threading.h 7.48 KB
Newer Older
Martin Reinecke's avatar
Martin Reinecke committed
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
336
#ifndef MRUTIL_THREADING_H
#define MRUTIL_THREADING_H

#include <cstdlib>
#include <mutex>
#include <condition_variable>
#include <thread>
#include <queue>
#include <atomic>
#include <functional>
#include <vector>

#ifdef MRUTIL_THREADING_PTHREADS
#  include <pthread.h>
#endif

namespace mr {

namespace threading {

namespace detail {

using namespace std;

thread_local size_t thread_id = 0;
thread_local size_t num_threads = 1;
static const size_t max_threads = max(1u, thread::hardware_concurrency());

class latch
  {
    atomic<size_t> num_left_;
    mutex mut_;
    condition_variable completed_;
    using lock_t = unique_lock<mutex>;

  public:
    latch(size_t n): num_left_(n) {}

    void count_down()
      {
      lock_t lock(mut_);
      if (--num_left_)
        return;
      completed_.notify_all();
      }

    void wait()
      {
      lock_t lock(mut_);
      completed_.wait(lock, [this]{ return is_ready(); });
      }
    bool is_ready() { return num_left_ == 0; }
  };

template <typename T> class concurrent_queue
  {
    queue<T> q_;
    mutex mut_;
    condition_variable item_added_;
    bool shutdown_;
    using lock_t = unique_lock<mutex>;

  public:
    concurrent_queue(): shutdown_(false) {}

    void push(T val)
      {
      {
      lock_t lock(mut_);
      if (shutdown_)
        throw runtime_error("Item added to queue after shutdown");
      q_.push(move(val));
      }
      item_added_.notify_one();
      }

    bool pop(T & val)
      {
      lock_t lock(mut_);
      item_added_.wait(lock, [this] { return (!q_.empty() || shutdown_); });
      if (q_.empty())
        return false;  // We are shutting down

      val = std::move(q_.front());
      q_.pop();
      return true;
      }

    void shutdown()
      {
      {
      lock_t lock(mut_);
      shutdown_ = true;
      }
      item_added_.notify_all();
      }

    void restart() { shutdown_ = false; }
  };

class thread_pool
  {
    concurrent_queue<function<void()>> work_queue_;
    vector<thread> threads_;

    void worker_main()
      {
      function<void()> work;
      while (work_queue_.pop(work))
        work();
      }

    void create_threads()
      {
      size_t nthreads = threads_.size();
      for (size_t i=0; i<nthreads; ++i)
        {
        try { threads_[i] = thread([this]{ worker_main(); }); }
        catch (...)
          {
          shutdown();
          throw;
          }
        }
      }

  public:
    explicit thread_pool(size_t nthreads):
      threads_(nthreads)
      { create_threads(); }

    thread_pool(): thread_pool(max_threads) {}

    ~thread_pool() { shutdown(); }

    void submit(function<void()> work)
      {
      work_queue_.push(move(work));
      }

    void shutdown()
      {
      work_queue_.shutdown();
      for (auto &thread : threads_)
        if (thread.joinable())
          thread.join();
      }

    void restart()
      {
      work_queue_.restart();
      create_threads();
      }
  };

thread_pool & get_pool()
  {
  static thread_pool pool;
#ifdef MR_THREADING_PTHREADS
  static once_flag f;
  call_once(f,
    []{
    pthread_atfork(
      +[]{ get_pool().shutdown(); },  // prepare
      +[]{ get_pool().restart(); },   // parent
      +[]{ get_pool().restart(); }    // child
      );
    });
#endif

  return pool;
  }

/** Map a function f over nthreads */
template <typename Func>
void thread_map(size_t nthreads, Func f)
  {
  if (nthreads == 0)
    nthreads = max_threads;

  if (nthreads == 1)
    { f(); return; }

  auto & pool = get_pool();
  latch counter(nthreads);
  exception_ptr ex;
  mutex ex_mut;
  for (size_t i=0; i<nthreads; ++i)
    {
    pool.submit(
      [&f, &counter, &ex, &ex_mut, i, nthreads] {
      thread_id = i;
      num_threads = nthreads;
      try { f(); }
      catch (...)
        {
        lock_guard<mutex> lock(ex_mut);
        ex = current_exception();
        }
      counter.count_down();
      });
    }
  counter.wait();
  if (ex)
    rethrow_exception(ex);
  }

class Scheduler
  {
  private:
    size_t nthreads_;
    mutex mut_;
    size_t nwork_;
    size_t cur_;
    size_t chunksize_;
    double fact_max_;
    vector<size_t> nextstart;
    typedef enum { SINGLE, STATIC, DYNAMIC } SchedMode;
    SchedMode mode;
    bool single_done;
    struct Range
      {
      size_t lo, hi;
      Range() : lo(0), hi(0) {}
      Range(size_t lo_, size_t hi_) : lo(lo_), hi(hi_) {}
      operator bool() const { return hi>lo; }
      };

  public:
    size_t nthreads() const { return nthreads_; }
    mutex &mut() { return mut_; }

    template<typename Func> void execSingle(size_t nwork, Func f)
      {
      mode = SINGLE;
      single_done = false;
      nwork_ = nwork;
      f(*this);
      }
    template<typename Func> void execStatic(size_t nwork,
      size_t nthreads, size_t chunksize, Func f)
      {
      mode = STATIC;
      nthreads_ = (nthreads==0) ? max_threads : nthreads;
      nwork_ = nwork;
      chunksize_ = (chunksize<1) ? (nwork_+nthreads_-1)/nthreads_
                                 : chunksize;
      if (chunksize_>=nwork_) return execSingle(nwork_, move(f));
      nextstart.resize(nthreads_);
      for (size_t i=0; i<nextstart.size(); ++i)
        nextstart[i] = i*chunksize_;
      thread_map(nthreads_, [&]() {f(*this);});
      }
    template<typename Func> void execDynamic(size_t nwork,
      size_t nthreads, size_t chunksize_min, double fact_max, Func f)
      {
      mode = DYNAMIC;
      nthreads_ = (nthreads==0) ? max_threads : nthreads;
      nwork_ = nwork;
      chunksize_ = (chunksize_min<1) ? 1 : chunksize_min;
      if (chunksize_*nthreads_>=nwork_)
        return execStatic(nwork, nthreads, 0, move(f));
      fact_max_ = fact_max;
      cur_ = 0;
      thread_map(nthreads_, [&]() {f(*this);});
      }
    Range getNext()
      {
      switch (mode)
        {
        case SINGLE:
          {
          if (single_done) return Range();
          single_done=true;
          return Range(0, nwork_);
          }
        case STATIC:
          {
          if (nextstart[thread_id]>=nwork_) return Range();
          size_t lo=nextstart[thread_id];
          size_t hi=min(lo+chunksize_,nwork_);
          nextstart[thread_id] += nthreads_*chunksize_;
          return Range(lo, hi);
          }
        case DYNAMIC:
          {
          unique_lock<mutex> lck(mut_);
          if (cur_>=nwork_) return Range();
          auto rem = nwork_-cur_;
          size_t tmp = size_t((fact_max_*rem)/nthreads_);
          auto sz = min(rem, max(chunksize_, tmp));
          size_t lo=cur_;
          cur_+=sz;
          size_t hi=cur_;
          return Range(lo, hi);
          }
        }
      return Range();
      }
  };

template<typename Func> void execSingle(size_t nwork, Func f)
  {
  Scheduler sched;
  sched.execSingle(nwork, move(f));
  }
template<typename Func> void execStatic(size_t nwork,
  size_t nthreads, size_t chunksize, Func f)
  {
  Scheduler sched;
  sched.execStatic(nwork, nthreads, chunksize, move(f));
  }
template<typename Func> void execDynamic(size_t nwork,
  size_t nthreads, size_t chunksize_min, Func f)
  {
  Scheduler sched;
  sched.execDynamic(nwork, nthreads, chunksize_min, 0., move(f));
  }
template<typename Func> void execGuided(size_t nwork,
  size_t nthreads, size_t chunksize_min, double fact_max, Func f)
  {
  Scheduler sched;
  sched.execDynamic(nwork, nthreads, chunksize_min, fact_max, move(f));
  }
} // end of namespace detail

using detail::Scheduler;
using detail::execSingle;
using detail::execStatic;
using detail::execDynamic;
using detail::execGuided;

} // end of namespace threading
} // end of namespace mr

#endif