
2007-09-11 Johannes Singler <singler@ira.uka.de> Leonor Frias Moya <lfrias@lsi.upc.edu> Felix Putze <kontakt@felix-putze.de> Marius Elvert <marius.elvert@ira.uka.de> Felix Bondarenko <f.bondarenko@web.de> Robert Geisberger <robert.geisberger@stud.uni-karlsruhe.de> Robin Dapp <r.dapp@freenet.de> Benjamin Kosnik <bkoz@redhat.com> Add parallel mode. * include/parallel: New. * include/parallel/iterator.h: New. * include/parallel/multiway_merge.h: New. * include/parallel/parallel.h: New. * include/parallel/algorithm * include/parallel/find_selectors.h: New. * include/parallel/losertree.h: New. * include/parallel/list_partition.h: New. * include/parallel/types.h: New. * include/parallel/for_each.h: New. * include/parallel/multiseq_selection.h: New. * include/parallel/workstealing.h: New. * include/parallel/base.h: New. * include/parallel/par_loop.h: New. * include/parallel/numeric * include/parallel/features.h: New. * include/parallel/quicksort.h: New. * include/parallel/algorithmfwd.h: New. * include/parallel/equally_split.h: New. * include/parallel/compiletime_settings.h: New. * include/parallel/for_each_selectors.h: New. * include/parallel/basic_iterator.h: New. * include/parallel/omp_loop_static.h: New. * include/parallel/random_shuffle.h: New. * include/parallel/balanced_quicksort.h: New. * include/parallel/set_operations.h: New. * include/parallel/tags.h: New. * include/parallel/merge.h: New. * include/parallel/tree.h: New. * include/parallel/settings.h: New. * include/parallel/unique_copy.h: New. * include/parallel/multiway_mergesort.h: New. * include/parallel/numericfwd.h: New. * include/parallel/search.h: New. * include/parallel/partition.h: New. * include/parallel/compatibility.h: New. * include/parallel/algobase.h: New. * include/parallel/find.h: New. * include/parallel/partial_sum.h: New. * include/parallel/algo.h: New. * include/parallel/omp_loop.h: New. * include/parallel/queue.h: New. * include/parallel/timing.h: New. * include/parallel/sort.h: New. * include/parallel/checkers.h: New. * include/parallel/random_number.h: New. * include/bits/algorithmfwd.h: New. * acinclude.m4 (GLIBCXX_ENABLE_PARALLEL): New. * configure.host: Add atomic_flags. * configure.ac: Export ATOMIC_FLAGS, call GLIBCXX_ENABLE_PARALLEL. * src/Makefile.am: Add parallel_list rules. * include/Makefile.am: Add parallel files. * testsuite/Makefile.am (check-parallel): Add. (check-performance-parallel): Add. * config.h.in: Regenerate. * configure: Same. * libsupc++/Makefile.in: Same. * testsuite/Makefile.in: Same. * Makefile.in: Same. * libmath/Makefile.in: Same. * include/Makefile.in: Same. * src/Makefile.in: Same. * po/Makefile.in: Same. * config/abi/pre/gnu.ver: Export parallel list bits. * docs/html/parallel_mode.html: New. * docs/html/documentation.html: Add link. * docs/doxygen/user.cfg.in: Adjust for new files and directory. * docs/doxygen/doxygroups.cc: Adjust namespace markup. * include/debug/set.h: Adjust for _GLIBCXX_STD_D or _P change. * include/debug/bitset: Same. * include/debug/multiset.h: Same. * include/debug/vector: Same. * include/debug/map.h: Same. * include/debug/deque: Same. * include/debug/list: Same. * include/debug/debug.h: Same. * include/debug/multimap.h: Same. * include/std/algorithm: Same. * include/std/numeric: Same. * include/std/bitset: Same. * include/std/string: Same. * include/ext/hash_map: Same. * include/ext/hash_set: Same. * include/bits/stl_list.h: Same. * include/bits/stl_map.h: Same. * include/bits/stl_algobase.h: Same. * include/bits/stl_set.h: Same. * include/bits/stl_multimap.h: Same. * include/bits/stl_vector.h: Same. * include/bits/stl_numeric.h: Same. * include/bits/stl_deque.h: Same. * include/bits/stl_multiset.h: Same. * include/bits/char_traits.h: Same. * include/bits/stl_algo.h: Same. * include/bits/c++config: Same. * include/bits/vector.tcc: Same. * include/bits/deque.tcc: Same. * include/bits/stl_bvector.h: Same. * include/bits/list.tcc: Same. * src/list.cc: Same. * src/parallel_list.cc: New. * testsuite/lib/libstdc++.exp (check_v3_target_parallel_mode): New. * testsuite/lib/dg-options.exp (dg-require-parallel-mode): New. * scripts/testsuite_flags.in (--cxxparallelflags): New. * scripts/check_performance: Adjust. * testsuite/25_algorithms/headers/parallel_algorithm.cc: New. * testsuite/25_algorithms/headers/algorithm_parallel_mode.cc: New. * testsuite/25_algorithms/headers/parallel_algorithm_mixed1.cc: New. * testsuite/25_algorithms/headers/parallel_algorithm_mixed2.cc: New. * testsuite/26_numerics/headers/numeric/parallel_numeric.cc: New. * testsuite/26_numerics/headers/numeric/numeric_parallel_mode.cc: New. * testsuite/26_numerics/headers/numeric/ parallel_numeric_mixed1.cc: New. * testsuite/26_numerics/headers/numeric/ parallel_numeric_mixed2.cc: New. Co-Authored-By: Benjamin Kosnik <bkoz@redhat.com> Co-Authored-By: Felix Bondarenko <f.bondarenko@web.de> Co-Authored-By: Felix Putze <kontakt@felix-putze.de> Co-Authored-By: Leonor Frias Moya <lfrias@lsi.upc.edu> Co-Authored-By: Marius Elvert <marius.elvert@ira.uka.de> Co-Authored-By: Robert Geisberger <robert.geisberger@stud.uni-karlsruhe.de> Co-Authored-By: Robin Dapp <r.dapp@freenet.de> From-SVN: r128395
181 lines
6 KiB
C++
181 lines
6 KiB
C++
// -*- C++ -*-
|
|
|
|
// Copyright (C) 2007 Free Software Foundation, Inc.
|
|
//
|
|
// This file is part of the GNU ISO C++ Library. This library is free
|
|
// software; you can redistribute it and/or modify it under the terms
|
|
// of the GNU General Public License as published by the Free Software
|
|
// Foundation; either version 2, or (at your option) any later
|
|
// version.
|
|
|
|
// This library is distributed in the hope that it will be useful, but
|
|
// WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
// General Public License for more details.
|
|
|
|
// You should have received a copy of the GNU General Public License
|
|
// along with this library; see the file COPYING. If not, write to
|
|
// the Free Software Foundation, 59 Temple Place - Suite 330, Boston,
|
|
// MA 02111-1307, USA.
|
|
|
|
// As a special exception, you may use this file as part of a free
|
|
// software library without restriction. Specifically, if other files
|
|
// instantiate templates or use macros or inline functions from this
|
|
// file, or you compile this file and link it with other files to
|
|
// produce an executable, this file does not by itself cause the
|
|
// resulting executable to be covered by the GNU General Public
|
|
// License. This exception does not however invalidate any other
|
|
// reasons why the executable file might be covered by the GNU General
|
|
// Public License.
|
|
|
|
/** @file parallel/list_partition.h
|
|
* @brief Functionality to split sequence referenced by only input
|
|
* iterators.
|
|
* This file is a GNU parallel extension to the Standard C++ Library.
|
|
*/
|
|
|
|
// Written by Leonor Frias Moya and Johannes Singler.
|
|
|
|
#ifndef _GLIBCXX_PARALLEL_LIST_PARTITION_H
|
|
#define _GLIBCXX_PARALLEL_LIST_PARTITION_H 1
|
|
|
|
#include <parallel/parallel.h>
|
|
#include <vector>
|
|
|
|
namespace __gnu_parallel
|
|
{
|
|
/** @brief Shrinks and doubles the ranges.
|
|
* @param os_starts Start positions worked on (oversampled).
|
|
* @param count_to_two Counts up to 2.
|
|
* @param range_length Current length of a chunk.
|
|
* @param make_twice Whether the @c os_starts is allowed to be
|
|
* grown or not
|
|
*/
|
|
template<typename InputIterator>
|
|
void
|
|
shrink_and_double(std::vector<InputIterator>& os_starts, size_t& count_to_two, size_t& range_length, const bool make_twice)
|
|
{
|
|
++count_to_two;
|
|
if (not make_twice or count_to_two < 2)
|
|
{
|
|
shrink(os_starts, count_to_two, range_length);
|
|
}
|
|
else
|
|
{
|
|
os_starts.resize((os_starts.size() - 1) * 2 + 1);
|
|
count_to_two = 0;
|
|
}
|
|
}
|
|
|
|
/** @brief Combines two ranges into one and thus halves the number of ranges.
|
|
* @param os_starts Start positions worked on (oversampled).
|
|
* @param count_to_two Counts up to 2.
|
|
* @param range_length Current length of a chunk. */
|
|
template<typename InputIterator>
|
|
void
|
|
shrink(std::vector<InputIterator>& os_starts, size_t& count_to_two,
|
|
size_t& range_length)
|
|
{
|
|
for (typename std::vector<InputIterator>::size_type i = 0; i <= (os_starts.size() / 2); ++i)
|
|
{
|
|
os_starts[i] = os_starts[i * 2];
|
|
}
|
|
range_length *= 2;
|
|
}
|
|
|
|
/** @brief Splits a sequence given by input iterators into parts of
|
|
* almost equal size
|
|
*
|
|
* The function needs only one pass over the sequence.
|
|
* @param begin Begin iterator of input sequence.
|
|
* @param end End iterator of input sequence.
|
|
* @param starts Start iterators for the resulting parts, dimension
|
|
* @c num_parts+1. For convenience, @c starts @c [num_parts]
|
|
* contains the end iterator of the sequence.
|
|
* @param lengths Length of the resulting parts.
|
|
* @param num_parts Number of parts to split the sequence into.
|
|
* @param f Functor to be applied to each element by traversing it
|
|
* @param oversampling Oversampling factor. If 0, then the
|
|
* partitions will differ in at most @f$ \sqrt{\mathrm{end} -
|
|
* \mathrm{begin}} @f$ elements. Otherwise, the ratio between the
|
|
* longest and the shortest part is bounded by @f$
|
|
* 1/(\mathrm{oversampling} \cdot \mathrm{num\_parts}) @f$.
|
|
* @return Length of the whole sequence.
|
|
*/
|
|
template<typename InputIterator, typename FunctorType>
|
|
size_t
|
|
list_partition(const InputIterator begin, const InputIterator end,
|
|
InputIterator* starts, size_t* lengths, const int num_parts,
|
|
FunctorType& f, int oversampling = 0)
|
|
{
|
|
bool make_twice = false;
|
|
|
|
// According to the oversampling factor, the resizing algorithm is chosen.
|
|
if (oversampling == 0)
|
|
{
|
|
make_twice = true;
|
|
oversampling = 1;
|
|
}
|
|
|
|
std::vector<InputIterator> os_starts(2 * oversampling * num_parts + 1);
|
|
|
|
os_starts[0]= begin;
|
|
InputIterator prev = begin, it = begin;
|
|
size_t dist_limit = 0, dist = 0;
|
|
size_t cur = 1, next = 1;
|
|
size_t range_length = 1;
|
|
size_t count_to_two = 0;
|
|
while (it != end){
|
|
cur = next;
|
|
for (; cur < os_starts.size() and it != end; ++cur)
|
|
{
|
|
for (dist_limit += range_length; dist < dist_limit and it != end; ++dist)
|
|
{
|
|
f(it);
|
|
++it;
|
|
}
|
|
os_starts[cur] = it;
|
|
}
|
|
|
|
// Must compare for end and not cur < os_starts.size() , because
|
|
// cur could be == os_starts.size() as well
|
|
if (it == end)
|
|
break;
|
|
|
|
shrink_and_double(os_starts, count_to_two, range_length, make_twice);
|
|
next = os_starts.size()/2 + 1;
|
|
}
|
|
|
|
// Calculation of the parts (one must be extracted from current
|
|
// because the partition beginning at end, consists only of
|
|
// itself).
|
|
size_t size_part = (cur - 1) / num_parts;
|
|
int size_greater = static_cast<int>((cur - 1) % num_parts);
|
|
starts[0] = os_starts[0];
|
|
|
|
size_t index = 0;
|
|
|
|
// Smallest partitions.
|
|
for (int i = 1; i < (num_parts + 1 - size_greater); ++i)
|
|
{
|
|
lengths[i-1] = size_part * range_length;
|
|
index += size_part;
|
|
starts[i] = os_starts[index];
|
|
}
|
|
|
|
// Biggest partitions.
|
|
for (int i = num_parts + 1 - size_greater; i <= num_parts; ++i)
|
|
{
|
|
lengths[i-1] = (size_part+1) * range_length;
|
|
index += (size_part+1);
|
|
starts[i] = os_starts[index];
|
|
}
|
|
|
|
// Correction of the end size (the end iteration has not finished).
|
|
lengths[num_parts - 1] -= (dist_limit - dist);
|
|
|
|
return dist;
|
|
}
|
|
}
|
|
|
|
#endif
|