OpenE2K
/
gcc
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
gcc/libstdc++-v3/doc/html/manual/bk01pt12ch31s04.html

98 lines
7.5 KiB
HTML
Raw Normal View History

*: Populate with regenerated files. 2008-02-11 Benjamin Kosnik <bkoz@redhat.com> * doc/html/*: Populate with regenerated files. From-SVN: r132251
2008-02-12 03:39:33 +01:00
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>Design</title><meta name="generator" content="DocBook XSL Stylesheets V1.73.2" /><meta name="keywords" content="&#10; C++&#10; , &#10; library&#10; , &#10; parallel&#10; " /><meta name="keywords" content="&#10; ISO C++&#10; , &#10; library&#10; " /><link rel="start" href="../spine.html" title="The GNU C++ Library Documentation" /><link rel="up" href="parallel_mode.html" title="Chapter 31. Parallel Mode" /><link rel="prev" href="bk01pt12ch31s03.html" title="Using" /><link rel="next" href="bk01pt12ch31s05.html" title="Testing" /></head><body><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">Design</th></tr><tr><td width="20%" align="left"><a accesskey="p" href="bk01pt12ch31s03.html">Prev</a> </td><th width="60%" align="center">Chapter 31. Parallel Mode</th><td width="20%" align="right"> <a accesskey="n" href="bk01pt12ch31s05.html">Next</a></td></tr></table><hr /></div><div class="sect1" lang="en" xml:lang="en"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="manual.ext.parallel_mode.design"></a>Design</h2></div></div></div><p>
</p><div class="sect2" lang="en" xml:lang="en"><div class="titlepage"><div><div><h3 class="title"><a id="manual.ext.parallel_mode.design.intro"></a>Interface Basics</h3></div></div></div><p>All parallel algorithms are intended to have signatures that are
equivalent to the ISO C++ algorithms replaced. For instance, the
<code class="code">std::adjacent_find</code> function is declared as:
</p><pre class="programlisting">
namespace std
{
template&lt;typename _FIter&gt;
_FIter
adjacent_find(_FIter, _FIter);
}
</pre><p>
Which means that there should be something equivalent for the parallel
version. Indeed, this is the case:
</p><pre class="programlisting">
namespace std
{
namespace __parallel
{
template&lt;typename _FIter&gt;
_FIter
adjacent_find(_FIter, _FIter);
...
}
}
</pre><p>But.... why the elipses?
</p><p> The elipses 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.
</p><p> Compile-time conditions are referred to as "embarrassingly
parallel," and are denoted with the appropriate dispatch object, ie
one of <code class="code">__gnu_parallel::sequential_tag</code>,
<code class="code">__gnu_parallel::parallel_tag</code>,
<code class="code">__gnu_parallel::balanced_tag</code>,
<code class="code">__gnu_parallel::unbalanced_tag</code>,
<code class="code">__gnu_parallel::omp_loop_tag</code>, or
<code class="code">__gnu_parallel::omp_loop_static_tag</code>.
</p><p> Run-time conditions depend on the hardware being used, the number
of threads available, etc., and are denoted by the use of the enum
<code class="code">__gnu_parallel::parallelism</code>. Values of this enum include
<code class="code">__gnu_parallel::sequential</code>,
<code class="code">__gnu_parallel::parallel_unbalanced</code>,
<code class="code">__gnu_parallel::parallel_balanced</code>,
<code class="code">__gnu_parallel::parallel_omp_loop</code>,
<code class="code">__gnu_parallel::parallel_omp_loop_static</code>, or
<code class="code">__gnu_parallel::parallel_taskqueue</code>.
</p><p> Putting all this together, the general view of overloads for the
parallel algorithms look like this:
</p><div class="itemizedlist"><ul type="disc"><li><p>ISO C++ signature</p></li><li><p>ISO C++ signature + sequential_tag argument</p></li><li><p>ISO C++ signature + parallelism argument</p></li></ul></div><p> Please note that the implementation may use additional functions
(designated with the <code class="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.
</p></div><div class="sect2" lang="en" xml:lang="en"><div class="titlepage"><div><div><h3 class="title"><a id="manual.ext.parallel_mode.design.tuning"></a>Configuration and Tuning</h3></div></div></div><p> Some algorithm variants can be enabled/disabled/selected at compile-time.
See <a class="ulink" href="latest-doxygen/compiletime__settings_8h.html" target="_top">
<code class="code">&lt;compiletime_settings.h&gt;</code></a> and
See <a class="ulink" href="latest-doxygen/compiletime__settings_8h.html" target="_top">
<code class="code">&lt;features.h&gt;</code></a> for details.
</p><p>
To specify the number of threads to be used for an algorithm,
use <code class="code">omp_set_num_threads</code>.
To force a function to execute sequentially,
even though parallelism is switched on in general,
add <code class="code">__gnu_parallel::sequential_tag()</code>
to the end of the argument list.
</p><p>
Parallelism always incurs some overhead. Thus, it is not
helpful to parallelize operations on very small sets of data.
There are measures to avoid parallelizing stuff that is not worth it.
For each algorithm, a minimum problem size can be stated,
usually using the variable
<code class="code">__gnu_parallel::Settings::[algorithm]_minimal_n</code>.
Please see <a class="ulink" href="latest-doxygen/settings_8h.html" target="_top">
<code class="code">&lt;settings.h&gt;</code></a> for details.</p></div><div class="sect2" lang="en" xml:lang="en"><div class="titlepage"><div><div><h3 class="title"><a id="manual.ext.parallel_mode.design.impl"></a>Implementation Namespaces</h3></div></div></div><p> One namespace contain versions of code that are explicitly sequential:
<code class="code">__gnu_serial</code>.
</p><p> Two namespaces contain the parallel mode:
<code class="code">std::__parallel</code> and <code class="code">__gnu_parallel</code>.
</p><p> Parallel implementations of standard components, including
template helpers to select parallelism, are defined in <code class="code">namespace
std::__parallel</code>. For instance, <code class="code">std::transform</code> from
&lt;algorithm&gt; has a parallel counterpart in
<code class="code">std::__parallel::transform</code> from
&lt;parallel/algorithm&gt;. In addition, these parallel
implementations are injected into <code class="code">namespace
__gnu_parallel</code> with using declarations.
</p><p> Support and general infrastructure is in <code class="code">namespace
__gnu_parallel</code>.
</p><p> 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.
</p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="bk01pt12ch31s03.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="parallel_mode.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="bk01pt12ch31s05.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">Using </td><td width="20%" align="center"><a accesskey="h" href="../spine.html">Home</a></td><td width="40%" align="right" valign="top"> Testing</td></tr></table></div></body></html>