Home | History | Annotate | Download | only in parallel
      1 // -*- C++ -*-
      2 
      3 // Copyright (C) 2007, 2008, 2009 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  *  This file is a GNU parallel extension to the Standard C++ Library.
     29  *
     30  *  @section parallelization_decision
     31  *  The decision 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  *
     65  *  a. force_sequential, meaning the sequential algorithm is executed.
     66  *  b. force_parallel, meaning the parallel algorithm is executed.
     67  *  c. heuristic
     68  *
     69  *  For heuristic, the parallel algorithm implementation is called
     70  *  only if the input size is sufficiently large.  For most
     71  *  algorithms, the input size is the (combined) length of the input
     72  *  sequence(s).  The threshold can be set by the user, individually
     73  *  for each algorithm.  The according variables are called
     74  *  __gnu_parallel::_Settings::[algorithm]_minimal_n .
     75  *
     76  *  For some of the algorithms, there are even more tuning options,
     77  *  e. g. the ability to choose from multiple algorithm variants.  See
     78  *  below for details.
     79  */
     80 
     81 // Written by Johannes Singler and Felix Putze.
     82 
     83 #ifndef _GLIBCXX_PARALLEL_SETTINGS_H
     84 #define _GLIBCXX_PARALLEL_SETTINGS_H 1
     85 
     86 #include <parallel/types.h>
     87 
     88 /**
     89   * @brief Determine at compile(?)-time if the parallel variant of an
     90   * algorithm should be called.
     91   * @param c A condition that is convertible to bool that is overruled by
     92   * __gnu_parallel::_Settings::algorithm_strategy. Usually a decision
     93   * based on the input size.
     94   */
     95 #define _GLIBCXX_PARALLEL_CONDITION(c) (__gnu_parallel::_Settings::get().algorithm_strategy != __gnu_parallel::force_sequential && ((__gnu_parallel::get_max_threads() > 1 && (c)) || __gnu_parallel::_Settings::get().algorithm_strategy == __gnu_parallel::force_parallel))
     96 
     97 /*
     98 inline bool
     99 parallel_condition(bool c)
    100 {
    101   bool ret = false;
    102   const _Settings& s = _Settings::get();
    103   if (s.algorithm_strategy != force_seqential)
    104     {
    105       if (s.algorithm_strategy == force_parallel)
    106 	ret = true;
    107       else
    108 	ret = get_max_threads() > 1 && c;
    109     }
    110   return ret;
    111 }
    112 */
    113 
    114 namespace __gnu_parallel
    115 {
    116   /// class _Settings
    117   /// Run-time settings for the parallel mode, including all tunable parameters.
    118   struct _Settings
    119   {
    120     _AlgorithmStrategy 		algorithm_strategy;
    121 
    122     _SortAlgorithm 		sort_algorithm;
    123     _PartialSumAlgorithm 	partial_sum_algorithm;
    124     _MultiwayMergeAlgorithm 	multiway_merge_algorithm;
    125     _FindAlgorithm 		find_algorithm;
    126 
    127     _SplittingAlgorithm 	sort_splitting;
    128     _SplittingAlgorithm 	merge_splitting;
    129     _SplittingAlgorithm 	multiway_merge_splitting;
    130 
    131     // Per-algorithm settings.
    132 
    133     /// Minimal input size for accumulate.
    134     sequence_index_t 		accumulate_minimal_n;
    135 
    136     /// Minimal input size for adjacent_difference.
    137     unsigned int 		adjacent_difference_minimal_n;
    138 
    139     /// Minimal input size for count and count_if.
    140     sequence_index_t 		count_minimal_n;
    141 
    142     /// Minimal input size for fill.
    143     sequence_index_t 		fill_minimal_n;
    144 
    145     /// Block size increase factor for find.
    146     double 			find_increasing_factor;
    147 
    148     /// Initial block size for find.
    149     sequence_index_t 		find_initial_block_size;
    150 
    151     /// Maximal block size for find.
    152     sequence_index_t 		find_maximum_block_size;
    153 
    154     /// Start with looking for this many elements sequentially, for find.
    155     sequence_index_t 		find_sequential_search_size;
    156 
    157     /// Minimal input size for for_each.
    158     sequence_index_t 		for_each_minimal_n;
    159 
    160     /// Minimal input size for generate.
    161     sequence_index_t 		generate_minimal_n;
    162 
    163     /// Minimal input size for max_element.
    164     sequence_index_t 		max_element_minimal_n;
    165 
    166     /// Minimal input size for merge.
    167     sequence_index_t 		merge_minimal_n;
    168 
    169     /// Oversampling factor for merge.
    170     unsigned int 		merge_oversampling;
    171 
    172     /// Minimal input size for min_element.
    173     sequence_index_t 		min_element_minimal_n;
    174 
    175     /// Minimal input size for multiway_merge.
    176     sequence_index_t 		multiway_merge_minimal_n;
    177 
    178     /// Oversampling factor for multiway_merge.
    179     int 			multiway_merge_minimal_k;
    180 
    181     /// Oversampling factor for multiway_merge.
    182     unsigned int 		multiway_merge_oversampling;
    183 
    184     /// Minimal input size for nth_element.
    185     sequence_index_t 		nth_element_minimal_n;
    186 
    187     /// Chunk size for partition.
    188     sequence_index_t 		partition_chunk_size;
    189 
    190     /// Chunk size for partition, relative to input size.  If > 0.0,
    191     /// this value overrides partition_chunk_size.
    192     double 			partition_chunk_share;
    193 
    194     /// Minimal input size for partition.
    195     sequence_index_t 		partition_minimal_n;
    196 
    197     /// Minimal input size for partial_sort.
    198     sequence_index_t 		partial_sort_minimal_n;
    199 
    200     /// Ratio for partial_sum. Assume "sum and write result" to be
    201     /// this factor slower than just "sum".
    202     float 			partial_sum_dilation;
    203 
    204     /// Minimal input size for partial_sum.
    205     unsigned int 		partial_sum_minimal_n;
    206 
    207     /// Minimal input size for random_shuffle.
    208     unsigned int 		random_shuffle_minimal_n;
    209 
    210     /// Minimal input size for replace and replace_if.
    211     sequence_index_t 		replace_minimal_n;
    212 
    213     /// Minimal input size for set_difference.
    214     sequence_index_t 		set_difference_minimal_n;
    215 
    216     /// Minimal input size for set_intersection.
    217     sequence_index_t 		set_intersection_minimal_n;
    218 
    219     /// Minimal input size for set_symmetric_difference.
    220     sequence_index_t 		set_symmetric_difference_minimal_n;
    221 
    222     /// Minimal input size for set_union.
    223     sequence_index_t 		set_union_minimal_n;
    224 
    225     /// Minimal input size for parallel sorting.
    226     sequence_index_t 		sort_minimal_n;
    227 
    228     /// Oversampling factor for parallel std::sort (MWMS).
    229     unsigned int 		sort_mwms_oversampling;
    230 
    231     /// Such many samples to take to find a good pivot (quicksort).
    232     unsigned int 		sort_qs_num_samples_preset;
    233 
    234     /// Maximal subsequence length to switch to unbalanced base case.
    235     /// Applies to std::sort with dynamically load-balanced quicksort.
    236     sequence_index_t 		sort_qsb_base_case_maximal_n;
    237 
    238     /// Minimal input size for parallel std::transform.
    239     sequence_index_t 		transform_minimal_n;
    240 
    241     /// Minimal input size for unique_copy.
    242     sequence_index_t 		unique_copy_minimal_n;
    243 
    244     sequence_index_t 		workstealing_chunk_size;
    245 
    246     // Hardware dependent tuning parameters.
    247 
    248     /// Size of the L1 cache in bytes (underestimation).
    249     unsigned long long 		L1_cache_size;
    250 
    251     /// Size of the L2 cache in bytes (underestimation).
    252     unsigned long long 		L2_cache_size;
    253 
    254     /// Size of the Translation Lookaside Buffer (underestimation).
    255     unsigned int 		TLB_size;
    256 
    257     /// Overestimation of cache line size.  Used to avoid false
    258     /// sharing, i. e. elements of different threads are at least this
    259     /// amount apart.
    260     unsigned int 		cache_line_size;
    261 
    262     // Statistics.
    263 
    264     /// The number of stolen ranges in load-balanced quicksort.
    265     sequence_index_t 		qsb_steals;
    266 
    267     /// Get the global settings.
    268     static const _Settings&
    269     get() throw();
    270 
    271     /// Set the global settings.
    272     static void
    273     set(_Settings&) throw();
    274 
    275     explicit
    276     _Settings() : algorithm_strategy(heuristic), sort_algorithm(MWMS), partial_sum_algorithm(LINEAR), multiway_merge_algorithm(LOSER_TREE), find_algorithm(CONSTANT_SIZE_BLOCKS), sort_splitting(EXACT), merge_splitting(EXACT), multiway_merge_splitting(EXACT), accumulate_minimal_n(1000), adjacent_difference_minimal_n(1000), count_minimal_n(1000), fill_minimal_n(1000), find_increasing_factor(2.0), find_initial_block_size(256), find_maximum_block_size(8192), find_sequential_search_size(256), for_each_minimal_n(1000), generate_minimal_n(1000), max_element_minimal_n(1000), merge_minimal_n(1000), merge_oversampling(10), min_element_minimal_n(1000), multiway_merge_minimal_n(1000), multiway_merge_minimal_k(2), multiway_merge_oversampling(10), nth_element_minimal_n(1000), partition_chunk_size(1000), partition_chunk_share(0.0), partition_minimal_n(1000), partial_sort_minimal_n(1000), partial_sum_dilation(1.0f), partial_sum_minimal_n(1000), random_shuffle_minimal_n(1000), replace_minimal_n(1000), set_difference_minimal_n(1000), set_intersection_minimal_n(1000), set_symmetric_difference_minimal_n(1000), set_union_minimal_n(1000), sort_minimal_n(1000), sort_mwms_oversampling(10), sort_qs_num_samples_preset(100), sort_qsb_base_case_maximal_n(100), transform_minimal_n(1000), unique_copy_minimal_n(10000), workstealing_chunk_size(100), L1_cache_size(16 << 10), L2_cache_size(256 << 10), TLB_size(128), cache_line_size(64), qsb_steals(0)
    277     { }
    278   };
    279 }
    280 
    281 #endif /* _GLIBCXX_PARALLEL_SETTINGS_H */
    282