1 <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
4 <meta http-equiv="Content-Type" content="text/html; charset=US-ASCII">
5 <title>Tutorial</title>
6 <link rel="stylesheet" href="../../../doc/src/boostbook.css" type="text/css">
7 <meta name="generator" content="DocBook XSL Stylesheets V1.79.1">
8 <link rel="home" href="../index.html" title="The Boost C++ Libraries BoostBook Documentation Subset">
9 <link rel="up" href="../mpi.html" title="Chapter 26. Boost.MPI">
10 <link rel="prev" href="getting_started.html" title="Getting started">
11 <link rel="next" href="c_mapping.html" title="Mapping from C MPI to Boost.MPI">
13 <body bgcolor="white" text="black" link="#0000FF" vlink="#840084" alink="#0000FF">
14 <table cellpadding="2" width="100%"><tr>
15 <td valign="top"><img alt="Boost C++ Libraries" width="277" height="86" src="../../../boost.png"></td>
16 <td align="center"><a href="../../../index.html">Home</a></td>
17 <td align="center"><a href="../../../libs/libraries.htm">Libraries</a></td>
18 <td align="center"><a href="http://www.boost.org/users/people.html">People</a></td>
19 <td align="center"><a href="http://www.boost.org/users/faq.html">FAQ</a></td>
20 <td align="center"><a href="../../../more/index.htm">More</a></td>
23 <div class="spirit-nav">
24 <a accesskey="p" href="getting_started.html"><img src="../../../doc/src/images/prev.png" alt="Prev"></a><a accesskey="u" href="../mpi.html"><img src="../../../doc/src/images/up.png" alt="Up"></a><a accesskey="h" href="../index.html"><img src="../../../doc/src/images/home.png" alt="Home"></a><a accesskey="n" href="c_mapping.html"><img src="../../../doc/src/images/next.png" alt="Next"></a>
27 <div class="titlepage"><div><div><h2 class="title" style="clear: both">
28 <a name="mpi.tutorial"></a><a class="link" href="tutorial.html" title="Tutorial">Tutorial</a>
29 </h2></div></div></div>
30 <div class="toc"><dl class="toc">
31 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.point_to_point">Point-to-Point communication</a></span></dt>
32 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.collectives">Collective operations</a></span></dt>
33 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.user_data_types">User-defined data types</a></span></dt>
34 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.communicators">Communicators</a></span></dt>
35 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.threading">Threads</a></span></dt>
36 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.skeleton_and_content">Separating structure
37 from content</a></span></dt>
38 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.performance_optimizations">Performance optimizations</a></span></dt>
41 A Boost.MPI program consists of many cooperating processes (possibly running
42 on different computers) that communicate among themselves by passing messages.
43 Boost.MPI is a library (as is the lower-level MPI), not a language, so the
44 first step in a Boost.MPI is to create an <code class="computeroutput"><a class="link" href="../boost/mpi/environment.html" title="Class environment">mpi::environment</a></code>
45 object that initializes the MPI environment and enables communication among
46 the processes. The <code class="computeroutput"><a class="link" href="../boost/mpi/environment.html" title="Class environment">mpi::environment</a></code>
47 object is initialized with the program arguments (which it may modify) in your
48 main program. The creation of this object initializes MPI, and its destruction
49 will finalize MPI. In the vast majority of Boost.MPI programs, an instance
50 of <code class="computeroutput"><a class="link" href="../boost/mpi/environment.html" title="Class environment">mpi::environment</a></code> will
51 be declared in <code class="computeroutput"><span class="identifier">main</span></code> at the
52 very beginning of the program.
54 <div class="warning"><table border="0" summary="Warning">
56 <td rowspan="2" align="center" valign="top" width="25"><img alt="[Warning]" src="../../../doc/src/images/warning.png"></td>
57 <th align="left">Warning</th>
59 <tr><td align="left" valign="top"><p>
60 Declaring an <code class="computeroutput"><a class="link" href="../boost/mpi/environment.html" title="Class environment">mpi::environment</a></code>
61 at global scope is undefined behavior. <a href="#ftn.mpi.tutorial.f0" class="footnote" name="mpi.tutorial.f0"><sup class="footnote">[11]</sup></a>
65 Communication with MPI always occurs over a <span class="bold"><strong>communicator</strong></span>,
66 which can be created be simply default-constructing an object of type <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html" title="Class communicator">mpi::communicator</a></code>. This communicator
67 can then be queried to determine how many processes are running (the "size"
68 of the communicator) and to give a unique number to each process, from zero
69 to the size of the communicator (i.e., the "rank" of the process):
71 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">environment</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
72 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">communicator</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
73 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
74 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
76 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
77 <span class="special">{</span>
78 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
79 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
80 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"I am process "</span> <span class="special"><<</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special"><<</span> <span class="string">" of "</span> <span class="special"><<</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">()</span>
81 <span class="special"><<</span> <span class="string">"."</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
82 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
83 <span class="special">}</span>
86 If you run this program with 7 processes, for instance, you will receive output
89 <pre class="programlisting">I am process 5 of 7.
98 Of course, the processes can execute in a different order each time, so the
99 ranks might not be strictly increasing. More interestingly, the text could
100 come out completely garbled, because one process can start writing "I
101 am a process" before another process has finished writing "of 7.".
104 If you should still have an MPI library supporting only MPI 1.1 you will need
105 to pass the command line arguments to the environment constructor as shown
108 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">environment</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
109 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">communicator</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
110 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
111 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
113 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">(</span><span class="keyword">int</span> <span class="identifier">argc</span><span class="special">,</span> <span class="keyword">char</span><span class="special">*</span> <span class="identifier">argv</span><span class="special">[])</span>
114 <span class="special">{</span>
115 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">(</span><span class="identifier">argc</span><span class="special">,</span> <span class="identifier">argv</span><span class="special">);</span>
116 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
117 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"I am process "</span> <span class="special"><<</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special"><<</span> <span class="string">" of "</span> <span class="special"><<</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">()</span>
118 <span class="special"><<</span> <span class="string">"."</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
119 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
120 <span class="special">}</span>
122 <div class="section">
123 <div class="titlepage"><div><div><h3 class="title">
124 <a name="mpi.tutorial.point_to_point"></a><a class="link" href="tutorial.html#mpi.tutorial.point_to_point" title="Point-to-Point communication">Point-to-Point communication</a>
125 </h3></div></div></div>
126 <div class="toc"><dl class="toc">
127 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.point_to_point.blocking">Blocking communication</a></span></dt>
128 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.point_to_point.nonblocking">Non-blocking
129 communication</a></span></dt>
131 <div class="section">
132 <div class="titlepage"><div><div><h4 class="title">
133 <a name="mpi.tutorial.point_to_point.blocking"></a><a class="link" href="tutorial.html#mpi.tutorial.point_to_point.blocking" title="Blocking communication">Blocking communication</a>
134 </h4></div></div></div>
136 As a message passing library, MPI's primary purpose is to routine messages
137 from one process to another, i.e., point-to-point. MPI contains routines
138 that can send messages, receive messages, and query whether messages are
139 available. Each message has a source process, a target process, a tag,
140 and a payload containing arbitrary data. The source and target processes
141 are the ranks of the sender and receiver of the message, respectively.
142 Tags are integers that allow the receiver to distinguish between different
143 messages coming from the same sender.
146 The following program uses two MPI processes to write "Hello, world!"
147 to the screen (<code class="computeroutput"><span class="identifier">hello_world</span><span class="special">.</span><span class="identifier">cpp</span></code>):
149 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
150 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
151 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">string</span><span class="special">></span>
152 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">serialization</span><span class="special">/</span><span class="identifier">string</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
153 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
155 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
156 <span class="special">{</span>
157 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
158 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
160 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
161 <span class="identifier">world</span><span class="special">.</span><span class="identifier">send</span><span class="special">(</span><span class="number">1</span><span class="special">,</span> <span class="number">0</span><span class="special">,</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">(</span><span class="string">"Hello"</span><span class="special">));</span>
162 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">msg</span><span class="special">;</span>
163 <span class="identifier">world</span><span class="special">.</span><span class="identifier">recv</span><span class="special">(</span><span class="number">1</span><span class="special">,</span> <span class="number">1</span><span class="special">,</span> <span class="identifier">msg</span><span class="special">);</span>
164 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="identifier">msg</span> <span class="special"><<</span> <span class="string">"!"</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
165 <span class="special">}</span> <span class="keyword">else</span> <span class="special">{</span>
166 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">msg</span><span class="special">;</span>
167 <span class="identifier">world</span><span class="special">.</span><span class="identifier">recv</span><span class="special">(</span><span class="number">0</span><span class="special">,</span> <span class="number">0</span><span class="special">,</span> <span class="identifier">msg</span><span class="special">);</span>
168 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="identifier">msg</span> <span class="special"><<</span> <span class="string">", "</span><span class="special">;</span>
169 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span><span class="special">.</span><span class="identifier">flush</span><span class="special">();</span>
170 <span class="identifier">world</span><span class="special">.</span><span class="identifier">send</span><span class="special">(</span><span class="number">0</span><span class="special">,</span> <span class="number">1</span><span class="special">,</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">(</span><span class="string">"world"</span><span class="special">));</span>
171 <span class="special">}</span>
173 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
174 <span class="special">}</span>
177 The first processor (rank 0) passes the message "Hello" to the
178 second processor (rank 1) using tag 0. The second processor prints the
179 string it receives, along with a comma, then passes the message "world"
180 back to processor 0 with a different tag. The first processor then writes
181 this message with the "!" and exits. All sends are accomplished
182 with the <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_4-bb">communicator::send</a></code>
183 method and all receives use a corresponding <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_9-bb">communicator::recv</a></code>
187 <div class="section">
188 <div class="titlepage"><div><div><h4 class="title">
189 <a name="mpi.tutorial.point_to_point.nonblocking"></a><a class="link" href="tutorial.html#mpi.tutorial.point_to_point.nonblocking" title="Non-blocking communication">Non-blocking
191 </h4></div></div></div>
193 The default MPI communication operations--<code class="computeroutput"><span class="identifier">send</span></code>
194 and <code class="computeroutput"><span class="identifier">recv</span></code>--may have to wait
195 until the entire transmission is completed before they can return. Sometimes
196 this <span class="bold"><strong>blocking</strong></span> behavior has a negative
197 impact on performance, because the sender could be performing useful computation
198 while it is waiting for the transmission to occur. More important, however,
199 are the cases where several communication operations must occur simultaneously,
200 e.g., a process will both send and receive at the same time.
203 Let's revisit our "Hello, world!" program from the previous
204 <a class="link" href="tutorial.html#mpi.tutorial.point_to_point.blocking" title="Blocking communication">section</a>. The
205 core of this program transmits two messages:
207 <pre class="programlisting"><span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
208 <span class="identifier">world</span><span class="special">.</span><span class="identifier">send</span><span class="special">(</span><span class="number">1</span><span class="special">,</span> <span class="number">0</span><span class="special">,</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">(</span><span class="string">"Hello"</span><span class="special">));</span>
209 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">msg</span><span class="special">;</span>
210 <span class="identifier">world</span><span class="special">.</span><span class="identifier">recv</span><span class="special">(</span><span class="number">1</span><span class="special">,</span> <span class="number">1</span><span class="special">,</span> <span class="identifier">msg</span><span class="special">);</span>
211 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="identifier">msg</span> <span class="special"><<</span> <span class="string">"!"</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
212 <span class="special">}</span> <span class="keyword">else</span> <span class="special">{</span>
213 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">msg</span><span class="special">;</span>
214 <span class="identifier">world</span><span class="special">.</span><span class="identifier">recv</span><span class="special">(</span><span class="number">0</span><span class="special">,</span> <span class="number">0</span><span class="special">,</span> <span class="identifier">msg</span><span class="special">);</span>
215 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="identifier">msg</span> <span class="special"><<</span> <span class="string">", "</span><span class="special">;</span>
216 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span><span class="special">.</span><span class="identifier">flush</span><span class="special">();</span>
217 <span class="identifier">world</span><span class="special">.</span><span class="identifier">send</span><span class="special">(</span><span class="number">0</span><span class="special">,</span> <span class="number">1</span><span class="special">,</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">(</span><span class="string">"world"</span><span class="special">));</span>
218 <span class="special">}</span>
221 The first process passes a message to the second process, then prepares
222 to receive a message. The second process does the send and receive in the
223 opposite order. However, this sequence of events is just that--a <span class="bold"><strong>sequence</strong></span>--meaning that there is essentially no parallelism.
224 We can use non-blocking communication to ensure that the two messages are
225 transmitted simultaneously (<code class="computeroutput"><span class="identifier">hello_world_nonblocking</span><span class="special">.</span><span class="identifier">cpp</span></code>):
227 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
228 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
229 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">string</span><span class="special">></span>
230 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">serialization</span><span class="special">/</span><span class="identifier">string</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
231 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
233 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
234 <span class="special">{</span>
235 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
236 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
238 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
239 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">request</span> <span class="identifier">reqs</span><span class="special">[</span><span class="number">2</span><span class="special">];</span>
240 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">msg</span><span class="special">,</span> <span class="identifier">out_msg</span> <span class="special">=</span> <span class="string">"Hello"</span><span class="special">;</span>
241 <span class="identifier">reqs</span><span class="special">[</span><span class="number">0</span><span class="special">]</span> <span class="special">=</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">isend</span><span class="special">(</span><span class="number">1</span><span class="special">,</span> <span class="number">0</span><span class="special">,</span> <span class="identifier">out_msg</span><span class="special">);</span>
242 <span class="identifier">reqs</span><span class="special">[</span><span class="number">1</span><span class="special">]</span> <span class="special">=</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">irecv</span><span class="special">(</span><span class="number">1</span><span class="special">,</span> <span class="number">1</span><span class="special">,</span> <span class="identifier">msg</span><span class="special">);</span>
243 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">wait_all</span><span class="special">(</span><span class="identifier">reqs</span><span class="special">,</span> <span class="identifier">reqs</span> <span class="special">+</span> <span class="number">2</span><span class="special">);</span>
244 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="identifier">msg</span> <span class="special"><<</span> <span class="string">"!"</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
245 <span class="special">}</span> <span class="keyword">else</span> <span class="special">{</span>
246 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">request</span> <span class="identifier">reqs</span><span class="special">[</span><span class="number">2</span><span class="special">];</span>
247 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">msg</span><span class="special">,</span> <span class="identifier">out_msg</span> <span class="special">=</span> <span class="string">"world"</span><span class="special">;</span>
248 <span class="identifier">reqs</span><span class="special">[</span><span class="number">0</span><span class="special">]</span> <span class="special">=</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">isend</span><span class="special">(</span><span class="number">0</span><span class="special">,</span> <span class="number">1</span><span class="special">,</span> <span class="identifier">out_msg</span><span class="special">);</span>
249 <span class="identifier">reqs</span><span class="special">[</span><span class="number">1</span><span class="special">]</span> <span class="special">=</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">irecv</span><span class="special">(</span><span class="number">0</span><span class="special">,</span> <span class="number">0</span><span class="special">,</span> <span class="identifier">msg</span><span class="special">);</span>
250 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">wait_all</span><span class="special">(</span><span class="identifier">reqs</span><span class="special">,</span> <span class="identifier">reqs</span> <span class="special">+</span> <span class="number">2</span><span class="special">);</span>
251 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="identifier">msg</span> <span class="special"><<</span> <span class="string">", "</span><span class="special">;</span>
252 <span class="special">}</span>
254 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
255 <span class="special">}</span>
258 We have replaced calls to the <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_4-bb">communicator::send</a></code>
259 and <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_9-bb">communicator::recv</a></code>
260 members with similar calls to their non-blocking counterparts, <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_16-bb">communicator::isend</a></code>
261 and <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_21-bb">communicator::irecv</a></code>.
262 The prefix <span class="bold"><strong>i</strong></span> indicates that the operations
263 return immediately with a <code class="computeroutput"><a class="link" href="../boost/mpi/request.html" title="Class request">mpi::request</a></code>
264 object, which allows one to query the status of a communication request
265 (see the <code class="computeroutput"><a class="link" href="../boost/mpi/request.html#id-1_3_27_7_21_2_1_1_7_2-bb">test</a></code>
266 method) or wait until it has completed (see the <code class="computeroutput"><a class="link" href="../boost/mpi/request.html#id-1_3_27_7_21_2_1_1_7_1-bb">wait</a></code>
267 method). Multiple requests can be completed at the same time with the
268 <code class="computeroutput"><a class="link" href="../boost/mpi/wait_all.html" title="Function wait_all">wait_all</a></code> operation.
270 <div class="important"><table border="0" summary="Important">
272 <td rowspan="2" align="center" valign="top" width="25"><img alt="[Important]" src="../../../doc/src/images/important.png"></td>
273 <th align="left">Important</th>
275 <tr><td align="left" valign="top"><p>
276 Regarding communication completion/progress: The MPI standard requires
277 users to keep the request handle for a non-blocking communication, and
278 to call the "wait" operation (or successfully test for completion)
279 to complete the send or receive. nlike most C MPI implementations, which
280 allow the user to discard the request for a non-blocking send, Boost.MPI
281 requires the user to call "wait" or "test", since
282 the request object might contain temporary buffers that have to be kept
283 until the send is completed. Moreover, the MPI standard does not guarantee
284 that the receive makes any progress before a call to "wait"
285 or "test", although most implementations of the C MPI do allow
286 receives to progress before the call to "wait" or "test".
287 Boost.MPI, on the other hand, generally requires "test" or
288 "wait" calls to make progress. More specifically, Boost.MPI
289 guarantee that calling "test" multiple time will eventually
290 complete the communication (this is due to the fact that serialized communication
291 are potentially a multi step operation.).
295 If you run this program multiple times, you may see some strange results:
296 namely, some runs will produce:
298 <pre class="programlisting"><span class="identifier">Hello</span><span class="special">,</span> <span class="identifier">world</span><span class="special">!</span>
301 while others will produce:
303 <pre class="programlisting"><span class="identifier">world</span><span class="special">!</span>
304 <span class="identifier">Hello</span><span class="special">,</span>
307 or even some garbled version of the letters in "Hello" and "world".
308 This indicates that there is some parallelism in the program, because after
309 both messages are (simultaneously) transmitted, both processes will concurrent
310 execute their print statements. For both performance and correctness, non-blocking
311 communication operations are critical to many parallel applications using
316 <div class="section">
317 <div class="titlepage"><div><div><h3 class="title">
318 <a name="mpi.tutorial.collectives"></a><a class="link" href="tutorial.html#mpi.tutorial.collectives" title="Collective operations">Collective operations</a>
319 </h3></div></div></div>
320 <div class="toc"><dl class="toc">
321 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.collectives.broadcast">Broadcast</a></span></dt>
322 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.collectives.gather">Gather</a></span></dt>
323 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.collectives.scatter">Scatter</a></span></dt>
324 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.collectives.reduce">Reduce</a></span></dt>
327 <a class="link" href="tutorial.html#mpi.tutorial.point_to_point" title="Point-to-Point communication">Point-to-point operations</a>
328 are the core message passing primitives in Boost.MPI. However, many message-passing
329 applications also require higher-level communication algorithms that combine
330 or summarize the data stored on many different processes. These algorithms
331 support many common tasks such as "broadcast this value to all processes",
332 "compute the sum of the values on all processors" or "find
335 <div class="section">
336 <div class="titlepage"><div><div><h4 class="title">
337 <a name="mpi.tutorial.collectives.broadcast"></a><a class="link" href="tutorial.html#mpi.tutorial.collectives.broadcast" title="Broadcast">Broadcast</a>
338 </h4></div></div></div>
340 The <code class="computeroutput"><a class="link" href="../boost/mpi/broadcast.html" title="Function broadcast">broadcast</a></code>
341 algorithm is by far the simplest collective operation. It broadcasts a
342 value from a single process to all other processes within a <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html" title="Class communicator">communicator</a></code>. For instance,
343 the following program broadcasts "Hello, World!" from process
344 0 to every other process. (<code class="computeroutput"><span class="identifier">hello_world_broadcast</span><span class="special">.</span><span class="identifier">cpp</span></code>)
346 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
347 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
348 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">string</span><span class="special">></span>
349 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">serialization</span><span class="special">/</span><span class="identifier">string</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
350 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
352 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
353 <span class="special">{</span>
354 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
355 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
357 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">value</span><span class="special">;</span>
358 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
359 <span class="identifier">value</span> <span class="special">=</span> <span class="string">"Hello, World!"</span><span class="special">;</span>
360 <span class="special">}</span>
362 <span class="identifier">broadcast</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">value</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
364 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"Process #"</span> <span class="special"><<</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special"><<</span> <span class="string">" says "</span> <span class="special"><<</span> <span class="identifier">value</span>
365 <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
366 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
367 <span class="special">}</span>
370 Running this program with seven processes will produce a result such as:
372 <pre class="programlisting">Process #0 says Hello, World!
373 Process #2 says Hello, World!
374 Process #1 says Hello, World!
375 Process #4 says Hello, World!
376 Process #3 says Hello, World!
377 Process #5 says Hello, World!
378 Process #6 says Hello, World!
381 <div class="section">
382 <div class="titlepage"><div><div><h4 class="title">
383 <a name="mpi.tutorial.collectives.gather"></a><a class="link" href="tutorial.html#mpi.tutorial.collectives.gather" title="Gather">Gather</a>
384 </h4></div></div></div>
386 The <code class="computeroutput"><a class="link" href="../boost/mpi/gather.html" title="Function gather">gather</a></code>
387 collective gathers the values produced by every process in a communicator
388 into a vector of values on the "root" process (specified by an
389 argument to <code class="computeroutput"><span class="identifier">gather</span></code>). The
390 /i/th element in the vector will correspond to the value gathered from
391 the /i/th process. For instance, in the following program each process
392 computes its own random number. All of these random numbers are gathered
393 at process 0 (the "root" in this case), which prints out the
394 values that correspond to each processor. (<code class="computeroutput"><span class="identifier">random_gather</span><span class="special">.</span><span class="identifier">cpp</span></code>)
396 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
397 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
398 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">vector</span><span class="special">></span>
399 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">cstdlib</span><span class="special">></span>
400 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
402 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
403 <span class="special">{</span>
404 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
405 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
407 <span class="identifier">std</span><span class="special">::</span><span class="identifier">srand</span><span class="special">(</span><span class="identifier">time</span><span class="special">(</span><span class="number">0</span><span class="special">)</span> <span class="special">+</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">());</span>
408 <span class="keyword">int</span> <span class="identifier">my_number</span> <span class="special">=</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">rand</span><span class="special">();</span>
409 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
410 <span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span> <span class="identifier">all_numbers</span><span class="special">;</span>
411 <span class="identifier">gather</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">my_number</span><span class="special">,</span> <span class="identifier">all_numbers</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
412 <span class="keyword">for</span> <span class="special">(</span><span class="keyword">int</span> <span class="identifier">proc</span> <span class="special">=</span> <span class="number">0</span><span class="special">;</span> <span class="identifier">proc</span> <span class="special"><</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">();</span> <span class="special">++</span><span class="identifier">proc</span><span class="special">)</span>
413 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"Process #"</span> <span class="special"><<</span> <span class="identifier">proc</span> <span class="special"><<</span> <span class="string">" thought of "</span>
414 <span class="special"><<</span> <span class="identifier">all_numbers</span><span class="special">[</span><span class="identifier">proc</span><span class="special">]</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
415 <span class="special">}</span> <span class="keyword">else</span> <span class="special">{</span>
416 <span class="identifier">gather</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">my_number</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
417 <span class="special">}</span>
419 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
420 <span class="special">}</span>
423 Executing this program with seven processes will result in output such
424 as the following. Although the random values will change from one run to
425 the next, the order of the processes in the output will remain the same
426 because only process 0 writes to <code class="computeroutput"><span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span></code>.
428 <pre class="programlisting">Process #0 thought of 332199874
429 Process #1 thought of 20145617
430 Process #2 thought of 1862420122
431 Process #3 thought of 480422940
432 Process #4 thought of 1253380219
433 Process #5 thought of 949458815
434 Process #6 thought of 650073868
437 The <code class="computeroutput"><span class="identifier">gather</span></code> operation collects
438 values from every process into a vector at one process. If instead the
439 values from every process need to be collected into identical vectors on
440 every process, use the <code class="computeroutput"><a class="link" href="../boost/mpi/all_gather.html" title="Function all_gather">all_gather</a></code> algorithm,
441 which is semantically equivalent to calling <code class="computeroutput"><span class="identifier">gather</span></code>
442 followed by a <code class="computeroutput"><span class="identifier">broadcast</span></code>
443 of the resulting vector.
446 <div class="section">
447 <div class="titlepage"><div><div><h4 class="title">
448 <a name="mpi.tutorial.collectives.scatter"></a><a class="link" href="tutorial.html#mpi.tutorial.collectives.scatter" title="Scatter">Scatter</a>
449 </h4></div></div></div>
451 The <code class="computeroutput"><a class="link" href="../boost/mpi/scatter.html" title="Function scatter">scatter</a></code>
452 collective scatters the values from a vector in the "root" process
453 in a communicator into values in all the processes of the communicator.
454 The /i/th element in the vector will correspond to the value received by
455 the /i/th process. For instance, in the following program, the root process
456 produces a vector of random nomber and send one value to each process that
457 will print it. (<code class="computeroutput"><span class="identifier">random_scatter</span><span class="special">.</span><span class="identifier">cpp</span></code>)
459 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
460 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">collectives</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
461 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
462 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">cstdlib</span><span class="special">></span>
463 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">vector</span><span class="special">></span>
465 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
467 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">(</span><span class="keyword">int</span> <span class="identifier">argc</span><span class="special">,</span> <span class="keyword">char</span><span class="special">*</span> <span class="identifier">argv</span><span class="special">[])</span>
468 <span class="special">{</span>
469 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">(</span><span class="identifier">argc</span><span class="special">,</span> <span class="identifier">argv</span><span class="special">);</span>
470 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
472 <span class="identifier">std</span><span class="special">::</span><span class="identifier">srand</span><span class="special">(</span><span class="identifier">time</span><span class="special">(</span><span class="number">0</span><span class="special">)</span> <span class="special">+</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">());</span>
473 <span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span> <span class="identifier">all</span><span class="special">;</span>
474 <span class="keyword">int</span> <span class="identifier">mine</span> <span class="special">=</span> <span class="special">-</span><span class="number">1</span><span class="special">;</span>
475 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
476 <span class="identifier">all</span><span class="special">.</span><span class="identifier">resize</span><span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">());</span>
477 <span class="identifier">std</span><span class="special">::</span><span class="identifier">generate</span><span class="special">(</span><span class="identifier">all</span><span class="special">.</span><span class="identifier">begin</span><span class="special">(),</span> <span class="identifier">all</span><span class="special">.</span><span class="identifier">end</span><span class="special">(),</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">rand</span><span class="special">);</span>
478 <span class="special">}</span>
479 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">scatter</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">all</span><span class="special">,</span> <span class="identifier">mine</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
480 <span class="keyword">for</span> <span class="special">(</span><span class="keyword">int</span> <span class="identifier">r</span> <span class="special">=</span> <span class="number">0</span><span class="special">;</span> <span class="identifier">r</span> <span class="special"><</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">();</span> <span class="special">++</span><span class="identifier">r</span><span class="special">)</span> <span class="special">{</span>
481 <span class="identifier">world</span><span class="special">.</span><span class="identifier">barrier</span><span class="special">();</span>
482 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">r</span> <span class="special">==</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">())</span> <span class="special">{</span>
483 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"Rank "</span> <span class="special"><<</span> <span class="identifier">r</span> <span class="special"><<</span> <span class="string">" got "</span> <span class="special"><<</span> <span class="identifier">mine</span> <span class="special"><<</span> <span class="char">'\n'</span><span class="special">;</span>
484 <span class="special">}</span>
485 <span class="special">}</span>
486 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
487 <span class="special">}</span>
490 Executing this program with seven processes will result in output such
491 as the following. Although the random values will change from one run to
492 the next, the order of the processes in the output will remain the same
493 because of the barrier.
495 <pre class="programlisting">Rank 0 got 1409381269
499 Rank 4 got 1827129182
500 Rank 5 got 1951746047
501 Rank 6 got 2117359639
504 <div class="section">
505 <div class="titlepage"><div><div><h4 class="title">
506 <a name="mpi.tutorial.collectives.reduce"></a><a class="link" href="tutorial.html#mpi.tutorial.collectives.reduce" title="Reduce">Reduce</a>
507 </h4></div></div></div>
509 The <code class="computeroutput"><a class="link" href="../boost/mpi/reduce.html" title="Function reduce">reduce</a></code>
510 collective summarizes the values from each process into a single value
511 at the user-specified "root" process. The Boost.MPI <code class="computeroutput"><span class="identifier">reduce</span></code> operation is similar in spirit
512 to the STL <a href="http://www.sgi.com/tech/stl/accumulate.html" target="_top"><code class="computeroutput"><span class="identifier">accumulate</span></code></a> operation, because
513 it takes a sequence of values (one per process) and combines them via a
514 function object. For instance, we can randomly generate values in each
515 process and the compute the minimum value over all processes via a call
516 to <code class="computeroutput"><a class="link" href="../boost/mpi/reduce.html" title="Function reduce">reduce</a></code>
517 (<code class="computeroutput"><span class="identifier">random_min</span><span class="special">.</span><span class="identifier">cpp</span></code>):
519 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
520 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
521 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">cstdlib</span><span class="special">></span>
522 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
524 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
525 <span class="special">{</span>
526 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
527 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
529 <span class="identifier">std</span><span class="special">::</span><span class="identifier">srand</span><span class="special">(</span><span class="identifier">time</span><span class="special">(</span><span class="number">0</span><span class="special">)</span> <span class="special">+</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">());</span>
530 <span class="keyword">int</span> <span class="identifier">my_number</span> <span class="special">=</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">rand</span><span class="special">();</span>
532 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
533 <span class="keyword">int</span> <span class="identifier">minimum</span><span class="special">;</span>
534 <span class="identifier">reduce</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">my_number</span><span class="special">,</span> <span class="identifier">minimum</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">minimum</span><span class="special"><</span><span class="keyword">int</span><span class="special">>(),</span> <span class="number">0</span><span class="special">);</span>
535 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"The minimum value is "</span> <span class="special"><<</span> <span class="identifier">minimum</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
536 <span class="special">}</span> <span class="keyword">else</span> <span class="special">{</span>
537 <span class="identifier">reduce</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">my_number</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">minimum</span><span class="special"><</span><span class="keyword">int</span><span class="special">>(),</span> <span class="number">0</span><span class="special">);</span>
538 <span class="special">}</span>
540 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
541 <span class="special">}</span>
544 The use of <code class="computeroutput"><span class="identifier">mpi</span><span class="special">::</span><span class="identifier">minimum</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span></code>
545 indicates that the minimum value should be computed. <code class="computeroutput"><span class="identifier">mpi</span><span class="special">::</span><span class="identifier">minimum</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span></code> is a binary function object that compares
546 its two parameters via <code class="computeroutput"><span class="special"><</span></code>
547 and returns the smaller value. Any associative binary function or function
548 object will work provided it's stateless. For instance, to concatenate
549 strings with <code class="computeroutput"><span class="identifier">reduce</span></code> one
550 could use the function object <code class="computeroutput"><span class="identifier">std</span><span class="special">::</span><span class="identifier">plus</span><span class="special"><</span><span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">></span></code> (<code class="computeroutput"><span class="identifier">string_cat</span><span class="special">.</span><span class="identifier">cpp</span></code>):
552 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
553 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
554 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">string</span><span class="special">></span>
555 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">functional</span><span class="special">></span>
556 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">serialization</span><span class="special">/</span><span class="identifier">string</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
557 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
559 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
560 <span class="special">{</span>
561 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
562 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
564 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">names</span><span class="special">[</span><span class="number">10</span><span class="special">]</span> <span class="special">=</span> <span class="special">{</span> <span class="string">"zero "</span><span class="special">,</span> <span class="string">"one "</span><span class="special">,</span> <span class="string">"two "</span><span class="special">,</span> <span class="string">"three "</span><span class="special">,</span>
565 <span class="string">"four "</span><span class="special">,</span> <span class="string">"five "</span><span class="special">,</span> <span class="string">"six "</span><span class="special">,</span> <span class="string">"seven "</span><span class="special">,</span>
566 <span class="string">"eight "</span><span class="special">,</span> <span class="string">"nine "</span> <span class="special">};</span>
568 <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span> <span class="identifier">result</span><span class="special">;</span>
569 <span class="identifier">reduce</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span>
570 <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special"><</span> <span class="number">10</span><span class="special">?</span> <span class="identifier">names</span><span class="special">[</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()]</span>
571 <span class="special">:</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">(</span><span class="string">"many "</span><span class="special">),</span>
572 <span class="identifier">result</span><span class="special">,</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">plus</span><span class="special"><</span><span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">>(),</span> <span class="number">0</span><span class="special">);</span>
574 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span>
575 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"The result is "</span> <span class="special"><<</span> <span class="identifier">result</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
577 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
578 <span class="special">}</span>
581 In this example, we compute a string for each process and then perform
582 a reduction that concatenates all of the strings together into one, long
583 string. Executing this program with seven processors yields the following
586 <pre class="programlisting">The result is zero one two three four five six
589 <a name="mpi.tutorial.collectives.reduce.h0"></a>
590 <span class="phrase"><a name="mpi.tutorial.collectives.reduce.binary_operations_for_reduce"></a></span><a class="link" href="tutorial.html#mpi.tutorial.collectives.reduce.binary_operations_for_reduce">Binary
591 operations for reduce</a>
594 Any kind of binary function objects can be used with <code class="computeroutput"><span class="identifier">reduce</span></code>.
595 For instance, and there are many such function objects in the C++ standard
596 <code class="computeroutput"><span class="special"><</span><span class="identifier">functional</span><span class="special">></span></code> header and the Boost.MPI header <code class="computeroutput"><span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">operations</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span></code>. Or, you can create your own function
597 object. Function objects used with <code class="computeroutput"><span class="identifier">reduce</span></code>
598 must be associative, i.e. <code class="computeroutput"><span class="identifier">f</span><span class="special">(</span><span class="identifier">x</span><span class="special">,</span>
599 <span class="identifier">f</span><span class="special">(</span><span class="identifier">y</span><span class="special">,</span> <span class="identifier">z</span><span class="special">))</span></code> must be equivalent to <code class="computeroutput"><span class="identifier">f</span><span class="special">(</span><span class="identifier">f</span><span class="special">(</span><span class="identifier">x</span><span class="special">,</span> <span class="identifier">y</span><span class="special">),</span> <span class="identifier">z</span><span class="special">)</span></code>. If they are also commutative (i..e,
600 <code class="computeroutput"><span class="identifier">f</span><span class="special">(</span><span class="identifier">x</span><span class="special">,</span> <span class="identifier">y</span><span class="special">)</span> <span class="special">==</span> <span class="identifier">f</span><span class="special">(</span><span class="identifier">y</span><span class="special">,</span>
601 <span class="identifier">x</span><span class="special">)</span></code>),
602 Boost.MPI can use a more efficient implementation of <code class="computeroutput"><span class="identifier">reduce</span></code>.
603 To state that a function object is commutative, you will need to specialize
604 the class <code class="computeroutput"><a class="link" href="../boost/mpi/is_commutative.html" title="Struct template is_commutative">is_commutative</a></code>.
605 For instance, we could modify the previous example by telling Boost.MPI
606 that string concatenation is commutative:
608 <pre class="programlisting"><span class="keyword">namespace</span> <span class="identifier">boost</span> <span class="special">{</span> <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">{</span>
610 <span class="keyword">template</span><span class="special"><></span>
611 <span class="keyword">struct</span> <span class="identifier">is_commutative</span><span class="special"><</span><span class="identifier">std</span><span class="special">::</span><span class="identifier">plus</span><span class="special"><</span><span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">>,</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span><span class="special">></span>
612 <span class="special">:</span> <span class="identifier">mpl</span><span class="special">::</span><span class="identifier">true_</span> <span class="special">{</span> <span class="special">};</span>
614 <span class="special">}</span> <span class="special">}</span> <span class="comment">// end namespace boost::mpi</span>
617 By adding this code prior to <code class="computeroutput"><span class="identifier">main</span><span class="special">()</span></code>, Boost.MPI will assume that string concatenation
618 is commutative and employ a different parallel algorithm for the <code class="computeroutput"><span class="identifier">reduce</span></code> operation. Using this algorithm,
619 the program outputs the following when run with seven processes:
621 <pre class="programlisting">The result is zero one four five six two three
624 Note how the numbers in the resulting string are in a different order:
625 this is a direct result of Boost.MPI reordering operations. The result
626 in this case differed from the non-commutative result because string concatenation
627 is not commutative: <code class="computeroutput"><span class="identifier">f</span><span class="special">(</span><span class="string">"x"</span><span class="special">,</span>
628 <span class="string">"y"</span><span class="special">)</span></code>
629 is not the same as <code class="computeroutput"><span class="identifier">f</span><span class="special">(</span><span class="string">"y"</span><span class="special">,</span>
630 <span class="string">"x"</span><span class="special">)</span></code>,
631 because argument order matters. For truly commutative operations (e.g.,
632 integer addition), the more efficient commutative algorithm will produce
633 the same result as the non-commutative algorithm. Boost.MPI also performs
634 direct mappings from function objects in <code class="computeroutput"><span class="special"><</span><span class="identifier">functional</span><span class="special">></span></code>
635 to <code class="computeroutput"><span class="identifier">MPI_Op</span></code> values predefined
636 by MPI (e.g., <code class="computeroutput"><span class="identifier">MPI_SUM</span></code>,
637 <code class="computeroutput"><span class="identifier">MPI_MAX</span></code>); if you have your
638 own function objects that can take advantage of this mapping, see the class
639 template <code class="computeroutput"><a class="link" href="../boost/mpi/is_mpi_op.html" title="Struct template is_mpi_op">is_mpi_op</a></code>.
641 <div class="warning"><table border="0" summary="Warning">
643 <td rowspan="2" align="center" valign="top" width="25"><img alt="[Warning]" src="../../../doc/src/images/warning.png"></td>
644 <th align="left">Warning</th>
646 <tr><td align="left" valign="top"><p>
647 Due to the underlying MPI limitations, it is important to note that the
648 operation must be stateless.
652 <a name="mpi.tutorial.collectives.reduce.h1"></a>
653 <span class="phrase"><a name="mpi.tutorial.collectives.reduce.all_process_variant"></a></span><a class="link" href="tutorial.html#mpi.tutorial.collectives.reduce.all_process_variant">All process
657 Like <a class="link" href="tutorial.html#mpi.tutorial.collectives.gather" title="Gather"><code class="computeroutput"><span class="identifier">gather</span></code></a>,
658 <code class="computeroutput"><span class="identifier">reduce</span></code> has an "all"
659 variant called <code class="computeroutput"><a class="link" href="../boost/mpi/all_reduce.html" title="Function all_reduce">all_reduce</a></code> that performs
660 the reduction operation and broadcasts the result to all processes. This
661 variant is useful, for instance, in establishing global minimum or maximum
665 The following code (<code class="computeroutput"><span class="identifier">global_min</span><span class="special">.</span><span class="identifier">cpp</span></code>)
666 shows a broadcasting version of the <code class="computeroutput"><span class="identifier">random_min</span><span class="special">.</span><span class="identifier">cpp</span></code>
669 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
670 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
671 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">cstdlib</span><span class="special">></span>
672 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
674 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">(</span><span class="keyword">int</span> <span class="identifier">argc</span><span class="special">,</span> <span class="keyword">char</span><span class="special">*</span> <span class="identifier">argv</span><span class="special">[])</span>
675 <span class="special">{</span>
676 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">(</span><span class="identifier">argc</span><span class="special">,</span> <span class="identifier">argv</span><span class="special">);</span>
677 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
679 <span class="identifier">std</span><span class="special">::</span><span class="identifier">srand</span><span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">());</span>
680 <span class="keyword">int</span> <span class="identifier">my_number</span> <span class="special">=</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">rand</span><span class="special">();</span>
681 <span class="keyword">int</span> <span class="identifier">minimum</span><span class="special">;</span>
683 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">all_reduce</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">my_number</span><span class="special">,</span> <span class="identifier">minimum</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">minimum</span><span class="special"><</span><span class="keyword">int</span><span class="special">>());</span>
685 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
686 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"The minimum value is "</span> <span class="special"><<</span> <span class="identifier">minimum</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
687 <span class="special">}</span>
689 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
690 <span class="special">}</span>
693 In that example we provide both input and output values, requiring twice
694 as much space, which can be a problem depending on the size of the transmitted
695 data. If there is no need to preserve the input value, the output value
696 can be omitted. In that case the input value will be overridden with the
697 output value and Boost.MPI is able, in some situation, to implement the
698 operation with a more space efficient solution (using the <code class="computeroutput"><span class="identifier">MPI_IN_PLACE</span></code> flag of the MPI C mapping),
699 as in the following example (<code class="computeroutput"><span class="identifier">in_place_global_min</span><span class="special">.</span><span class="identifier">cpp</span></code>):
701 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
702 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
703 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">cstdlib</span><span class="special">></span>
704 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
706 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">(</span><span class="keyword">int</span> <span class="identifier">argc</span><span class="special">,</span> <span class="keyword">char</span><span class="special">*</span> <span class="identifier">argv</span><span class="special">[])</span>
707 <span class="special">{</span>
708 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">(</span><span class="identifier">argc</span><span class="special">,</span> <span class="identifier">argv</span><span class="special">);</span>
709 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
711 <span class="identifier">std</span><span class="special">::</span><span class="identifier">srand</span><span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">());</span>
712 <span class="keyword">int</span> <span class="identifier">my_number</span> <span class="special">=</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">rand</span><span class="special">();</span>
714 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">all_reduce</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">my_number</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">minimum</span><span class="special"><</span><span class="keyword">int</span><span class="special">>());</span>
716 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special">==</span> <span class="number">0</span><span class="special">)</span> <span class="special">{</span>
717 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"The minimum value is "</span> <span class="special"><<</span> <span class="identifier">my_number</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
718 <span class="special">}</span>
720 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
721 <span class="special">}</span>
725 <div class="section">
726 <div class="titlepage"><div><div><h3 class="title">
727 <a name="mpi.tutorial.user_data_types"></a><a class="link" href="tutorial.html#mpi.tutorial.user_data_types" title="User-defined data types">User-defined data types</a>
728 </h3></div></div></div>
730 The inclusion of <code class="computeroutput"><span class="identifier">boost</span><span class="special">/</span><span class="identifier">serialization</span><span class="special">/</span><span class="identifier">string</span><span class="special">.</span><span class="identifier">hpp</span></code> in the previous examples is very important:
731 it makes values of type <code class="computeroutput"><span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span></code>
732 serializable, so that they can be be transmitted using Boost.MPI. In general,
733 built-in C++ types (<code class="computeroutput"><span class="keyword">int</span></code>s, <code class="computeroutput"><span class="keyword">float</span></code>s, characters, etc.) can be transmitted
734 over MPI directly, while user-defined and library-defined types will need
735 to first be serialized (packed) into a format that is amenable to transmission.
736 Boost.MPI relies on the <a href="../../../libs/serialization/doc" target="_top">Boost.Serialization</a>
737 library to serialize and deserialize data types.
740 For types defined by the standard library (such as <code class="computeroutput"><span class="identifier">std</span><span class="special">::</span><span class="identifier">string</span></code>
741 or <code class="computeroutput"><span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span></code>) and some types in Boost (such as
742 <code class="computeroutput"><span class="identifier">boost</span><span class="special">::</span><span class="identifier">variant</span></code>), the <a href="../../../libs/serialization/doc" target="_top">Boost.Serialization</a>
743 library already contains all of the required serialization code. In these
744 cases, you need only include the appropriate header from the <code class="computeroutput"><span class="identifier">boost</span><span class="special">/</span><span class="identifier">serialization</span></code> directory.
747 For types that do not already have a serialization header, you will first
748 need to implement serialization code before the types can be transmitted
749 using Boost.MPI. Consider a simple class <a class="link" href="tutorial.html#gps_position"><code class="computeroutput"><span class="identifier">gps_position</span></code></a> that contains members
750 <code class="computeroutput"><span class="identifier">degrees</span></code>, <code class="computeroutput"><span class="identifier">minutes</span></code>,
751 and <code class="computeroutput"><span class="identifier">seconds</span></code>. This class is
752 made serializable by making it a friend of <code class="computeroutput"><span class="identifier">boost</span><span class="special">::</span><span class="identifier">serialization</span><span class="special">::</span><span class="identifier">access</span></code>
753 and introducing the templated <code class="computeroutput"><span class="identifier">serialize</span><span class="special">()</span></code> function, as follows:<a name="gps_position"></a>
755 <pre class="programlisting"><span class="keyword">class</span> <span class="identifier">gps_position</span>
756 <span class="special">{</span>
757 <span class="keyword">private</span><span class="special">:</span>
758 <span class="keyword">friend</span> <span class="keyword">class</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">serialization</span><span class="special">::</span><span class="identifier">access</span><span class="special">;</span>
760 <span class="keyword">template</span><span class="special"><</span><span class="keyword">class</span> <span class="identifier">Archive</span><span class="special">></span>
761 <span class="keyword">void</span> <span class="identifier">serialize</span><span class="special">(</span><span class="identifier">Archive</span> <span class="special">&</span> <span class="identifier">ar</span><span class="special">,</span> <span class="keyword">const</span> <span class="keyword">unsigned</span> <span class="keyword">int</span> <span class="identifier">version</span><span class="special">)</span>
762 <span class="special">{</span>
763 <span class="identifier">ar</span> <span class="special">&</span> <span class="identifier">degrees</span><span class="special">;</span>
764 <span class="identifier">ar</span> <span class="special">&</span> <span class="identifier">minutes</span><span class="special">;</span>
765 <span class="identifier">ar</span> <span class="special">&</span> <span class="identifier">seconds</span><span class="special">;</span>
766 <span class="special">}</span>
768 <span class="keyword">int</span> <span class="identifier">degrees</span><span class="special">;</span>
769 <span class="keyword">int</span> <span class="identifier">minutes</span><span class="special">;</span>
770 <span class="keyword">float</span> <span class="identifier">seconds</span><span class="special">;</span>
771 <span class="keyword">public</span><span class="special">:</span>
772 <span class="identifier">gps_position</span><span class="special">(){};</span>
773 <span class="identifier">gps_position</span><span class="special">(</span><span class="keyword">int</span> <span class="identifier">d</span><span class="special">,</span> <span class="keyword">int</span> <span class="identifier">m</span><span class="special">,</span> <span class="keyword">float</span> <span class="identifier">s</span><span class="special">)</span> <span class="special">:</span>
774 <span class="identifier">degrees</span><span class="special">(</span><span class="identifier">d</span><span class="special">),</span> <span class="identifier">minutes</span><span class="special">(</span><span class="identifier">m</span><span class="special">),</span> <span class="identifier">seconds</span><span class="special">(</span><span class="identifier">s</span><span class="special">)</span>
775 <span class="special">{}</span>
776 <span class="special">};</span>
779 Complete information about making types serializable is beyond the scope
780 of this tutorial. For more information, please see the <a href="../../../libs/serialization/doc" target="_top">Boost.Serialization</a>
781 library tutorial from which the above example was extracted. One important
782 side benefit of making types serializable for Boost.MPI is that they become
783 serializable for any other usage, such as storing the objects to disk and
784 manipulated them in XML.
787 Some serializable types, like <a class="link" href="tutorial.html#gps_position"><code class="computeroutput"><span class="identifier">gps_position</span></code></a> above, have a fixed
788 amount of data stored at fixed offsets and are fully defined by the values
789 of their data member (most POD with no pointers are a good example). When
790 this is the case, Boost.MPI can optimize their serialization and transmission
791 by avoiding extraneous copy operations. To enable this optimization, users
792 must specialize the type trait <code class="computeroutput"><a class="link" href="../boost/mpi/is_mpi_datatype.html" title="Struct template is_mpi_datatype">is_mpi_datatype</a></code>, e.g.:
794 <pre class="programlisting"><span class="keyword">namespace</span> <span class="identifier">boost</span> <span class="special">{</span> <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">{</span>
795 <span class="keyword">template</span> <span class="special"><></span>
796 <span class="keyword">struct</span> <span class="identifier">is_mpi_datatype</span><span class="special"><</span><span class="identifier">gps_position</span><span class="special">></span> <span class="special">:</span> <span class="identifier">mpl</span><span class="special">::</span><span class="identifier">true_</span> <span class="special">{</span> <span class="special">};</span>
797 <span class="special">}</span> <span class="special">}</span>
800 For non-template types we have defined a macro to simplify declaring a type
803 <pre class="programlisting"><span class="identifier">BOOST_IS_MPI_DATATYPE</span><span class="special">(</span><span class="identifier">gps_position</span><span class="special">)</span>
806 For composite traits, the specialization of <code class="computeroutput"><a class="link" href="../boost/mpi/is_mpi_datatype.html" title="Struct template is_mpi_datatype">is_mpi_datatype</a></code> may depend
807 on <code class="computeroutput"><span class="identifier">is_mpi_datatype</span></code> itself.
808 For instance, a <code class="computeroutput"><span class="identifier">boost</span><span class="special">::</span><span class="identifier">array</span></code> object is fixed only when the type
809 of the parameter it stores is fixed:
811 <pre class="programlisting"><span class="keyword">namespace</span> <span class="identifier">boost</span> <span class="special">{</span> <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">{</span>
812 <span class="keyword">template</span> <span class="special"><</span><span class="keyword">typename</span> <span class="identifier">T</span><span class="special">,</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">size_t</span> <span class="identifier">N</span><span class="special">></span>
813 <span class="keyword">struct</span> <span class="identifier">is_mpi_datatype</span><span class="special"><</span><span class="identifier">array</span><span class="special"><</span><span class="identifier">T</span><span class="special">,</span> <span class="identifier">N</span><span class="special">></span> <span class="special">></span>
814 <span class="special">:</span> <span class="keyword">public</span> <span class="identifier">is_mpi_datatype</span><span class="special"><</span><span class="identifier">T</span><span class="special">></span> <span class="special">{</span> <span class="special">};</span>
815 <span class="special">}</span> <span class="special">}</span>
818 The redundant copy elimination optimization can only be applied when the
819 shape of the data type is completely fixed. Variable-length types (e.g.,
820 strings, linked lists) and types that store pointers cannot use the optimization,
821 but Boost.MPI will be unable to detect this error at compile time. Attempting
822 to perform this optimization when it is not correct will likely result in
823 segmentation faults and other strange program behavior.
826 Boost.MPI can transmit any user-defined data type from one process to another.
827 Built-in types can be transmitted without any extra effort; library-defined
828 types require the inclusion of a serialization header; and user-defined types
829 will require the addition of serialization code. Fixed data types can be
830 optimized for transmission using the <code class="computeroutput"><a class="link" href="../boost/mpi/is_mpi_datatype.html" title="Struct template is_mpi_datatype">is_mpi_datatype</a></code> type trait.
833 <div class="section">
834 <div class="titlepage"><div><div><h3 class="title">
835 <a name="mpi.tutorial.communicators"></a><a class="link" href="tutorial.html#mpi.tutorial.communicators" title="Communicators">Communicators</a>
836 </h3></div></div></div>
837 <div class="toc"><dl class="toc">
838 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.communicators.managing">Managing communicators</a></span></dt>
839 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.communicators.cartesian_communicator">Cartesian
840 communicator</a></span></dt>
842 <div class="section">
843 <div class="titlepage"><div><div><h4 class="title">
844 <a name="mpi.tutorial.communicators.managing"></a><a class="link" href="tutorial.html#mpi.tutorial.communicators.managing" title="Managing communicators">Managing communicators</a>
845 </h4></div></div></div>
847 Communication with Boost.MPI always occurs over a communicator. A communicator
848 contains a set of processes that can send messages among themselves and
849 perform collective operations. There can be many communicators within a
850 single program, each of which contains its own isolated communication space
851 that acts independently of the other communicators.
854 When the MPI environment is initialized, only the "world" communicator
855 (called <code class="computeroutput"><span class="identifier">MPI_COMM_WORLD</span></code>
856 in the MPI C and Fortran bindings) is available. The "world"
857 communicator, accessed by default-constructing a <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html" title="Class communicator">mpi::communicator</a></code>
858 object, contains all of the MPI processes present when the program begins
859 execution. Other communicators can then be constructed by duplicating or
860 building subsets of the "world" communicator. For instance, in
861 the following program we split the processes into two groups: one for processes
862 generating data and the other for processes that will collect the data.
863 (<code class="computeroutput"><span class="identifier">generate_collect</span><span class="special">.</span><span class="identifier">cpp</span></code>)
865 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
866 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
867 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">cstdlib</span><span class="special">></span>
868 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">serialization</span><span class="special">/</span><span class="identifier">vector</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
869 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
871 <span class="keyword">enum</span> <span class="identifier">message_tags</span> <span class="special">{</span><span class="identifier">msg_data_packet</span><span class="special">,</span> <span class="identifier">msg_broadcast_data</span><span class="special">,</span> <span class="identifier">msg_finished</span><span class="special">};</span>
873 <span class="keyword">void</span> <span class="identifier">generate_data</span><span class="special">(</span><span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">local</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">);</span>
874 <span class="keyword">void</span> <span class="identifier">collect_data</span><span class="special">(</span><span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">local</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">);</span>
876 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
877 <span class="special">{</span>
878 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
879 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
881 <span class="keyword">bool</span> <span class="identifier">is_generator</span> <span class="special">=</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special"><</span> <span class="number">2</span> <span class="special">*</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">()</span> <span class="special">/</span> <span class="number">3</span><span class="special">;</span>
882 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">local</span> <span class="special">=</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">split</span><span class="special">(</span><span class="identifier">is_generator</span><span class="special">?</span> <span class="number">0</span> <span class="special">:</span> <span class="number">1</span><span class="special">);</span>
883 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">is_generator</span><span class="special">)</span> <span class="identifier">generate_data</span><span class="special">(</span><span class="identifier">local</span><span class="special">,</span> <span class="identifier">world</span><span class="special">);</span>
884 <span class="keyword">else</span> <span class="identifier">collect_data</span><span class="special">(</span><span class="identifier">local</span><span class="special">,</span> <span class="identifier">world</span><span class="special">);</span>
886 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
887 <span class="special">}</span>
890 When communicators are split in this way, their processes retain membership
891 in both the original communicator (which is not altered by the split) and
892 the new communicator. However, the ranks of the processes may be different
893 from one communicator to the next, because the rank values within a communicator
894 are always contiguous values starting at zero. In the example above, the
895 first two thirds of the processes become "generators" and the
896 remaining processes become "collectors". The ranks of the "collectors"
897 in the <code class="computeroutput"><span class="identifier">world</span></code> communicator
898 will be 2/3 <code class="computeroutput"><span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">()</span></code>
899 and greater, whereas the ranks of the same collector processes in the
900 <code class="computeroutput"><span class="identifier">local</span></code> communicator will
901 start at zero. The following excerpt from <code class="computeroutput"><span class="identifier">collect_data</span><span class="special">()</span></code> (in <code class="computeroutput"><span class="identifier">generate_collect</span><span class="special">.</span><span class="identifier">cpp</span></code>)
902 illustrates how to manage multiple communicators:
904 <pre class="programlisting"><span class="identifier">mpi</span><span class="special">::</span><span class="identifier">status</span> <span class="identifier">msg</span> <span class="special">=</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">probe</span><span class="special">();</span>
905 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">msg</span><span class="special">.</span><span class="identifier">tag</span><span class="special">()</span> <span class="special">==</span> <span class="identifier">msg_data_packet</span><span class="special">)</span> <span class="special">{</span>
906 <span class="comment">// Receive the packet of data</span>
907 <span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span> <span class="identifier">data</span><span class="special">;</span>
908 <span class="identifier">world</span><span class="special">.</span><span class="identifier">recv</span><span class="special">(</span><span class="identifier">msg</span><span class="special">.</span><span class="identifier">source</span><span class="special">(),</span> <span class="identifier">msg</span><span class="special">.</span><span class="identifier">tag</span><span class="special">(),</span> <span class="identifier">data</span><span class="special">);</span>
910 <span class="comment">// Tell each of the collectors that we'll be broadcasting some data</span>
911 <span class="keyword">for</span> <span class="special">(</span><span class="keyword">int</span> <span class="identifier">dest</span> <span class="special">=</span> <span class="number">1</span><span class="special">;</span> <span class="identifier">dest</span> <span class="special"><</span> <span class="identifier">local</span><span class="special">.</span><span class="identifier">size</span><span class="special">();</span> <span class="special">++</span><span class="identifier">dest</span><span class="special">)</span>
912 <span class="identifier">local</span><span class="special">.</span><span class="identifier">send</span><span class="special">(</span><span class="identifier">dest</span><span class="special">,</span> <span class="identifier">msg_broadcast_data</span><span class="special">,</span> <span class="identifier">msg</span><span class="special">.</span><span class="identifier">source</span><span class="special">());</span>
914 <span class="comment">// Broadcast the actual data.</span>
915 <span class="identifier">broadcast</span><span class="special">(</span><span class="identifier">local</span><span class="special">,</span> <span class="identifier">data</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
916 <span class="special">}</span>
919 The code in this except is executed by the "master" collector,
920 e.g., the node with rank 2/3 <code class="computeroutput"><span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">()</span></code> in the <code class="computeroutput"><span class="identifier">world</span></code>
921 communicator and rank 0 in the <code class="computeroutput"><span class="identifier">local</span></code>
922 (collector) communicator. It receives a message from a generator via the
923 <code class="computeroutput"><span class="identifier">world</span></code> communicator, then
924 broadcasts the message to each of the collectors via the <code class="computeroutput"><span class="identifier">local</span></code> communicator.
927 For more control in the creation of communicators for subgroups of processes,
928 the Boost.MPI <code class="computeroutput"><a class="link" href="../boost/mpi/group.html" title="Class group">group</a></code>
929 provides facilities to compute the union (<code class="computeroutput"><span class="special">|</span></code>),
930 intersection (<code class="computeroutput"><span class="special">&</span></code>), and
931 difference (<code class="computeroutput"><span class="special">-</span></code>) of two groups,
932 generate arbitrary subgroups, etc.
935 <div class="section">
936 <div class="titlepage"><div><div><h4 class="title">
937 <a name="mpi.tutorial.communicators.cartesian_communicator"></a><a class="link" href="tutorial.html#mpi.tutorial.communicators.cartesian_communicator" title="Cartesian communicator">Cartesian
939 </h4></div></div></div>
941 A communicator can be organised as a cartesian grid, here a basic example:
943 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">vector</span><span class="special">></span>
944 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
946 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">communicator</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
947 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">collectives</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
948 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">environment</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
949 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">cartesian_communicator</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
951 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">test</span><span class="special">/</span><span class="identifier">minimal</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
953 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
954 <span class="keyword">int</span> <span class="identifier">test_main</span><span class="special">(</span><span class="keyword">int</span> <span class="identifier">argc</span><span class="special">,</span> <span class="keyword">char</span><span class="special">*</span> <span class="identifier">argv</span><span class="special">[])</span>
955 <span class="special">{</span>
956 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">;</span>
957 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
959 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">()</span> <span class="special">!=</span> <span class="number">24</span><span class="special">)</span> <span class="keyword">return</span> <span class="special">-</span><span class="number">1</span><span class="special">;</span>
960 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">cartesian_dimension</span> <span class="identifier">dims</span><span class="special">[]</span> <span class="special">=</span> <span class="special">{{</span><span class="number">2</span><span class="special">,</span> <span class="keyword">true</span><span class="special">},</span> <span class="special">{</span><span class="number">3</span><span class="special">,</span><span class="keyword">true</span><span class="special">},</span> <span class="special">{</span><span class="number">4</span><span class="special">,</span><span class="keyword">true</span><span class="special">}};</span>
961 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">cartesian_communicator</span> <span class="identifier">cart</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">cartesian_topology</span><span class="special">(</span><span class="identifier">dims</span><span class="special">));</span>
962 <span class="keyword">for</span> <span class="special">(</span><span class="keyword">int</span> <span class="identifier">r</span> <span class="special">=</span> <span class="number">0</span><span class="special">;</span> <span class="identifier">r</span> <span class="special"><</span> <span class="identifier">cart</span><span class="special">.</span><span class="identifier">size</span><span class="special">();</span> <span class="special">++</span><span class="identifier">r</span><span class="special">)</span> <span class="special">{</span>
963 <span class="identifier">cart</span><span class="special">.</span><span class="identifier">barrier</span><span class="special">();</span>
964 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">r</span> <span class="special">==</span> <span class="identifier">cart</span><span class="special">.</span><span class="identifier">rank</span><span class="special">())</span> <span class="special">{</span>
965 <span class="identifier">std</span><span class="special">::</span><span class="identifier">vector</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span> <span class="identifier">c</span> <span class="special">=</span> <span class="identifier">cart</span><span class="special">.</span><span class="identifier">coordinates</span><span class="special">(</span><span class="identifier">r</span><span class="special">);</span>
966 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"rk :"</span> <span class="special"><<</span> <span class="identifier">r</span> <span class="special"><<</span> <span class="string">" coords: "</span>
967 <span class="special"><<</span> <span class="identifier">c</span><span class="special">[</span><span class="number">0</span><span class="special">]</span> <span class="special"><<</span> <span class="char">' '</span> <span class="special"><<</span> <span class="identifier">c</span><span class="special">[</span><span class="number">1</span><span class="special">]</span> <span class="special"><<</span> <span class="char">' '</span> <span class="special"><<</span> <span class="identifier">c</span><span class="special">[</span><span class="number">2</span><span class="special">]</span> <span class="special"><<</span> <span class="char">'\n'</span><span class="special">;</span>
968 <span class="special">}</span>
969 <span class="special">}</span>
970 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
971 <span class="special">}</span>
975 <div class="section">
976 <div class="titlepage"><div><div><h3 class="title">
977 <a name="mpi.tutorial.threading"></a><a class="link" href="tutorial.html#mpi.tutorial.threading" title="Threads">Threads</a>
978 </h3></div></div></div>
980 There are an increasing number of hybrid parallel applications that mix distributed
981 and shared memory parallelism. To know how to support that model, one need
982 to know what level of threading support is guaranteed by the MPI implementation.
983 There are 4 ordered level of possible threading support described by <code class="computeroutput"><a class="link" href="../boost/mpi/threading/level.html" title="Type level">mpi::threading::level</a></code>. At the
984 lowest level, you should not use threads at all, at the highest level, any
985 thread can perform MPI call.
988 If you want to use multi-threading in your MPI application, you should indicate
989 in the environment constructor your preferred threading support. Then probe
990 the one the library did provide, and decide what you can do with it (it could
991 be nothing, then aborting is a valid option):
993 <pre class="programlisting"><span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">environment</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
994 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">communicator</span><span class="special">.</span><span class="identifier">hpp</span><span class="special">></span>
995 <span class="preprocessor">#include</span> <span class="special"><</span><span class="identifier">iostream</span><span class="special">></span>
996 <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">=</span> <span class="identifier">boost</span><span class="special">::</span><span class="identifier">mpi</span><span class="special">;</span>
997 <span class="keyword">namespace</span> <span class="identifier">mt</span> <span class="special">=</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">threading</span><span class="special">;</span>
999 <span class="keyword">int</span> <span class="identifier">main</span><span class="special">()</span>
1000 <span class="special">{</span>
1001 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">environment</span> <span class="identifier">env</span><span class="special">(</span><span class="identifier">mt</span><span class="special">::</span><span class="identifier">funneled</span><span class="special">);</span>
1002 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">env</span><span class="special">.</span><span class="identifier">thread_level</span><span class="special">()</span> <span class="special"><</span> <span class="identifier">mt</span><span class="special">::</span><span class="identifier">funneled</span><span class="special">)</span> <span class="special">{</span>
1003 <span class="identifier">env</span><span class="special">.</span><span class="identifier">abort</span><span class="special">(-</span><span class="number">1</span><span class="special">);</span>
1004 <span class="special">}</span>
1005 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">communicator</span> <span class="identifier">world</span><span class="special">;</span>
1006 <span class="identifier">std</span><span class="special">::</span><span class="identifier">cout</span> <span class="special"><<</span> <span class="string">"I am process "</span> <span class="special"><<</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">rank</span><span class="special">()</span> <span class="special"><<</span> <span class="string">" of "</span> <span class="special"><<</span> <span class="identifier">world</span><span class="special">.</span><span class="identifier">size</span><span class="special">()</span>
1007 <span class="special"><<</span> <span class="string">"."</span> <span class="special"><<</span> <span class="identifier">std</span><span class="special">::</span><span class="identifier">endl</span><span class="special">;</span>
1008 <span class="keyword">return</span> <span class="number">0</span><span class="special">;</span>
1009 <span class="special">}</span>
1012 <div class="section">
1013 <div class="titlepage"><div><div><h3 class="title">
1014 <a name="mpi.tutorial.skeleton_and_content"></a><a class="link" href="tutorial.html#mpi.tutorial.skeleton_and_content" title="Separating structure from content">Separating structure
1016 </h3></div></div></div>
1018 When communicating data types over MPI that are not fundamental to MPI (such
1019 as strings, lists, and user-defined data types), Boost.MPI must first serialize
1020 these data types into a buffer and then communicate them; the receiver then
1021 copies the results into a buffer before deserializing into an object on the
1022 other end. For some data types, this overhead can be eliminated by using
1023 <code class="computeroutput"><a class="link" href="../boost/mpi/is_mpi_datatype.html" title="Struct template is_mpi_datatype">is_mpi_datatype</a></code>.
1024 However, variable-length data types such as strings and lists cannot be MPI
1028 Boost.MPI supports a second technique for improving performance by separating
1029 the structure of these variable-length data structures from the content stored
1030 in the data structures. This feature is only beneficial when the shape of
1031 the data structure remains the same but the content of the data structure
1032 will need to be communicated several times. For instance, in a finite element
1033 analysis the structure of the mesh may be fixed at the beginning of computation
1034 but the various variables on the cells of the mesh (temperature, stress,
1035 etc.) will be communicated many times within the iterative analysis process.
1036 In this case, Boost.MPI allows one to first send the "skeleton"
1037 of the mesh once, then transmit the "content" multiple times. Since
1038 the content need not contain any information about the structure of the data
1039 type, it can be transmitted without creating separate communication buffers.
1042 To illustrate the use of skeletons and content, we will take a somewhat more
1043 limited example wherein a master process generates random number sequences
1044 into a list and transmits them to several slave processes. The length of
1045 the list will be fixed at program startup, so the content of the list (i.e.,
1046 the current sequence of numbers) can be transmitted efficiently. The complete
1047 example is available in <code class="computeroutput"><span class="identifier">example</span><span class="special">/</span><span class="identifier">random_content</span><span class="special">.</span><span class="identifier">cpp</span></code>. We
1048 being with the master process (rank 0), which builds a list, communicates
1049 its structure via a <code class="computeroutput"><a class="link" href="reference.html#boost.mpi.skeleton">skeleton</a></code>, then repeatedly
1050 generates random number sequences to be broadcast to the slave processes
1051 via <code class="computeroutput">content</code>:
1053 <pre class="programlisting"><span class="comment">// Generate the list and broadcast its structure</span>
1054 <span class="identifier">std</span><span class="special">::</span><span class="identifier">list</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span> <span class="identifier">l</span><span class="special">(</span><span class="identifier">list_len</span><span class="special">);</span>
1055 <span class="identifier">broadcast</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">skeleton</span><span class="special">(</span><span class="identifier">l</span><span class="special">),</span> <span class="number">0</span><span class="special">);</span>
1057 <span class="comment">// Generate content several times and broadcast out that content</span>
1058 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">content</span> <span class="identifier">c</span> <span class="special">=</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">get_content</span><span class="special">(</span><span class="identifier">l</span><span class="special">);</span>
1059 <span class="keyword">for</span> <span class="special">(</span><span class="keyword">int</span> <span class="identifier">i</span> <span class="special">=</span> <span class="number">0</span><span class="special">;</span> <span class="identifier">i</span> <span class="special"><</span> <span class="identifier">iterations</span><span class="special">;</span> <span class="special">++</span><span class="identifier">i</span><span class="special">)</span> <span class="special">{</span>
1060 <span class="comment">// Generate new random values</span>
1061 <span class="identifier">std</span><span class="special">::</span><span class="identifier">generate</span><span class="special">(</span><span class="identifier">l</span><span class="special">.</span><span class="identifier">begin</span><span class="special">(),</span> <span class="identifier">l</span><span class="special">.</span><span class="identifier">end</span><span class="special">(),</span> <span class="special">&</span><span class="identifier">random</span><span class="special">);</span>
1063 <span class="comment">// Broadcast the new content of l</span>
1064 <span class="identifier">broadcast</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">c</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
1065 <span class="special">}</span>
1067 <span class="comment">// Notify the slaves that we're done by sending all zeroes</span>
1068 <span class="identifier">std</span><span class="special">::</span><span class="identifier">fill</span><span class="special">(</span><span class="identifier">l</span><span class="special">.</span><span class="identifier">begin</span><span class="special">(),</span> <span class="identifier">l</span><span class="special">.</span><span class="identifier">end</span><span class="special">(),</span> <span class="number">0</span><span class="special">);</span>
1069 <span class="identifier">broadcast</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">c</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
1072 The slave processes have a very similar structure to the master. They receive
1073 (via the <code class="computeroutput"><a class="link" href="../boost/mpi/broadcast.html" title="Function broadcast">broadcast()</a></code> call) the skeleton of the
1074 data structure, then use it to build their own lists of integers. In each
1075 iteration, they receive via another <code class="computeroutput"><span class="identifier">broadcast</span><span class="special">()</span></code> the new content in the data structure and
1076 compute some property of the data:
1078 <pre class="programlisting"><span class="comment">// Receive the content and build up our own list</span>
1079 <span class="identifier">std</span><span class="special">::</span><span class="identifier">list</span><span class="special"><</span><span class="keyword">int</span><span class="special">></span> <span class="identifier">l</span><span class="special">;</span>
1080 <span class="identifier">broadcast</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">skeleton</span><span class="special">(</span><span class="identifier">l</span><span class="special">),</span> <span class="number">0</span><span class="special">);</span>
1082 <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">content</span> <span class="identifier">c</span> <span class="special">=</span> <span class="identifier">mpi</span><span class="special">::</span><span class="identifier">get_content</span><span class="special">(</span><span class="identifier">l</span><span class="special">);</span>
1083 <span class="keyword">int</span> <span class="identifier">i</span> <span class="special">=</span> <span class="number">0</span><span class="special">;</span>
1084 <span class="keyword">do</span> <span class="special">{</span>
1085 <span class="identifier">broadcast</span><span class="special">(</span><span class="identifier">world</span><span class="special">,</span> <span class="identifier">c</span><span class="special">,</span> <span class="number">0</span><span class="special">);</span>
1087 <span class="keyword">if</span> <span class="special">(</span><span class="identifier">std</span><span class="special">::</span><span class="identifier">find_if</span>
1088 <span class="special">(</span><span class="identifier">l</span><span class="special">.</span><span class="identifier">begin</span><span class="special">(),</span> <span class="identifier">l</span><span class="special">.</span><span class="identifier">end</span><span class="special">(),</span>
1089 <span class="identifier">std</span><span class="special">::</span><span class="identifier">bind1st</span><span class="special">(</span><span class="identifier">std</span><span class="special">::</span><span class="identifier">not_equal_to</span><span class="special"><</span><span class="keyword">int</span><span class="special">>(),</span> <span class="number">0</span><span class="special">))</span> <span class="special">==</span> <span class="identifier">l</span><span class="special">.</span><span class="identifier">end</span><span class="special">())</span>
1090 <span class="keyword">break</span><span class="special">;</span>
1092 <span class="comment">// Compute some property of the data.</span>
1094 <span class="special">++</span><span class="identifier">i</span><span class="special">;</span>
1095 <span class="special">}</span> <span class="keyword">while</span> <span class="special">(</span><span class="keyword">true</span><span class="special">);</span>
1098 The skeletons and content of any Serializable data type can be transmitted
1099 either via the <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_4-bb">send</a></code> and <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html#id-1_3_27_7_6_2_1_1_3_9-bb">recv</a></code> members of the <code class="computeroutput"><a class="link" href="../boost/mpi/communicator.html" title="Class communicator">communicator</a></code>
1100 class (for point-to-point communicators) or broadcast via the <code class="computeroutput"><a class="link" href="../boost/mpi/broadcast.html" title="Function broadcast">broadcast()</a></code> collective. When separating
1101 a data structure into a skeleton and content, be careful not to modify the
1102 data structure (either on the sender side or the receiver side) without transmitting
1103 the skeleton again. Boost.MPI can not detect these accidental modifications
1104 to the data structure, which will likely result in incorrect data being transmitted
1105 or unstable programs.
1108 <div class="section">
1109 <div class="titlepage"><div><div><h3 class="title">
1110 <a name="mpi.tutorial.performance_optimizations"></a><a class="link" href="tutorial.html#mpi.tutorial.performance_optimizations" title="Performance optimizations">Performance optimizations</a>
1111 </h3></div></div></div>
1112 <div class="toc"><dl class="toc">
1113 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.performance_optimizations.serialization_optimizations">Serialization
1114 optimizations</a></span></dt>
1115 <dt><span class="section"><a href="tutorial.html#mpi.tutorial.performance_optimizations.homogeneous_machines">Homogeneous
1116 Machines</a></span></dt>
1118 <div class="section">
1119 <div class="titlepage"><div><div><h4 class="title">
1120 <a name="mpi.tutorial.performance_optimizations.serialization_optimizations"></a><a class="link" href="tutorial.html#mpi.tutorial.performance_optimizations.serialization_optimizations" title="Serialization optimizations">Serialization
1122 </h4></div></div></div>
1124 To obtain optimal performance for small fixed-length data types not containing
1125 any pointers it is very important to mark them using the type traits of
1126 Boost.MPI and Boost.Serialization.
1129 It was already discussed that fixed length types containing no pointers
1130 can be using as <code class="computeroutput"><a class="link" href="../boost/mpi/is_mpi_datatype.html" title="Struct template is_mpi_datatype">is_mpi_datatype</a></code>, e.g.:
1132 <pre class="programlisting"><span class="keyword">namespace</span> <span class="identifier">boost</span> <span class="special">{</span> <span class="keyword">namespace</span> <span class="identifier">mpi</span> <span class="special">{</span>
1133 <span class="keyword">template</span> <span class="special"><></span>
1134 <span class="keyword">struct</span> <span class="identifier">is_mpi_datatype</span><span class="special"><</span><span class="identifier">gps_position</span><span class="special">></span> <span class="special">:</span> <span class="identifier">mpl</span><span class="special">::</span><span class="identifier">true_</span> <span class="special">{</span> <span class="special">};</span>
1135 <span class="special">}</span> <span class="special">}</span>
1138 or the equivalent macro
1140 <pre class="programlisting"><span class="identifier">BOOST_IS_MPI_DATATYPE</span><span class="special">(</span><span class="identifier">gps_position</span><span class="special">)</span>
1143 In addition it can give a substantial performance gain to turn off tracking
1144 and versioning for these types, if no pointers to these types are used,
1145 by using the traits classes or helper macros of Boost.Serialization:
1147 <pre class="programlisting"><span class="identifier">BOOST_CLASS_TRACKING</span><span class="special">(</span><span class="identifier">gps_position</span><span class="special">,</span><span class="identifier">track_never</span><span class="special">)</span>
1148 <span class="identifier">BOOST_CLASS_IMPLEMENTATION</span><span class="special">(</span><span class="identifier">gps_position</span><span class="special">,</span><span class="identifier">object_serializable</span><span class="special">)</span>
1151 <div class="section">
1152 <div class="titlepage"><div><div><h4 class="title">
1153 <a name="mpi.tutorial.performance_optimizations.homogeneous_machines"></a><a class="link" href="tutorial.html#mpi.tutorial.performance_optimizations.homogeneous_machines" title="Homogeneous Machines">Homogeneous
1155 </h4></div></div></div>
1157 More optimizations are possible on homogeneous machines, by avoiding MPI_Pack/MPI_Unpack
1158 calls but using direct bitwise copy. This feature is enabled by default
1159 by defining the macro <code class="computeroutput"><a class="link" href="../BOOST_MPI_HOMOGENEOUS.html" title="Macro BOOST_MPI_HOMOGENEOUS">BOOST_MPI_HOMOGENEOUS</a></code>
1160 in the include file <code class="computeroutput"><span class="identifier">boost</span><span class="special">/</span><span class="identifier">mpi</span><span class="special">/</span><span class="identifier">config</span><span class="special">.</span><span class="identifier">hpp</span></code>.
1161 That definition must be consistent when building Boost.MPI and when building
1165 In addition all classes need to be marked both as is_mpi_datatype and as
1166 is_bitwise_serializable, by using the helper macro of Boost.Serialization:
1168 <pre class="programlisting"><span class="identifier">BOOST_IS_BITWISE_SERIALIZABLE</span><span class="special">(</span><span class="identifier">gps_position</span><span class="special">)</span>
1171 Usually it is safe to serialize a class for which is_mpi_datatype is true
1172 by using binary copy of the bits. The exception are classes for which some
1173 members should be skipped for serialization.
1177 <div class="footnotes">
1178 <br><hr style="width:100; text-align:left;margin-left: 0">
1179 <div id="ftn.mpi.tutorial.f0" class="footnote"><p><a href="#mpi.tutorial.f0" class="para"><sup class="para">[11] </sup></a>
1180 According to the MPI standard, initialization must take place at user's
1181 initiative after once the main function has been called.
1185 <table xmlns:rev="http://www.cs.rpi.edu/~gregod/boost/tools/doc/revision" width="100%"><tr>
1186 <td align="left"></td>
1187 <td align="right"><div class="copyright-footer">Copyright © 2005-2007 Douglas Gregor,
1188 Matthias Troyer, Trustees of Indiana University<p>
1189 Distributed under the Boost Software License, Version 1.0. (See accompanying
1190 file LICENSE_1_0.txt or copy at <ulink url="http://www.boost.org/LICENSE_1_0.txt">
1191 http://www.boost.org/LICENSE_1_0.txt </ulink>)
1196 <div class="spirit-nav">
1197 <a accesskey="p" href="getting_started.html"><img src="../../../doc/src/images/prev.png" alt="Prev"></a><a accesskey="u" href="../mpi.html"><img src="../../../doc/src/images/up.png" alt="Up"></a><a accesskey="h" href="../index.html"><img src="../../../doc/src/images/home.png" alt="Home"></a><a accesskey="n" href="c_mapping.html"><img src="../../../doc/src/images/next.png" alt="Next"></a>