libstdc++-v3/include/parallel/settings.h

   1 // -*- C++ -*-
   2
   3 // Copyright (C) 2007-2024 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/settings.h
  26  *  @brief Runtime settings and tuning parameters, heuristics to decide
  27  *  whether to use parallelized algorithms.
  28  *
  29  *  This file is a GNU parallel extension to the Standard C++ Library.
  30  *
  31  *  @section parallelization_decision Deciding whether to run an algorithm in parallel.
  32  *
  33  *  There are several ways the user can switch on and off the parallel
  34  *  execution of an algorithm, both at compile- and run-time.
  35  *
  36  *  Only sequential execution can be forced at compile-time.  This
  37  *  reduces code size and protects code parts that have
  38  *  non-thread-safe side effects.
  39  *
  40  *  Ultimately, forcing parallel execution at compile-time makes
  41  *  sense.  Often, the sequential algorithm implementation is used as
  42  *  a subroutine, so no reduction in code size can be achieved.  Also,
  43  *  the machine the program is run on might have only one processor
  44  *  core, so to avoid overhead, the algorithm is executed
  45  *  sequentially.
  46  *
  47  *  To force sequential execution of an algorithm ultimately at
  48  *  compile-time, the user must add the tag
  49 *  gnu_parallel::sequential_tag() to the end of the parameter list,
  50  *  e. g.
  51  *
  52  *  \code
  53  *  std::sort(__v.begin(), __v.end(), __gnu_parallel::sequential_tag());
  54  *  \endcode
  55  *
  56  *  This is compatible with all overloaded algorithm variants.  No
  57  *  additional code will be instantiated, at all.  The same holds for
  58  *  most algorithm calls with iterators not providing random access.
  59  *
  60  *  If the algorithm call is not forced to be executed sequentially
  61  *  at compile-time, the decision is made at run-time.
  62  *  The global variable __gnu_parallel::_Settings::algorithm_strategy
  63  *  is checked. It is a tristate variable corresponding to:
  64  *    - a. force_sequential, meaning the sequential algorithm is executed.
  65  *    - b. force_parallel, meaning the parallel algorithm is executed.
  66  *    - c. heuristic
  67  *
  68  *  For heuristic, the parallel algorithm implementation is called
  69  *  only if the input size is sufficiently large.  For most
  70  *  algorithms, the input size is the (combined) length of the input
  71  *  sequence(__s).  The threshold can be set by the user, individually
  72  *  for each algorithm.  The according variables are called
  73  *  gnu_parallel::_Settings::[algorithm]_minimal_n .
  74  *
  75  *  For some of the algorithms, there are even more tuning options,
  76  *  e. g. the ability to choose from multiple algorithm variants.  See
  77  *  below for details.
  78  */
  79
  80 // Written by Johannes Singler and Felix Putze.
  81
  82 #ifndef _GLIBCXX_PARALLEL_SETTINGS_H
  83 #define _GLIBCXX_PARALLEL_SETTINGS_H 1
  84
  85 #include <parallel/types.h>
  86
  87 /**
  88   * @brief Determine at compile(?)-time if the parallel variant of an
  89   * algorithm should be called.
  90   * @param __c A condition that is convertible to bool that is overruled by
  91   * __gnu_parallel::_Settings::algorithm_strategy. Usually a decision
  92   * based on the input size.
  93   */
  94 #define _GLIBCXX_PARALLEL_CONDITION(__c) \
  95   (__gnu_parallel::_Settings::get().algorithm_strategy \
  96     != __gnu_parallel::force_sequential \
  97   && ((__gnu_parallel::__get_max_threads() > 1 && (__c)) \
  98      || __gnu_parallel::_Settings::get().algorithm_strategy \
  99         == __gnu_parallel::force_parallel))
 100
 101 /*
 102 inline bool
 103 parallel_condition(bool __c)
 104 {
 105   bool ret = false;
 106   const _Settings& __s = _Settings::get();
 107   if (__s.algorithm_strategy != force_seqential)
 108     {
 109       if (__s.algorithm_strategy == force_parallel)
 110         ret = true;
 111       else
 112         ret = __get_max_threads() > 1 && __c;
 113     }
 114   return ret;
 115 }
 116 */
 117
 118 namespace __gnu_parallel
 119 {
 120   /// class _Settings
 121   /// Run-time settings for the parallel mode including all tunable parameters.
 122   struct _Settings
 123   {
 124     _AlgorithmStrategy          algorithm_strategy;
 125
 126     _SortAlgorithm              sort_algorithm;
 127     _PartialSumAlgorithm        partial_sum_algorithm;
 128     _MultiwayMergeAlgorithm     multiway_merge_algorithm;
 129     _FindAlgorithm              find_algorithm;
 130
 131     _SplittingAlgorithm         sort_splitting;
 132     _SplittingAlgorithm         merge_splitting;
 133     _SplittingAlgorithm         multiway_merge_splitting;
 134
 135     // Per-algorithm settings.
 136
 137     /// Minimal input size for accumulate.
 138     _SequenceIndex              accumulate_minimal_n;
 139
 140     /// Minimal input size for adjacent_difference.
 141     unsigned int                adjacent_difference_minimal_n;
 142
 143     /// Minimal input size for count and count_if.
 144     _SequenceIndex              count_minimal_n;
 145
 146     /// Minimal input size for fill.
 147     _SequenceIndex              fill_minimal_n;
 148
 149     /// Block size increase factor for find.
 150     double                      find_increasing_factor;
 151
 152     /// Initial block size for find.
 153     _SequenceIndex              find_initial_block_size;
 154
 155     /// Maximal block size for find.
 156     _SequenceIndex              find_maximum_block_size;
 157
 158     /// Start with looking for this many elements sequentially, for find.
 159     _SequenceIndex              find_sequential_search_size;
 160
 161     /// Minimal input size for for_each.
 162     _SequenceIndex              for_each_minimal_n;
 163
 164     /// Minimal input size for generate.
 165     _SequenceIndex              generate_minimal_n;
 166
 167     /// Minimal input size for max_element.
 168     _SequenceIndex              max_element_minimal_n;
 169
 170     /// Minimal input size for merge.
 171     _SequenceIndex              merge_minimal_n;
 172
 173     /// Oversampling factor for merge.
 174     unsigned int                merge_oversampling;
 175
 176     /// Minimal input size for min_element.
 177     _SequenceIndex              min_element_minimal_n;
 178
 179     /// Minimal input size for multiway_merge.
 180     _SequenceIndex              multiway_merge_minimal_n;
 181
 182     /// Oversampling factor for multiway_merge.
 183     int                         multiway_merge_minimal_k;
 184
 185     /// Oversampling factor for multiway_merge.
 186     unsigned int                multiway_merge_oversampling;
 187
 188     /// Minimal input size for nth_element.
 189     _SequenceIndex              nth_element_minimal_n;
 190
 191     /// Chunk size for partition.
 192     _SequenceIndex              partition_chunk_size;
 193
 194     /// Chunk size for partition, relative to input size.  If > 0.0,
 195     /// this value overrides partition_chunk_size.
 196     double                      partition_chunk_share;
 197
 198     /// Minimal input size for partition.
 199     _SequenceIndex              partition_minimal_n;
 200
 201     /// Minimal input size for partial_sort.
 202     _SequenceIndex              partial_sort_minimal_n;
 203
 204     /// Ratio for partial_sum. Assume "sum and write result" to be
 205     /// this factor slower than just "sum".
 206     float                       partial_sum_dilation;
 207
 208     /// Minimal input size for partial_sum.
 209     unsigned int                partial_sum_minimal_n;
 210
 211     /// Minimal input size for random_shuffle.
 212     unsigned int                random_shuffle_minimal_n;
 213
 214     /// Minimal input size for replace and replace_if.
 215     _SequenceIndex              replace_minimal_n;
 216
 217     /// Minimal input size for set_difference.
 218     _SequenceIndex              set_difference_minimal_n;
 219
 220     /// Minimal input size for set_intersection.
 221     _SequenceIndex              set_intersection_minimal_n;
 222
 223     /// Minimal input size for set_symmetric_difference.
 224     _SequenceIndex              set_symmetric_difference_minimal_n;
 225
 226     /// Minimal input size for set_union.
 227     _SequenceIndex              set_union_minimal_n;
 228
 229     /// Minimal input size for parallel sorting.
 230     _SequenceIndex              sort_minimal_n;
 231
 232     /// Oversampling factor for parallel std::sort (MWMS).
 233     unsigned int                sort_mwms_oversampling;
 234
 235     /// Such many samples to take to find a good pivot (quicksort).
 236     unsigned int                sort_qs_num_samples_preset;
 237
 238     /// Maximal subsequence __length to switch to unbalanced __base case.
 239     /// Applies to std::sort with dynamically load-balanced quicksort.
 240     _SequenceIndex              sort_qsb_base_case_maximal_n;
 241
 242     /// Minimal input size for parallel std::transform.
 243     _SequenceIndex              transform_minimal_n;
 244
 245     /// Minimal input size for unique_copy.
 246     _SequenceIndex              unique_copy_minimal_n;
 247
 248     _SequenceIndex              workstealing_chunk_size;
 249
 250     // Hardware dependent tuning parameters.
 251
 252     /// size of the L1 cache in bytes (underestimation).
 253     unsigned long long          L1_cache_size;
 254
 255     /// size of the L2 cache in bytes (underestimation).
 256     unsigned long long          L2_cache_size;
 257
 258     /// size of the Translation Lookaside Buffer (underestimation).
 259     unsigned int                TLB_size;
 260
 261     /// Overestimation of cache line size.  Used to avoid false
 262     /// sharing, i.e. elements of different threads are at least this
 263     /// amount apart.
 264     unsigned int                cache_line_size;
 265
 266     // Statistics.
 267
 268     /// The number of stolen ranges in load-balanced quicksort.
 269     _SequenceIndex              qsb_steals;
 270
 271     /// Minimal input size for search and search_n.
 272     _SequenceIndex              search_minimal_n;
 273
 274     /// Block size scale-down factor with respect to current position.
 275     float                       find_scale_factor;
 276
 277     /// Get the global settings.
 278     _GLIBCXX_CONST static const _Settings&
 279     get() throw();
 280
 281     /// Set the global settings.
 282     static void
 283     set(_Settings&) throw();
 284
 285     explicit
 286     _Settings() :
 287             algorithm_strategy(heuristic),
 288             sort_algorithm(MWMS),
 289             partial_sum_algorithm(LINEAR),
 290             multiway_merge_algorithm(LOSER_TREE),
 291             find_algorithm(CONSTANT_SIZE_BLOCKS),
 292             sort_splitting(EXACT),
 293             merge_splitting(EXACT),
 294             multiway_merge_splitting(EXACT),
 295             accumulate_minimal_n(1000),
 296             adjacent_difference_minimal_n(1000),
 297             count_minimal_n(1000),
 298             fill_minimal_n(1000),
 299             find_increasing_factor(2.0),
 300             find_initial_block_size(256),
 301             find_maximum_block_size(8192),
 302             find_sequential_search_size(256),
 303             for_each_minimal_n(1000),
 304             generate_minimal_n(1000),
 305             max_element_minimal_n(1000),
 306             merge_minimal_n(1000),
 307             merge_oversampling(10),
 308             min_element_minimal_n(1000),
 309             multiway_merge_minimal_n(1000),
 310             multiway_merge_minimal_k(2), multiway_merge_oversampling(10),
 311             nth_element_minimal_n(1000),
 312             partition_chunk_size(1000),
 313             partition_chunk_share(0.0),
 314             partition_minimal_n(1000),
 315             partial_sort_minimal_n(1000),
 316             partial_sum_dilation(1.0f),
 317             partial_sum_minimal_n(1000),
 318             random_shuffle_minimal_n(1000),
 319             replace_minimal_n(1000),
 320             set_difference_minimal_n(1000),
 321             set_intersection_minimal_n(1000),
 322             set_symmetric_difference_minimal_n(1000),
 323             set_union_minimal_n(1000),
 324             sort_minimal_n(1000),
 325             sort_mwms_oversampling(10),
 326             sort_qs_num_samples_preset(100),
 327             sort_qsb_base_case_maximal_n(100),
 328             transform_minimal_n(1000),
 329             unique_copy_minimal_n(10000),
 330             workstealing_chunk_size(100),
 331             L1_cache_size(16 << 10),
 332             L2_cache_size(256 << 10),
 333             TLB_size(128),
 334             cache_line_size(64),
 335             qsb_steals(0),
 336             search_minimal_n(1000),
 337             find_scale_factor(0.01f)
 338     { }
 339   };
 340 }
 341
 342 #endif /* _GLIBCXX_PARALLEL_SETTINGS_H */