<chapter xmlns="http://docbook.org/ns/docbook" version="5.0"
	 xml:id="manual.ext.parallel_mode" xreflabel="Parallel Mode">
<?dbhtml filename="parallel_mode.html"?>

<info><title>Parallel Mode</title>
  <keywordset>
    <keyword>C++</keyword>
    <keyword>library</keyword>
    <keyword>parallel</keyword>
  </keywordset>
</info>


<para> The libstdc++ parallel mode is an experimental parallel
implementation of many algorithms of the C++ Standard Library.
</para>

<para>
Several of the standard algorithms, for instance
<function>std::sort</function>, are made parallel using OpenMP
annotations. These parallel mode constructs can be invoked by
explicit source declaration or by compiling existing sources with a
specific compiler flag.
</para>

<note>
  <para>
    The parallel mode has not been kept up to date with recent C++ standards
    and so it only conforms to the C++03 requirements.
    That means that move-only predicates may not work with parallel mode
    algorithms, and for C++20 most of the algorithms cannot be used in
    <code>constexpr</code> functions.
  </para>
  <para>
    For C++17 and above there are new overloads of the standard algorithms
    which take an execution policy argument. You should consider using those
    instead of the non-standard parallel mode extensions.
  </para>
</note>

<section xml:id="manual.ext.parallel_mode.intro" xreflabel="Intro"><info><title>Intro</title></info>


<para>The following library components in the include
<filename class="headerfile">numeric</filename> are included in the parallel mode:</para>
<itemizedlist>
  <listitem><para><function>std::accumulate</function></para></listitem>
  <listitem><para><function>std::adjacent_difference</function></para></listitem>
  <listitem><para><function>std::inner_product</function></para></listitem>
  <listitem><para><function>std::partial_sum</function></para></listitem>
</itemizedlist>

<para>The following library components in the include
<filename class="headerfile">algorithm</filename> are included in the parallel mode:</para>
<itemizedlist>
  <listitem><para><function>std::adjacent_find</function></para></listitem>
  <listitem><para><function>std::count</function></para></listitem>
  <listitem><para><function>std::count_if</function></para></listitem>
  <listitem><para><function>std::equal</function></para></listitem>
  <listitem><para><function>std::find</function></para></listitem>
  <listitem><para><function>std::find_if</function></para></listitem>
  <listitem><para><function>std::find_first_of</function></para></listitem>
  <listitem><para><function>std::for_each</function></para></listitem>
  <listitem><para><function>std::generate</function></para></listitem>
  <listitem><para><function>std::generate_n</function></para></listitem>
  <listitem><para><function>std::lexicographical_compare</function></para></listitem>
  <listitem><para><function>std::mismatch</function></para></listitem>
  <listitem><para><function>std::search</function></para></listitem>
  <listitem><para><function>std::search_n</function></para></listitem>
  <listitem><para><function>std::transform</function></para></listitem>
  <listitem><para><function>std::replace</function></para></listitem>
  <listitem><para><function>std::replace_if</function></para></listitem>
  <listitem><para><function>std::max_element</function></para></listitem>
  <listitem><para><function>std::merge</function></para></listitem>
  <listitem><para><function>std::min_element</function></para></listitem>
  <listitem><para><function>std::nth_element</function></para></listitem>
  <listitem><para><function>std::partial_sort</function></para></listitem>
  <listitem><para><function>std::partition</function></para></listitem>
  <listitem><para><function>std::random_shuffle</function></para></listitem>
  <listitem><para><function>std::set_union</function></para></listitem>
  <listitem><para><function>std::set_intersection</function></para></listitem>
  <listitem><para><function>std::set_symmetric_difference</function></para></listitem>
  <listitem><para><function>std::set_difference</function></para></listitem>
  <listitem><para><function>std::sort</function></para></listitem>
  <listitem><para><function>std::stable_sort</function></para></listitem>
  <listitem><para><function>std::unique_copy</function></para></listitem>
</itemizedlist>

</section>

<section xml:id="manual.ext.parallel_mode.semantics" xreflabel="Semantics"><info><title>Semantics</title></info>
<?dbhtml filename="parallel_mode_semantics.html"?>


<para> The parallel mode STL algorithms are currently not exception-safe,
i.e. user-defined functors must not throw exceptions.
Also, the order of execution is not guaranteed for some functions, of course.
Therefore, user-defined functors should not have any concurrent side effects.
</para>

<para> Since the current GCC OpenMP implementation does not support
OpenMP parallel regions in concurrent threads,
it is not possible to call parallel STL algorithm in
concurrent threads, either.
It might work with other compilers, though.</para>

</section>

<section xml:id="manual.ext.parallel_mode.using" xreflabel="Using"><info><title>Using</title></info>
<?dbhtml filename="parallel_mode_using.html"?>


<section xml:id="parallel_mode.using.prereq_flags"><info><title>Prerequisite Compiler Flags</title></info>


<para>
  Any use of parallel functionality requires additional compiler
  and runtime support, in particular support for OpenMP. Adding this support is
  not difficult: just compile your application with the compiler
  flag <literal>-fopenmp</literal>. This will link
  in <code>libgomp</code>, the
  <link xmlns:xlink="http://www.w3.org/1999/xlink"
    xlink:href="http://gcc.gnu.org/onlinedocs/libgomp/">GNU Offloading and
    Multi Processing Runtime Library</link>,
  whose presence is mandatory.
</para>

<para>
In addition, hardware that supports atomic operations and a compiler
  capable of producing atomic operations is mandatory: GCC defaults to no
  support for atomic operations on some common hardware
  architectures. Activating atomic operations may require explicit
  compiler flags on some targets (like sparc and x86), such
  as <literal>-march=i686</literal>,
  <literal>-march=native</literal> or <literal>-mcpu=v9</literal>. See
  the GCC manual for more information.
</para>

</section>

<section xml:id="parallel_mode.using.parallel_mode"><info><title>Using Parallel Mode</title></info>


<para>
  To use the libstdc++ parallel mode, compile your application with
  the prerequisite flags as detailed above, and in addition
  add <constant>-D_GLIBCXX_PARALLEL</constant>. This will convert all
  use of the standard (sequential) algorithms to the appropriate parallel
  equivalents. Please note that this doesn't necessarily mean that
  everything will end up being executed in a parallel manner, but
  rather that the heuristics and settings coded into the parallel
  versions will be used to determine if all, some, or no algorithms
  will be executed using parallel variants.
</para>

<para>Note that the <constant>_GLIBCXX_PARALLEL</constant> define may change the
  sizes and behavior of standard class templates such as
  <function>std::search</function>, and therefore one can only link code
  compiled with parallel mode and code compiled without parallel mode
  if no instantiation of a container is passed between the two
  translation units. Parallel mode functionality has distinct linkage,
  and cannot be confused with normal mode symbols.
</para>
</section>

<section xml:id="parallel_mode.using.specific"><info><title>Using Specific Parallel Components</title></info>


<para>When it is not feasible to recompile your entire application, or
  only specific algorithms need to be parallel-aware, individual
  parallel algorithms can be made available explicitly. These
  parallel algorithms are functionally equivalent to the standard
  drop-in algorithms used in parallel mode, but they are available in
  a separate namespace as GNU extensions and may be used in programs
  compiled with either release mode or with parallel mode.
</para>


<para>An example of using a parallel version
of <function>std::sort</function>, but no other parallel algorithms, is:
</para>

<programlisting>
#include &lt;vector&gt;
#include &lt;parallel/algorithm&gt;

int main()
{
  std::vector&lt;int&gt; v(100);

  // ...

  // Explicitly force a call to parallel sort.
  __gnu_parallel::sort(v.begin(), v.end());
  return 0;
}
</programlisting>

<para>
Then compile this code with the prerequisite compiler flags
(<literal>-fopenmp</literal> and any necessary architecture-specific
flags for atomic operations.)
</para>

<para> The following table provides the names and headers of all the
  parallel algorithms that can be used in a similar manner:
</para>

<table frame="all" xml:id="table.parallel_algos">
<title>Parallel Algorithms</title>

<tgroup cols="4" align="left" colsep="1" rowsep="1">
<colspec colname="c1"/>
<colspec colname="c2"/>
<colspec colname="c3"/>
<colspec colname="c4"/>

<thead>
  <row>
    <entry>Algorithm</entry>
    <entry>Header</entry>
    <entry>Parallel algorithm</entry>
    <entry>Parallel header</entry>
  </row>
</thead>

<tbody>
  <row>
    <entry><function>std::accumulate</function></entry>
    <entry><filename class="headerfile">numeric</filename></entry>
    <entry><function>__gnu_parallel::accumulate</function></entry>
    <entry><filename class="headerfile">parallel/numeric</filename></entry>
  </row>
  <row>
    <entry><function>std::adjacent_difference</function></entry>
    <entry><filename class="headerfile">numeric</filename></entry>
    <entry><function>__gnu_parallel::adjacent_difference</function></entry>
    <entry><filename class="headerfile">parallel/numeric</filename></entry>
  </row>
  <row>
    <entry><function>std::inner_product</function></entry>
    <entry><filename class="headerfile">numeric</filename></entry>
    <entry><function>__gnu_parallel::inner_product</function></entry>
    <entry><filename class="headerfile">parallel/numeric</filename></entry>
  </row>
  <row>
    <entry><function>std::partial_sum</function></entry>
    <entry><filename class="headerfile">numeric</filename></entry>
    <entry><function>__gnu_parallel::partial_sum</function></entry>
    <entry><filename class="headerfile">parallel/numeric</filename></entry>
  </row>
  <row>
    <entry><function>std::adjacent_find</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::adjacent_find</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::count</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::count</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::count_if</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::count_if</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::equal</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::equal</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::find</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::find</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::find_if</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::find_if</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::find_first_of</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::find_first_of</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::for_each</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::for_each</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::generate</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::generate</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::generate_n</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::generate_n</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::lexicographical_compare</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::lexicographical_compare</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::mismatch</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::mismatch</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::search</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::search</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::search_n</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::search_n</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::transform</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::transform</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::replace</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::replace</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::replace_if</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::replace_if</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::max_element</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::max_element</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::merge</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::merge</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::min_element</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::min_element</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::nth_element</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::nth_element</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::partial_sort</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::partial_sort</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::partition</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::partition</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::random_shuffle</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::random_shuffle</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::set_union</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::set_union</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::set_intersection</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::set_intersection</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::set_symmetric_difference</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::set_symmetric_difference</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::set_difference</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::set_difference</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::sort</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::sort</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::stable_sort</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::stable_sort</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>

  <row>
    <entry><function>std::unique_copy</function></entry>
    <entry><filename class="headerfile">algorithm</filename></entry>
    <entry><function>__gnu_parallel::unique_copy</function></entry>
    <entry><filename class="headerfile">parallel/algorithm</filename></entry>
  </row>
</tbody>
</tgroup>
</table>

</section>

</section>

<section xml:id="manual.ext.parallel_mode.design" xreflabel="Design"><info><title>Design</title></info>
<?dbhtml filename="parallel_mode_design.html"?>

  <para>
  </para>
<section xml:id="parallel_mode.design.intro" xreflabel="Intro"><info><title>Interface Basics</title></info>


<para>
All parallel algorithms are intended to have signatures that are
equivalent to the ISO C++ algorithms replaced. For instance, the
<function>std::adjacent_find</function> function is declared as:
</para>
<programlisting>
namespace std
{
  template&lt;typename _FIter&gt;
    _FIter
    adjacent_find(_FIter, _FIter);
}
</programlisting>

<para>
Which means that there should be something equivalent for the parallel
version. Indeed, this is the case:
</para>

<programlisting>
namespace std
{
  namespace __parallel
  {
    template&lt;typename _FIter&gt;
      _FIter
      adjacent_find(_FIter, _FIter);

    ...
  }
}
</programlisting>

<para>But.... why the ellipses?
</para>

<para> The ellipses in the example above represent additional overloads
required for the parallel version of the function. These additional
overloads are used to dispatch calls from the ISO C++ function
signature to the appropriate parallel function (or sequential
function, if no parallel functions are deemed worthy), based on either
compile-time or run-time conditions.
</para>

<para> The available signature options are specific for the different
algorithms/algorithm classes.</para>

<para> The general view of overloads for the parallel algorithms look like this:
</para>
<itemizedlist>
   <listitem><para>ISO C++ signature</para></listitem>
   <listitem><para>ISO C++ signature + sequential_tag argument</para></listitem>
   <listitem><para>ISO C++ signature + algorithm-specific tag type
    (several signatures)</para></listitem>
</itemizedlist>

<para> Please note that the implementation may use additional functions
(designated with the <code>_switch</code> suffix) to dispatch from the
ISO C++ signature to the correct parallel version. Also, some of the
algorithms do not have support for run-time conditions, so the last
overload is therefore missing.
</para>


</section>

<section xml:id="parallel_mode.design.tuning" xreflabel="Tuning"><info><title>Configuration and Tuning</title></info>


<section xml:id="parallel_mode.design.tuning.omp" xreflabel="OpenMP Environment"><info><title>Setting up the OpenMP Environment</title></info>


<para>
Several aspects of the overall runtime environment can be manipulated
by standard OpenMP function calls.
</para>

<para>
To specify the number of threads to be used for the algorithms globally,
use the function <function>omp_set_num_threads</function>. An example:
</para>

<programlisting>
#include &lt;stdlib.h&gt;
#include &lt;omp.h&gt;

int main()
{
  // Explicitly set number of threads.
  const int threads_wanted = 20;
  omp_set_dynamic(false);
  omp_set_num_threads(threads_wanted);

  // Call parallel mode algorithms.

  return 0;
}
</programlisting>

<para>
 Some algorithms allow the number of threads being set for a particular call,
 by augmenting the algorithm variant.
 See the next section for further information.
</para>

<para>
Other parts of the runtime environment able to be manipulated include
nested parallelism (<function>omp_set_nested</function>), schedule kind
(<function>omp_set_schedule</function>), and others. See the OpenMP
documentation for more information.
</para>

</section>

<section xml:id="parallel_mode.design.tuning.compile" xreflabel="Compile Switches"><info><title>Compile Time Switches</title></info>


<para>
To force an algorithm to execute sequentially, even though parallelism
is switched on in general via the macro <constant>_GLIBCXX_PARALLEL</constant>,
add <classname>__gnu_parallel::sequential_tag()</classname> to the end
of the algorithm's argument list.
</para>

<para>
Like so:
</para>

<programlisting>
std::sort(v.begin(), v.end(), __gnu_parallel::sequential_tag());
</programlisting>

<para>
Some parallel algorithm variants can be excluded from compilation by
preprocessor defines. See the doxygen documentation on
<code>compiletime_settings.h</code> and <code>features.h</code> for details.
</para>

<para>
For some algorithms, the desired variant can be chosen at compile-time by
appending a tag object. The available options are specific to the particular
algorithm (class).
</para>

<para>
For the "embarrassingly parallel" algorithms, there is only one "tag object
type", the enum _Parallelism.
It takes one of the following values,
<code>__gnu_parallel::parallel_tag</code>,
<code>__gnu_parallel::balanced_tag</code>,
<code>__gnu_parallel::unbalanced_tag</code>,
<code>__gnu_parallel::omp_loop_tag</code>,
<code>__gnu_parallel::omp_loop_static_tag</code>.
This means that the actual parallelization strategy is chosen at run-time.
(Choosing the variants at compile-time will come soon.)
</para>

<para>
For the following algorithms in general, we have
<code>__gnu_parallel::parallel_tag</code> and
<code>__gnu_parallel::default_parallel_tag</code>, in addition to
<code>__gnu_parallel::sequential_tag</code>.
<code>__gnu_parallel::default_parallel_tag</code> chooses the default
algorithm at compiletime, as does omitting the tag.
<code>__gnu_parallel::parallel_tag</code> postpones the decision to runtime
(see next section).
For all tags, the number of threads desired for this call can optionally be
passed to the respective tag's constructor.
</para>

<para>
The <code>multiway_merge</code> algorithm comes with the additional choices,
<code>__gnu_parallel::exact_tag</code> and
<code>__gnu_parallel::sampling_tag</code>.
Exact and sampling are the two available splitting strategies.
</para>

<para>
For the <code>sort</code> and <code>stable_sort</code> algorithms, there are
several additional choices, namely
<code>__gnu_parallel::multiway_mergesort_tag</code>,
<code>__gnu_parallel::multiway_mergesort_exact_tag</code>,
<code>__gnu_parallel::multiway_mergesort_sampling_tag</code>,
<code>__gnu_parallel::quicksort_tag</code>, and
<code>__gnu_parallel::balanced_quicksort_tag</code>.
Multiway mergesort comes with the two splitting strategies for multi-way
merging. The quicksort options cannot be used for <code>stable_sort</code>.
</para>

</section>

<section xml:id="parallel_mode.design.tuning.settings" xreflabel="_Settings"><info><title>Run Time Settings and Defaults</title></info>


<para>
The default parallelization strategy, the choice of specific algorithm
strategy, the minimum threshold limits for individual parallel
algorithms, and aspects of the underlying hardware can be specified as
desired via manipulation
of <classname>__gnu_parallel::_Settings</classname> member data.
</para>

<para>
First off, the choice of parallelization strategy: serial, parallel,
or heuristically deduced. This corresponds
to <code>__gnu_parallel::_Settings::algorithm_strategy</code> and is a
value of enum <type>__gnu_parallel::_AlgorithmStrategy</type>
type. Choices
include: <type>heuristic</type>, <type>force_sequential</type>,
and <type>force_parallel</type>. The default is <type>heuristic</type>.
</para>


<para>
Next, the sub-choices for algorithm variant, if not fixed at compile-time.
Specific algorithms like <function>find</function> or <function>sort</function>
can be implemented in multiple ways: when this is the case,
a <classname>__gnu_parallel::_Settings</classname> member exists to
pick the default strategy. For
example, <code>__gnu_parallel::_Settings::sort_algorithm</code> can
have any values of
enum <type>__gnu_parallel::_SortAlgorithm</type>: <type>MWMS</type>, <type>QS</type>,
or <type>QS_BALANCED</type>.
</para>

<para>
Likewise for setting the minimal threshold for algorithm
parallelization.  Parallelism always incurs some overhead. Thus, it is
not helpful to parallelize operations on very small sets of
data. Because of this, measures are taken to avoid parallelizing below
a certain, pre-determined threshold. For each algorithm, a minimum
problem size is encoded as a variable in the
active <classname>__gnu_parallel::_Settings</classname> object.  This
threshold variable follows the following naming scheme:
<code>__gnu_parallel::_Settings::[algorithm]_minimal_n</code>.  So,
for <function>fill</function>, the threshold variable
is <code>__gnu_parallel::_Settings::fill_minimal_n</code>,
</para>

<para>
Finally, hardware details like L1/L2 cache size can be hardwired
via <code>__gnu_parallel::_Settings::L1_cache_size</code> and friends.
</para>

<para>
</para>

<para>
All these configuration variables can be changed by the user, if
desired.
There exists one global instance of the class <classname>_Settings</classname>,
i. e. it is a singleton. It can be read and written by calling
<code>__gnu_parallel::_Settings::get</code> and
<code>__gnu_parallel::_Settings::set</code>, respectively.
Please note that the first call return a const object, so direct manipulation
is forbidden.
See <link xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="http://gcc.gnu.org/onlinedocs/libstdc++/latest-doxygen/index.html">
  <filename class="headerfile">&lt;parallel/settings.h&gt;</filename></link>
for complete details.
</para>

<para>
A small example of tuning the default:
</para>

<programlisting>
#include &lt;parallel/algorithm&gt;
#include &lt;parallel/settings.h&gt;

int main()
{
  __gnu_parallel::_Settings s;
  s.algorithm_strategy = __gnu_parallel::force_parallel;
  __gnu_parallel::_Settings::set(s);

  // Do work... all algorithms will be parallelized, always.

  return 0;
}
</programlisting>

</section>

</section>

<section xml:id="parallel_mode.design.impl" xreflabel="Impl"><info><title>Implementation Namespaces</title></info>


<para> One namespace contain versions of code that are always
explicitly sequential:
<code>__gnu_serial</code>.
</para>

<para> Two namespaces contain the parallel mode:
<code>std::__parallel</code> and <code>__gnu_parallel</code>.
</para>

<para> Parallel implementations of standard components, including
template helpers to select parallelism, are defined in <code>namespace
std::__parallel</code>. For instance, <function>std::transform</function> from <filename class="headerfile">algorithm</filename> has a parallel counterpart in
<function>std::__parallel::transform</function> from <filename class="headerfile">parallel/algorithm</filename>. In addition, these parallel
implementations are injected into <code>namespace
__gnu_parallel</code> with using declarations.
</para>

<para> Support and general infrastructure is in <code>namespace
__gnu_parallel</code>.
</para>

<para> More information, and an organized index of types and functions
related to the parallel mode on a per-namespace basis, can be found in
the generated source documentation.
</para>

</section>

</section>

<section xml:id="manual.ext.parallel_mode.test" xreflabel="Testing"><info><title>Testing</title></info>
<?dbhtml filename="parallel_mode_test.html"?>


  <para>
    Both the normal conformance and regression tests and the
    supplemental performance tests work.
  </para>

  <para>
    To run the conformance and regression tests with the parallel mode
    active,
  </para>

  <screen>
  <userinput>make check-parallel</userinput>
  </screen>

  <para>
    The log and summary files for conformance testing are in the
    <filename class="directory">testsuite/parallel</filename> directory.
  </para>

  <para>
    To run the performance tests with the parallel mode active,
  </para>

  <screen>
  <userinput>make check-performance-parallel</userinput>
  </screen>

  <para>
    The result file for performance testing are in the
    <filename class="directory">testsuite</filename> directory, in the file
    <filename>libstdc++_performance.sum</filename>. In addition, the
    policy-based containers have their own visualizations, which have
    additional software dependencies than the usual bare-boned text
    file, and can be generated by using the <code>make
    doc-performance</code> rule in the testsuite's Makefile.
</para>
</section>

<bibliography xml:id="parallel_mode.biblio"><info><title>Bibliography</title></info>


  <biblioentry>
    <citetitle>
      Parallelization of Bulk Operations for STL Dictionaries
    </citetitle>

    <author><personname><firstname>Johannes</firstname><surname>Singler</surname></personname></author>
    <author><personname><firstname>Leonor</firstname><surname>Frias</surname></personname></author>

    <copyright>
      <year>2007</year>
      <holder/>
    </copyright>

    <publisher>
      <publishername>
	Workshop on Highly Parallel Processing on a Chip (HPPC) 2007. (LNCS)
      </publishername>
    </publisher>
  </biblioentry>

  <biblioentry>
    <citetitle>
      The Multi-Core Standard Template Library
    </citetitle>

    <author><personname><firstname>Johannes</firstname><surname>Singler</surname></personname></author>
    <author><personname><firstname>Peter</firstname><surname>Sanders</surname></personname></author>
    <author><personname><firstname>Felix</firstname><surname>Putze</surname></personname></author>

    <copyright>
      <year>2007</year>
      <holder/>
    </copyright>

    <publisher>
      <publishername>
	 Euro-Par 2007: Parallel Processing. (LNCS 4641)
      </publishername>
    </publisher>
  </biblioentry>

</bibliography>

</chapter>