentthreadengine.h>

#include <iterator>

QT_BEGIN_NAMESPACE



namespace QtConcurrent {

/*
    The BlockSizeManager class manages how many iterations a thread should
    reserve and process at a time. This is done by measuring the time spent
    in the user code versus the control part code, and then increasing
    the block size if the ratio between them is to small. The block size
    management is done on the basis of the median of several timing measurements,
    and it is done individually for each thread.
*/
class Q_CONCURRENT_EXPORT BlockSizeManager
{
public:
    explicit BlockSizeManager(QThreadPool *pool, int iterationCount);

    void timeBeforeUser();
    void timeAfterUser();
    int blockSize();

private:
    inline bool blockSizeMaxed()
    {
        return (m_blockSize >= maxBlockSize);
    }

    const int maxBlockSize;
    qint64 beforeUser;
    qint64 afterUser;
    Median controlPartElapsed;
    Median userPartElapsed;
    int m_blockSize;

    Q_DISABLE_COPY(BlockSizeManager)
};

template <typename T>
class ResultReporter
{
public:
    ResultReporter(ThreadEngine<T> *_threadEngine, T &_defaultValue)
        : threadEngine(_threadEngine), defaultValue(_defaultValue)
    {
    }

    void reserveSpace(int resultCount)
    {
        currentResultCount = resultCount;
        resizeList(qMax(resultCount, vector.count()));
    }

    void reportResults(int begin)
    {
        const int useVectorThreshold = 4; // Tunable parameter.
        if (currentResultCount > useVectorThreshold) {
            resizeList(currentResultCount);
            threadEngine->reportResults(vector, begin);
        } else {
            for (int i = 0; i < currentResultCount; ++i)
                threadEngine->reportResult(&vector.at(i), begin + i);
        }
    }

    inline T * getPointer()
    {
        return vector.data();
    }

    int currentResultCount;
    ThreadEngine<T> *threadEngine;
    QList<T> vector;

private:
    void resizeList(qsizetype size)
    {
        if constexpr (std::is_default_constructible_v<T>)
            vector.resize(size);
        else
            vector.resize(size, defaultValue);
    }

    T &defaultValue;
};

template <>
class ResultReporter<void>
{
public:
    inline ResultReporter(ThreadEngine<void> *) { }
    inline void reserveSpace(int) { }
    inline void reportResults(int) { }
    inline void * getPointer() { return nullptr; }
};

template<typename T>
struct DefaultValueContainer
{
    template<typename U = T>
    DefaultValueContainer(U &&_value) : value(std::forward<U>(_value))
    {
    }

    T value;
};

template<>
struct DefaultValueContainer<void>
{
};

inline bool selectIteration(std::bidirectional_iterator_tag)
{
    return false; // while
}

inline bool selectIteration(std::forward_iterator_tag)
{
    return false; // while
}

inline bool selectIteration(std::random_access_iterator_tag)
{
    return true; // for
}

template <typename Iterator, typename T>
class IterateKernel : public ThreadEngine<T>
{
    using IteratorCategory = typename std::iterator_traits<Iterator>::iterator_category;

public:
    typedef T ResultType;

    template<typename U = T, std::enable_if_t<std::is_same_v<U, void>, bool> = true>
    IterateKernel(QThreadPool *pool, Iterator _begin, Iterator _end)
        : ThreadEngine<U>(pool),
          begin(_begin),
          end(_end),
          current(_begin),
          iterationCount(selectIteration(IteratorCategory()) ? std::distance(_begin, _end) : 0),
          forIteration(selectIteration(IteratorCategory())),
          progressReportingEnabled(true)
    {
    }

    template<typename U = T, std::enable_if_t<!std::is_same_v<U, void>, bool> = true>
    IterateKernel(QThreadPool *pool, Iterator _begin, Iterator _end)
        : ThreadEngine<U>(pool),
          begin(_begin),
          end(_end),
          current(_begin),
          iterationCount(selectIteration(IteratorCategory()) ? std::distance(_begin, _end) : 0),
          forIteration(selectIteration(IteratorCategory())),
          progressReportingEnabled(true),
          defaultValue(U())
    {
    }

    template<typename U = T, std::enable_if_t<!std::is_same_v<U, void>, bool> = true>
    IterateKernel(QThreadPool *pool, Iterator _begin, Iterator _end, U &&_defaultValue)
        : ThreadEngine<U>(pool),
          begin(_begin),
          end(_end),
          current(_begin),
          iterationCount(selectIteration(IteratorCategory()) ? std::distance(_begin, _end) : 0),
          forIteration(selectIteration(IteratorCategory())),
          progressReportingEnabled(true),
          defaultValue(std::forward<U>(_defaultValue))
    {
    }

    virtual ~IterateKernel() { }

    virtual bool runIteration(Iterator, int , T *) { return false; }
    virtual bool runIterations(Iterator, int, int, T *) { return false; }

    void start() override
    {
        progressReportingEnabled = this->isProgressReportingEnabled();
        if (progressReportingEnabled && iterationCount > 0)
            this->setProgressRange(0, iterationCount);
    }

    bool shouldStartThread() override
    {
        if (forIteration)
            return (currentIndex.loadRelaxed() < iterationCount) && !this->shouldThrottleThread();
        else // whileIteration
            return (iteratorThreads.loadRelaxed() == 0);
    }

    ThreadFunctionResult threadFunction() override
    {
        if (forIteration)
            return this->forThreadFunction();
        else // whileIteration
            return this->whileThreadFunction();
    }

    ThreadFunctionResult forThreadFunction()
    {
        BlockSizeManager blockSizeManager(ThreadEngineBase::threadPool, iterationCount);
        ResultReporter<T> resultReporter = createResultsReporter();

        for(;;) {
            if (this->isCanceled())
                break;

            const int currentBlockSize = blockSizeManager.blockSize();

            if (currentIndex.loadRelaxed() >= iterationCount)
                break;

            // Atomically reserve a block of iterationCount for this thread.
            const int beginIndex = currentIndex.fetchAndAddRelease(currentBlockSize);
            const int endIndex = qMin(beginIndex + currentBlockSize, iterationCount);

            if (beginIndex >= endIndex) {
                // No more work
                break;
            }

            this->waitForResume(); // (only waits if the qfuture is paused.)

            if (shouldStartThread())
                this->startThread();

            const int finalBlockSize = endIndex - beginIndex; // block size adjusted for possible end-of-range
            resultReporter.reserveSpace(finalBlockSize);

            // Call user code with the current iteration range.
            blockSizeManager.timeBeforeUser();
            const bool resultsAvailable = this->runIterations(begin, beginIndex, endIndex, resultReporter.getPointer());
            blockSizeManager.timeAfterUser();

            if (resultsAvailable)
                resultReporter.reportResults(beginIndex);

            // Report progress if progress reporting enabled.
            if (progressReportingEnabled) {
                completed.fetchAndAddAcquire(finalBlockSize);
                this->setProgressValue(this->completed.loadRelaxed());
            }

            if (this->shouldThrottleThread())
                return ThrottleThread;
        }
        return ThreadFinished;
    }

    ThreadFunctionResult whileThreadFunction()
    {
        if (iteratorThreads.testAndSetAcquire(0, 1) == false)
            return ThreadFinished;

        ResultReporter<T> resultReporter = createResultsReporter();
        resultReporter.reserveSpace(1);

        while (current != end) {
            // The following two lines breaks support for input iterators according to
            // the sgi docs: dereferencing prev after calling ++current is not allowed
            // on input iterators. (prev is dereferenced inside user.runIteration())
            Iterator prev = current;
            ++current;
            int index = currentIndex.fetchAndAddRelaxed(1);
            iteratorThreads.testAndSetRelease(1, 0);

            this->waitForResume(); // (only waits if the qfuture is paused.)

            if (shouldStartThread())
                this->startThread();

            const bool resultAavailable = this->runIteration(prev, index, resultReporter.getPointer());
            if (resultAavailable)
                resultReporter.reportResults(index);

            if (this->shouldThrottleThread())
                return ThrottleThread;

            if (iteratorThreads.testAndSetAcquire(0, 1) == false)
                return ThreadFinished;
        }

        return ThreadFinished;
    }

private:
    ResultReporter<T> createResultsReporter()
    {
        if constexpr (!std::is_same_v<T, void>)
            return ResultReporter<T>(this, defaultValue.value);
        else
            return ResultReporter<T>(this);
    }

public:
    const Iterator begin;
    const Iterator end;
    Iterator current;
    QAtomicInt currentIndex;
    QAtomicInt iteratorThreads;
    QAtomicInt completed;
    const int iterationCount;
    const bool forIteration;
    bool progressReportingEnabled;
    DefaultValueContainer<ResultType> defaultValue;
};

} // namespace QtConcurrent


QT_END_NAMESPACE

#endif // QT_NO_CONCURRENT

#endif