]> oss.titaniummirror.com Git - msp430-gcc.git/blobdiff - libstdc++-v3/include/parallel/par_loop.h
Imported gcc-4.4.3
[msp430-gcc.git] / libstdc++-v3 / include / parallel / par_loop.h
diff --git a/libstdc++-v3/include/parallel/par_loop.h b/libstdc++-v3/include/parallel/par_loop.h
new file mode 100644 (file)
index 0000000..be61d4c
--- /dev/null
@@ -0,0 +1,135 @@
+// -*- C++ -*-
+
+// Copyright (C) 2007, 2008, 2009 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 3, 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.
+
+// Under Section 7 of GPL version 3, you are granted additional
+// permissions described in the GCC Runtime Library Exception, version
+// 3.1, as published by the Free Software Foundation.
+
+// You should have received a copy of the GNU General Public License and
+// a copy of the GCC Runtime Library Exception along with this program;
+// see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+// <http://www.gnu.org/licenses/>.
+
+/** @file parallel/par_loop.h
+ *  @brief Parallelization of embarrassingly parallel execution by
+ *  means of equal splitting.
+ *  This file is a GNU parallel extension to the Standard C++ Library.
+ */
+
+// Written by Felix Putze.
+
+#ifndef _GLIBCXX_PARALLEL_PAR_LOOP_H
+#define _GLIBCXX_PARALLEL_PAR_LOOP_H 1
+
+#include <omp.h>
+#include <parallel/settings.h>
+#include <parallel/base.h>
+#include <parallel/equally_split.h>
+
+namespace __gnu_parallel
+{
+
+/** @brief Embarrassingly parallel algorithm for random access
+  * iterators, using hand-crafted parallelization by equal splitting
+  * the work.
+  *
+  *  @param begin Begin iterator of element sequence.
+  *  @param end End iterator of element sequence.
+  *  @param o User-supplied functor (comparator, predicate, adding
+  *  functor, ...)
+  *  @param f Functor to "process" an element with op (depends on
+  *  desired functionality, e. g. for std::for_each(), ...).
+  *  @param r Functor to "add" a single result to the already
+  *  processed elements (depends on functionality).
+  *  @param base Base value for reduction.
+  *  @param output Pointer to position where final result is written to
+  *  @param bound Maximum number of elements processed (e. g. for
+  *  std::count_n()).
+  *  @return User-supplied functor (that may contain a part of the result).
+  */
+template<typename RandomAccessIterator,
+        typename Op,
+        typename Fu,
+        typename Red,
+        typename Result>
+  Op
+  for_each_template_random_access_ed(RandomAccessIterator begin,
+                                    RandomAccessIterator end,
+                                    Op o, Fu& f, Red r, Result base,
+                                    Result& output,
+                                    typename std::iterator_traits
+                                    <RandomAccessIterator>::
+                                    difference_type bound)
+  {
+    typedef std::iterator_traits<RandomAccessIterator> traits_type;
+    typedef typename traits_type::difference_type difference_type;
+    const difference_type length = end - begin;
+    Result *thread_results;
+    bool* constructed;
+
+    thread_index_t num_threads =
+      __gnu_parallel::min<difference_type>(get_max_threads(), length);
+
+#   pragma omp parallel num_threads(num_threads)
+      {
+#       pragma omp single
+          {
+            num_threads = omp_get_num_threads();
+            thread_results = static_cast<Result*>(
+                                ::operator new(num_threads * sizeof(Result)));
+            constructed = new bool[num_threads];
+          }
+
+        thread_index_t iam = omp_get_thread_num();
+
+        // Neutral element.
+        Result* reduct = static_cast<Result*>(::operator new(sizeof(Result)));
+
+        difference_type
+            start = equally_split_point(length, num_threads, iam),
+            stop = equally_split_point(length, num_threads, iam + 1);
+
+        if (start < stop)
+          {
+            new(reduct) Result(f(o, begin + start));
+            ++start;
+            constructed[iam] = true;
+          }
+        else
+          constructed[iam] = false;
+
+        for (; start < stop; ++start)
+          *reduct = r(*reduct, f(o, begin + start));
+
+        thread_results[iam] = *reduct;
+      } //parallel
+
+    for (thread_index_t i = 0; i < num_threads; ++i)
+        if (constructed[i])
+            output = r(output, thread_results[i]);
+
+    // Points to last element processed (needed as return value for
+    // some algorithms like transform).
+    f.finish_iterator = begin + length;
+
+    delete[] thread_results;
+    delete[] constructed;
+
+    return o;
+  }
+
+} // end namespace
+
+#endif /* _GLIBCXX_PARALLEL_PAR_LOOP_H */