diff options
Diffstat (limited to 'libstdc++-v3/docs/html')
| -rw-r--r-- | libstdc++-v3/docs/html/parallel_mode.html | 71 |
1 files changed, 61 insertions, 10 deletions
diff --git a/libstdc++-v3/docs/html/parallel_mode.html b/libstdc++-v3/docs/html/parallel_mode.html index 5843ae8c3d1..0ada39b6de6 100644 --- a/libstdc++-v3/docs/html/parallel_mode.html +++ b/libstdc++-v3/docs/html/parallel_mode.html @@ -35,7 +35,7 @@ implementation of many algorithms the C++ Standard Library. <p> Several of the standard algorithms, for instance -<code>std::search</code>, are made parallel using OpenMP +<code>std::sort</code>, are made parallel using OpenMP annotations. These parallel mode constructs and can be invoked by explicit source declaration or by compiling existing sources with a specific compiler flag. @@ -43,7 +43,7 @@ specific compiler flag. <h3 class="left"><a name="parallel">The libstdc++ parallel mode</a></h3> -<p>The libstdc++ parallel mode performs parallization of algorithms, +<p>The libstdc++ parallel mode performs parallelization of algorithms, function objects, classes, and functions in the C++ Standard.</p> <h4 class="left">Using the libstdc++ parallel mode</h4> @@ -53,7 +53,7 @@ function objects, classes, and functions in the C++ Standard.</p> will link in <code>libgomp</code>, the GNU OpenMP <a href="http://gcc.gnu.org/onlinedocs/libgomp">implementation</a>, whose presence is mandatory. In addition, hardware capable of atomic - operations is de rigueur. Actually activating these atomic + operations is mandatory. Actually activating these atomic operations may require explicit compiler flags on some targets (like sparc and x86), such as <code>-march=i686</code>, <code>-march=native</code> or <code>-mcpu=v9</code>. @@ -113,6 +113,13 @@ function objects, classes, and functions in the C++ Standard.</p> <li><code>std::unique_copy</code></li> </ul> +<p>The following library components in the includes +<code><set></code> and <code><map></code> are included in the parallel mode:</p> +<ul> + <li><code>std::(multi_)map/set<T>::(multi_)map/set(Iterator begin, Iterator end)</code> (bulk construction)</li> + <li><code>std::(multi_)map/set<T>::insert(Iterator begin, Iterator end)</code> (bulk insertion)</li> +</ul> + <h4 class="left">Using the parallel algorithms without parallel mode</h4> @@ -380,13 +387,47 @@ function objects, classes, and functions in the C++ Standard.</p> <h4 class="left">Parallel mode semantics</h4> -<p> Something about exception safety, interaction with threads, -etc. Goal is to have the usual constraints of the STL with respect to -exception safety and threads, but add in support for parallel -computing.</p> -<p> Something about compile-time settings and configuration, ie using -<code>__gnu_parallel::Settings</code>. XXX Up in the air.</p> +<p> The parallel mode STL algorithms are currently not exception-safe, +i. e. user-defined functors must not throw exceptions. +</p> + +<p> 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.</p> + + +<h4 class="left">Configuration and Tuning</h4> + +<p> Some algorithm variants can be enabled/disabled/selected at compile-time. +See <a href="latest-doxygen/compiletime__settings_8h.html"> +<code><compiletime_settings.h></code></a> and +See <a href="latest-doxygen/compiletime__settings_8h.html"> +<code><features.h></code></a> for details. +</p> + +<p> +To specify the number of threads to be used for an algorithm, +use <code>omp_set_num_threads</code>. +To force a function to execute sequentially, +even though parallelism is switched on in general, +add <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>__gnu_parallel::Settings::[algorithm]_minimal_n</code>. +Please see <a href="latest-doxygen/settings_8h.html"> +<code><settings.h></code><a> for details.</p> + + <h4 class="left">Interface basics and general design</h4> @@ -485,7 +526,7 @@ std::__parallel</code>. For instance, <code>std::transform</code> from <algorithm> has a parallel counterpart in <code>std::__parallel::transform</code> from <parallel/algorithm>. In addition, these parallel -implementatations are injected into <code>namespace +implementations are injected into <code>namespace __gnu_parallel</code> with using declarations. </p> @@ -526,6 +567,16 @@ testsuite's Makefile.</p> </p> +<h4 class="left">References / Further Reading</h4> + +<p> +Johannes Singler, Peter Sanders, Felix Putze. The Multi-Core Standard Template Library. Euro-Par 2007: Parallel Processing. (LNCS 4641) +</p> + +<p> +Leonor Frias, Johannes Singler: Parallelization of Bulk Operations for STL Dictionaries. Workshop on Highly Parallel Processing on a Chip (HPPC) 2007. (LNCS) +</p> + <!-- ####################################################### --> <hr /> |
