OpenCores
URL https://opencores.org/ocsvn/altor32/altor32/trunk

Subversion Repositories altor32

[/] [altor32/] [trunk/] [gcc-x64/] [or1knd-elf/] [or1knd-elf/] [include/] [c++/] [4.8.0/] [parallel/] [random_shuffle.h] - Blame information for rev 35

Details | Compare with Previous | View Log

Line No. Rev Author Line
1 35 ultra_embe
// -*- C++ -*-
2
 
3
// Copyright (C) 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc.
4
//
5
// This file is part of the GNU ISO C++ Library.  This library is free
6
// software; you can redistribute it and/or modify it under the terms
7
// of the GNU General Public License as published by the Free Software
8
// Foundation; either version 3, or (at your option) any later
9
// version.
10
 
11
// This library is distributed in the hope that it will be useful, but
12
// WITHOUT ANY WARRANTY; without even the implied warranty of
13
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
// General Public License for more details.
15
 
16
// Under Section 7 of GPL version 3, you are granted additional
17
// permissions described in the GCC Runtime Library Exception, version
18
// 3.1, as published by the Free Software Foundation.
19
 
20
// You should have received a copy of the GNU General Public License and
21
// a copy of the GCC Runtime Library Exception along with this program;
22
// see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
23
// <http://www.gnu.org/licenses/>.
24
 
25
/** @file parallel/random_shuffle.h
26
 *  @brief Parallel implementation of std::random_shuffle().
27
 *  This file is a GNU parallel extension to the Standard C++ Library.
28
 */
29
 
30
// Written by Johannes Singler.
31
 
32
#ifndef _GLIBCXX_PARALLEL_RANDOM_SHUFFLE_H
33
#define _GLIBCXX_PARALLEL_RANDOM_SHUFFLE_H 1
34
 
35
#include <limits>
36
#include <bits/stl_numeric.h>
37
#include <parallel/parallel.h>
38
#include <parallel/random_number.h>
39
 
40
namespace __gnu_parallel
41
{
42
  /** @brief Type to hold the index of a bin.
43
    *
44
    *  Since many variables of this type are allocated, it should be
45
    *  chosen as small as possible.
46
    */
47
  typedef unsigned short _BinIndex;
48
 
49
  /** @brief Data known to every thread participating in
50
      __gnu_parallel::__parallel_random_shuffle(). */
51
  template<typename _RAIter>
52
    struct _DRandomShufflingGlobalData
53
    {
54
      typedef std::iterator_traits<_RAIter> _TraitsType;
55
      typedef typename _TraitsType::value_type _ValueType;
56
      typedef typename _TraitsType::difference_type _DifferenceType;
57
 
58
      /** @brief Begin iterator of the __source. */
59
      _RAIter& _M_source;
60
 
61
      /** @brief Temporary arrays for each thread. */
62
      _ValueType** _M_temporaries;
63
 
64
      /** @brief Two-dimensional array to hold the thread-bin distribution.
65
       *
66
       *  Dimensions (_M_num_threads + 1) __x (_M_num_bins + 1). */
67
      _DifferenceType** _M_dist;
68
 
69
      /** @brief Start indexes of the threads' __chunks. */
70
      _DifferenceType* _M_starts;
71
 
72
      /** @brief Number of the thread that will further process the
73
          corresponding bin. */
74
      _ThreadIndex* _M_bin_proc;
75
 
76
      /** @brief Number of bins to distribute to. */
77
      int _M_num_bins;
78
 
79
      /** @brief Number of bits needed to address the bins. */
80
      int _M_num_bits;
81
 
82
      /** @brief Constructor. */
83
      _DRandomShufflingGlobalData(_RAIter& __source)
84
      : _M_source(__source) { }
85
    };
86
 
87
  /** @brief Local data for a thread participating in
88
      __gnu_parallel::__parallel_random_shuffle().
89
    */
90
  template<typename _RAIter, typename _RandomNumberGenerator>
91
    struct _DRSSorterPU
92
    {
93
      /** @brief Number of threads participating in total. */
94
      int _M_num_threads;
95
 
96
      /** @brief Begin index for bins taken care of by this thread. */
97
      _BinIndex _M_bins_begin;
98
 
99
      /** @brief End index for bins taken care of by this thread. */
100
      _BinIndex __bins_end;
101
 
102
      /** @brief Random _M_seed for this thread. */
103
      uint32_t _M_seed;
104
 
105
      /** @brief Pointer to global data. */
106
      _DRandomShufflingGlobalData<_RAIter>* _M_sd;
107
    };
108
 
109
  /** @brief Generate a random number in @c [0,2^__logp).
110
    *  @param __logp Logarithm (basis 2) of the upper range __bound.
111
    *  @param __rng Random number generator to use.
112
    */
113
  template<typename _RandomNumberGenerator>
114
    inline int
115
    __random_number_pow2(int __logp, _RandomNumberGenerator& __rng)
116
    { return __rng.__genrand_bits(__logp); }
117
 
118
  /** @brief Random shuffle code executed by each thread.
119
    *  @param __pus Array of thread-local data records. */
120
  template<typename _RAIter, typename _RandomNumberGenerator>
121
    void
122
    __parallel_random_shuffle_drs_pu(_DRSSorterPU<_RAIter,
123
                                     _RandomNumberGenerator>* __pus)
124
    {
125
      typedef std::iterator_traits<_RAIter> _TraitsType;
126
      typedef typename _TraitsType::value_type _ValueType;
127
      typedef typename _TraitsType::difference_type _DifferenceType;
128
 
129
      _ThreadIndex __iam = omp_get_thread_num();
130
      _DRSSorterPU<_RAIter, _RandomNumberGenerator>* __d = &__pus[__iam];
131
      _DRandomShufflingGlobalData<_RAIter>* __sd = __d->_M_sd;
132
 
133
      // Indexing: _M_dist[bin][processor]
134
      _DifferenceType __length = (__sd->_M_starts[__iam + 1]
135
                                  - __sd->_M_starts[__iam]);
136
      _BinIndex* __oracles = new _BinIndex[__length];
137
      _DifferenceType* __dist = new _DifferenceType[__sd->_M_num_bins + 1];
138
      _BinIndex* __bin_proc = new _BinIndex[__sd->_M_num_bins];
139
      _ValueType** __temporaries = new _ValueType*[__d->_M_num_threads];
140
 
141
      // Compute oracles and count appearances.
142
      for (_BinIndex __b = 0; __b < __sd->_M_num_bins + 1; ++__b)
143
        __dist[__b] = 0;
144
      int __num_bits = __sd->_M_num_bits;
145
 
146
      _RandomNumber __rng(__d->_M_seed);
147
 
148
      // First main loop.
149
      for (_DifferenceType __i = 0; __i < __length; ++__i)
150
        {
151
          _BinIndex __oracle = __random_number_pow2(__num_bits, __rng);
152
          __oracles[__i] = __oracle;
153
 
154
          // To allow prefix (partial) sum.
155
          ++(__dist[__oracle + 1]);
156
        }
157
 
158
      for (_BinIndex __b = 0; __b < __sd->_M_num_bins + 1; ++__b)
159
        __sd->_M_dist[__b][__iam + 1] = __dist[__b];
160
 
161
#     pragma omp barrier
162
 
163
#     pragma omp single
164
      {
165
        // Sum up bins, __sd->_M_dist[__s + 1][__d->_M_num_threads] now
166
        // contains the total number of items in bin __s
167
        for (_BinIndex __s = 0; __s < __sd->_M_num_bins; ++__s)
168
          __gnu_sequential::partial_sum(__sd->_M_dist[__s + 1],
169
                                        __sd->_M_dist[__s + 1]
170
                                        + __d->_M_num_threads + 1,
171
                                        __sd->_M_dist[__s + 1]);
172
      }
173
 
174
#     pragma omp barrier
175
 
176
      _SequenceIndex __offset = 0, __global_offset = 0;
177
      for (_BinIndex __s = 0; __s < __d->_M_bins_begin; ++__s)
178
        __global_offset += __sd->_M_dist[__s + 1][__d->_M_num_threads];
179
 
180
#     pragma omp barrier
181
 
182
      for (_BinIndex __s = __d->_M_bins_begin; __s < __d->__bins_end; ++__s)
183
        {
184
          for (int __t = 0; __t < __d->_M_num_threads + 1; ++__t)
185
            __sd->_M_dist[__s + 1][__t] += __offset;
186
          __offset = __sd->_M_dist[__s + 1][__d->_M_num_threads];
187
        }
188
 
189
      __sd->_M_temporaries[__iam] = static_cast<_ValueType*>
190
        (::operator new(sizeof(_ValueType) * __offset));
191
 
192
#     pragma omp barrier
193
 
194
      // Draw local copies to avoid false sharing.
195
      for (_BinIndex __b = 0; __b < __sd->_M_num_bins + 1; ++__b)
196
        __dist[__b] = __sd->_M_dist[__b][__iam];
197
      for (_BinIndex __b = 0; __b < __sd->_M_num_bins; ++__b)
198
        __bin_proc[__b] = __sd->_M_bin_proc[__b];
199
      for (_ThreadIndex __t = 0; __t < __d->_M_num_threads; ++__t)
200
        __temporaries[__t] = __sd->_M_temporaries[__t];
201
 
202
      _RAIter __source = __sd->_M_source;
203
      _DifferenceType __start = __sd->_M_starts[__iam];
204
 
205
      // Distribute according to oracles, second main loop.
206
      for (_DifferenceType __i = 0; __i < __length; ++__i)
207
        {
208
          _BinIndex __target_bin = __oracles[__i];
209
          _ThreadIndex __target_p = __bin_proc[__target_bin];
210
 
211
          // Last column [__d->_M_num_threads] stays unchanged.
212
          ::new(&(__temporaries[__target_p][__dist[__target_bin + 1]++]))
213
              _ValueType(*(__source + __i + __start));
214
        }
215
 
216
      delete[] __oracles;
217
      delete[] __dist;
218
      delete[] __bin_proc;
219
      delete[] __temporaries;
220
 
221
#     pragma omp barrier
222
 
223
      // Shuffle bins internally.
224
      for (_BinIndex __b = __d->_M_bins_begin; __b < __d->__bins_end; ++__b)
225
        {
226
          _ValueType* __begin =
227
            (__sd->_M_temporaries[__iam]
228
             + (__b == __d->_M_bins_begin
229
                ? 0 : __sd->_M_dist[__b][__d->_M_num_threads])),
230
            *__end = (__sd->_M_temporaries[__iam]
231
                      + __sd->_M_dist[__b + 1][__d->_M_num_threads]);
232
 
233
          __sequential_random_shuffle(__begin, __end, __rng);
234
          std::copy(__begin, __end, __sd->_M_source + __global_offset
235
                    + (__b == __d->_M_bins_begin
236
                       ? 0 : __sd->_M_dist[__b][__d->_M_num_threads]));
237
        }
238
 
239
      for (_SequenceIndex __i = 0; __i < __offset; ++__i)
240
        __sd->_M_temporaries[__iam][__i].~_ValueType();
241
      ::operator delete(__sd->_M_temporaries[__iam]);
242
    }
243
 
244
  /** @brief Round up to the next greater power of 2.
245
    *  @param __x _Integer to round up */
246
  template<typename _Tp>
247
    _Tp
248
    __round_up_to_pow2(_Tp __x)
249
    {
250
      if (__x <= 1)
251
        return 1;
252
      else
253
        return (_Tp)1 << (__rd_log2(__x - 1) + 1);
254
    }
255
 
256
  /** @brief Main parallel random shuffle step.
257
    *  @param __begin Begin iterator of sequence.
258
    *  @param __end End iterator of sequence.
259
    *  @param __n Length of sequence.
260
    *  @param __num_threads Number of threads to use.
261
    *  @param __rng Random number generator to use.
262
    */
263
  template<typename _RAIter, typename _RandomNumberGenerator>
264
    void
265
    __parallel_random_shuffle_drs(_RAIter __begin, _RAIter __end,
266
                                  typename std::iterator_traits
267
                                  <_RAIter>::difference_type __n,
268
                                  _ThreadIndex __num_threads,
269
                                  _RandomNumberGenerator& __rng)
270
    {
271
      typedef std::iterator_traits<_RAIter> _TraitsType;
272
      typedef typename _TraitsType::value_type _ValueType;
273
      typedef typename _TraitsType::difference_type _DifferenceType;
274
 
275
      _GLIBCXX_CALL(__n)
276
 
277
      const _Settings& __s = _Settings::get();
278
 
279
      if (__num_threads > __n)
280
        __num_threads = static_cast<_ThreadIndex>(__n);
281
 
282
      _BinIndex __num_bins, __num_bins_cache;
283
 
284
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_L1
285
      // Try the L1 cache first.
286
 
287
      // Must fit into L1.
288
      __num_bins_cache =
289
        std::max<_DifferenceType>(1, __n / (__s.L1_cache_size_lb
290
                                            / sizeof(_ValueType)));
291
      __num_bins_cache = __round_up_to_pow2(__num_bins_cache);
292
 
293
      // No more buckets than TLB entries, power of 2
294
      // Power of 2 and at least one element per bin, at most the TLB size.
295
      __num_bins = std::min<_DifferenceType>(__n, __num_bins_cache);
296
 
297
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_TLB
298
      // 2 TLB entries needed per bin.
299
      __num_bins = std::min<_DifferenceType>(__s.TLB_size / 2, __num_bins);
300
#endif
301
      __num_bins = __round_up_to_pow2(__num_bins);
302
 
303
      if (__num_bins < __num_bins_cache)
304
        {
305
#endif
306
          // Now try the L2 cache
307
          // Must fit into L2
308
          __num_bins_cache = static_cast<_BinIndex>
309
            (std::max<_DifferenceType>(1, __n / (__s.L2_cache_size
310
                                                 / sizeof(_ValueType))));
311
          __num_bins_cache = __round_up_to_pow2(__num_bins_cache);
312
 
313
          // No more buckets than TLB entries, power of 2.
314
          __num_bins = static_cast<_BinIndex>
315
            (std::min(__n, static_cast<_DifferenceType>(__num_bins_cache)));
316
          // Power of 2 and at least one element per bin, at most the TLB size.
317
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_TLB
318
          // 2 TLB entries needed per bin.
319
          __num_bins = std::min(static_cast<_DifferenceType>(__s.TLB_size / 2),
320
                                __num_bins);
321
#endif
322
            __num_bins = __round_up_to_pow2(__num_bins);
323
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_L1
324
        }
325
#endif
326
 
327
      __num_bins = __round_up_to_pow2(
328
                        std::max<_BinIndex>(__num_threads, __num_bins));
329
 
330
      if (__num_threads <= 1)
331
      {
332
        _RandomNumber __derived_rng(
333
                            __rng(std::numeric_limits<uint32_t>::max()));
334
        __sequential_random_shuffle(__begin, __end, __derived_rng);
335
        return;
336
      }
337
 
338
      _DRandomShufflingGlobalData<_RAIter> __sd(__begin);
339
      _DRSSorterPU<_RAIter, _RandomNumber >* __pus;
340
      _DifferenceType* __starts;
341
 
342
#     pragma omp parallel num_threads(__num_threads)
343
      {
344
        _ThreadIndex __num_threads = omp_get_num_threads();
345
#       pragma omp single
346
        {
347
          __pus = new _DRSSorterPU<_RAIter, _RandomNumber>[__num_threads];
348
 
349
          __sd._M_temporaries = new _ValueType*[__num_threads];
350
          __sd._M_dist = new _DifferenceType*[__num_bins + 1];
351
          __sd._M_bin_proc = new _ThreadIndex[__num_bins];
352
          for (_BinIndex __b = 0; __b < __num_bins + 1; ++__b)
353
            __sd._M_dist[__b] = new _DifferenceType[__num_threads + 1];
354
          for (_BinIndex __b = 0; __b < (__num_bins + 1); ++__b)
355
            {
356
              __sd._M_dist[0][0] = 0;
357
              __sd._M_dist[__b][0] = 0;
358
            }
359
          __starts = __sd._M_starts = new _DifferenceType[__num_threads + 1];
360
          int __bin_cursor = 0;
361
          __sd._M_num_bins = __num_bins;
362
          __sd._M_num_bits = __rd_log2(__num_bins);
363
 
364
          _DifferenceType __chunk_length = __n / __num_threads,
365
                                 __split = __n % __num_threads,
366
                                 __start = 0;
367
          _DifferenceType __bin_chunk_length = __num_bins / __num_threads,
368
                                 __bin_split = __num_bins % __num_threads;
369
          for (_ThreadIndex __i = 0; __i < __num_threads; ++__i)
370
            {
371
              __starts[__i] = __start;
372
              __start += (__i < __split
373
                          ? (__chunk_length + 1) : __chunk_length);
374
              int __j = __pus[__i]._M_bins_begin = __bin_cursor;
375
 
376
              // Range of bins for this processor.
377
              __bin_cursor += (__i < __bin_split
378
                               ? (__bin_chunk_length + 1)
379
                               : __bin_chunk_length);
380
              __pus[__i].__bins_end = __bin_cursor;
381
              for (; __j < __bin_cursor; ++__j)
382
                __sd._M_bin_proc[__j] = __i;
383
              __pus[__i]._M_num_threads = __num_threads;
384
              __pus[__i]._M_seed = __rng(std::numeric_limits<uint32_t>::max());
385
              __pus[__i]._M_sd = &__sd;
386
            }
387
          __starts[__num_threads] = __start;
388
        } //single
389
          // Now shuffle in parallel.
390
        __parallel_random_shuffle_drs_pu(__pus);
391
      }  // parallel
392
 
393
      delete[] __starts;
394
      delete[] __sd._M_bin_proc;
395
      for (int __s = 0; __s < (__num_bins + 1); ++__s)
396
        delete[] __sd._M_dist[__s];
397
      delete[] __sd._M_dist;
398
      delete[] __sd._M_temporaries;
399
 
400
      delete[] __pus;
401
    }
402
 
403
  /** @brief Sequential cache-efficient random shuffle.
404
   *  @param __begin Begin iterator of sequence.
405
   *  @param __end End iterator of sequence.
406
   *  @param __rng Random number generator to use.
407
   */
408
  template<typename _RAIter, typename _RandomNumberGenerator>
409
    void
410
    __sequential_random_shuffle(_RAIter __begin, _RAIter __end,
411
                                _RandomNumberGenerator& __rng)
412
    {
413
      typedef std::iterator_traits<_RAIter> _TraitsType;
414
      typedef typename _TraitsType::value_type _ValueType;
415
      typedef typename _TraitsType::difference_type _DifferenceType;
416
 
417
      _DifferenceType __n = __end - __begin;
418
      const _Settings& __s = _Settings::get();
419
 
420
      _BinIndex __num_bins, __num_bins_cache;
421
 
422
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_L1
423
      // Try the L1 cache first, must fit into L1.
424
      __num_bins_cache = std::max<_DifferenceType>
425
        (1, __n / (__s.L1_cache_size_lb / sizeof(_ValueType)));
426
      __num_bins_cache = __round_up_to_pow2(__num_bins_cache);
427
 
428
      // No more buckets than TLB entries, power of 2
429
      // Power of 2 and at least one element per bin, at most the TLB size
430
      __num_bins = std::min(__n, (_DifferenceType)__num_bins_cache);
431
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_TLB
432
      // 2 TLB entries needed per bin
433
      __num_bins = std::min((_DifferenceType)__s.TLB_size / 2, __num_bins);
434
#endif
435
      __num_bins = __round_up_to_pow2(__num_bins);
436
 
437
      if (__num_bins < __num_bins_cache)
438
        {
439
#endif
440
          // Now try the L2 cache, must fit into L2.
441
          __num_bins_cache = static_cast<_BinIndex>
442
            (std::max<_DifferenceType>(1, __n / (__s.L2_cache_size
443
                                                 / sizeof(_ValueType))));
444
          __num_bins_cache = __round_up_to_pow2(__num_bins_cache);
445
 
446
          // No more buckets than TLB entries, power of 2
447
          // Power of 2 and at least one element per bin, at most the TLB size.
448
          __num_bins = static_cast<_BinIndex>
449
            (std::min(__n, static_cast<_DifferenceType>(__num_bins_cache)));
450
 
451
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_TLB
452
          // 2 TLB entries needed per bin
453
          __num_bins = std::min<_DifferenceType>(__s.TLB_size / 2, __num_bins);
454
#endif
455
          __num_bins = __round_up_to_pow2(__num_bins);
456
#if _GLIBCXX_RANDOM_SHUFFLE_CONSIDER_L1
457
        }
458
#endif
459
 
460
      int __num_bits = __rd_log2(__num_bins);
461
 
462
      if (__num_bins > 1)
463
        {
464
          _ValueType* __target =
465
            static_cast<_ValueType*>(::operator new(sizeof(_ValueType) * __n));
466
          _BinIndex* __oracles = new _BinIndex[__n];
467
          _DifferenceType* __dist0 = new _DifferenceType[__num_bins + 1],
468
                         * __dist1 = new _DifferenceType[__num_bins + 1];
469
 
470
          for (int __b = 0; __b < __num_bins + 1; ++__b)
471
            __dist0[__b] = 0;
472
 
473
          _RandomNumber __bitrng(__rng(0xFFFFFFFF));
474
 
475
          for (_DifferenceType __i = 0; __i < __n; ++__i)
476
            {
477
              _BinIndex __oracle = __random_number_pow2(__num_bits, __bitrng);
478
              __oracles[__i] = __oracle;
479
 
480
              // To allow prefix (partial) sum.
481
              ++(__dist0[__oracle + 1]);
482
            }
483
 
484
          // Sum up bins.
485
          __gnu_sequential::partial_sum(__dist0, __dist0 + __num_bins + 1,
486
                                        __dist0);
487
 
488
          for (int __b = 0; __b < __num_bins + 1; ++__b)
489
            __dist1[__b] = __dist0[__b];
490
 
491
          // Distribute according to oracles.
492
          for (_DifferenceType __i = 0; __i < __n; ++__i)
493
            ::new(&(__target[(__dist0[__oracles[__i]])++]))
494
                _ValueType(*(__begin + __i));
495
 
496
          for (int __b = 0; __b < __num_bins; ++__b)
497
            __sequential_random_shuffle(__target + __dist1[__b],
498
                                        __target + __dist1[__b + 1], __rng);
499
 
500
          // Copy elements back.
501
          std::copy(__target, __target + __n, __begin);
502
 
503
          delete[] __dist0;
504
          delete[] __dist1;
505
          delete[] __oracles;
506
 
507
          for (_DifferenceType __i = 0; __i < __n; ++__i)
508
            __target[__i].~_ValueType();
509
          ::operator delete(__target);
510
        }
511
      else
512
        __gnu_sequential::random_shuffle(__begin, __end, __rng);
513
    }
514
 
515
  /** @brief Parallel random public call.
516
   *  @param __begin Begin iterator of sequence.
517
   *  @param __end End iterator of sequence.
518
   *  @param __rng Random number generator to use.
519
   */
520
  template<typename _RAIter, typename _RandomNumberGenerator>
521
    inline void
522
    __parallel_random_shuffle(_RAIter __begin, _RAIter __end,
523
                              _RandomNumberGenerator __rng = _RandomNumber())
524
    {
525
      typedef std::iterator_traits<_RAIter> _TraitsType;
526
      typedef typename _TraitsType::difference_type _DifferenceType;
527
      _DifferenceType __n = __end - __begin;
528
      __parallel_random_shuffle_drs(__begin, __end, __n,
529
                                    __get_max_threads(), __rng);
530
    }
531
}
532
 
533
#endif /* _GLIBCXX_PARALLEL_RANDOM_SHUFFLE_H */

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.