cannam@127
|
1 <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
|
cannam@127
|
2 <html>
|
cannam@127
|
3 <!-- This manual is for FFTW
|
cannam@127
|
4 (version 3.3.5, 30 July 2016).
|
cannam@127
|
5
|
cannam@127
|
6 Copyright (C) 2003 Matteo Frigo.
|
cannam@127
|
7
|
cannam@127
|
8 Copyright (C) 2003 Massachusetts Institute of Technology.
|
cannam@127
|
9
|
cannam@127
|
10 Permission is granted to make and distribute verbatim copies of this
|
cannam@127
|
11 manual provided the copyright notice and this permission notice are
|
cannam@127
|
12 preserved on all copies.
|
cannam@127
|
13
|
cannam@127
|
14 Permission is granted to copy and distribute modified versions of this
|
cannam@127
|
15 manual under the conditions for verbatim copying, provided that the
|
cannam@127
|
16 entire resulting derived work is distributed under the terms of a
|
cannam@127
|
17 permission notice identical to this one.
|
cannam@127
|
18
|
cannam@127
|
19 Permission is granted to copy and distribute translations of this manual
|
cannam@127
|
20 into another language, under the above conditions for modified versions,
|
cannam@127
|
21 except that this permission notice may be stated in a translation
|
cannam@127
|
22 approved by the Free Software Foundation. -->
|
cannam@127
|
23 <!-- Created by GNU Texinfo 5.2, http://www.gnu.org/software/texinfo/ -->
|
cannam@127
|
24 <head>
|
cannam@127
|
25 <title>FFTW 3.3.5: FFTW MPI Wisdom</title>
|
cannam@127
|
26
|
cannam@127
|
27 <meta name="description" content="FFTW 3.3.5: FFTW MPI Wisdom">
|
cannam@127
|
28 <meta name="keywords" content="FFTW 3.3.5: FFTW MPI Wisdom">
|
cannam@127
|
29 <meta name="resource-type" content="document">
|
cannam@127
|
30 <meta name="distribution" content="global">
|
cannam@127
|
31 <meta name="Generator" content="makeinfo">
|
cannam@127
|
32 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
|
cannam@127
|
33 <link href="index.html#Top" rel="start" title="Top">
|
cannam@127
|
34 <link href="Concept-Index.html#Concept-Index" rel="index" title="Concept Index">
|
cannam@127
|
35 <link href="index.html#SEC_Contents" rel="contents" title="Table of Contents">
|
cannam@127
|
36 <link href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" rel="up" title="Distributed-memory FFTW with MPI">
|
cannam@127
|
37 <link href="Avoiding-MPI-Deadlocks.html#Avoiding-MPI-Deadlocks" rel="next" title="Avoiding MPI Deadlocks">
|
cannam@127
|
38 <link href="An-improved-replacement-for-MPI_005fAlltoall.html#An-improved-replacement-for-MPI_005fAlltoall" rel="prev" title="An improved replacement for MPI_Alltoall">
|
cannam@127
|
39 <style type="text/css">
|
cannam@127
|
40 <!--
|
cannam@127
|
41 a.summary-letter {text-decoration: none}
|
cannam@127
|
42 blockquote.smallquotation {font-size: smaller}
|
cannam@127
|
43 div.display {margin-left: 3.2em}
|
cannam@127
|
44 div.example {margin-left: 3.2em}
|
cannam@127
|
45 div.indentedblock {margin-left: 3.2em}
|
cannam@127
|
46 div.lisp {margin-left: 3.2em}
|
cannam@127
|
47 div.smalldisplay {margin-left: 3.2em}
|
cannam@127
|
48 div.smallexample {margin-left: 3.2em}
|
cannam@127
|
49 div.smallindentedblock {margin-left: 3.2em; font-size: smaller}
|
cannam@127
|
50 div.smalllisp {margin-left: 3.2em}
|
cannam@127
|
51 kbd {font-style:oblique}
|
cannam@127
|
52 pre.display {font-family: inherit}
|
cannam@127
|
53 pre.format {font-family: inherit}
|
cannam@127
|
54 pre.menu-comment {font-family: serif}
|
cannam@127
|
55 pre.menu-preformatted {font-family: serif}
|
cannam@127
|
56 pre.smalldisplay {font-family: inherit; font-size: smaller}
|
cannam@127
|
57 pre.smallexample {font-size: smaller}
|
cannam@127
|
58 pre.smallformat {font-family: inherit; font-size: smaller}
|
cannam@127
|
59 pre.smalllisp {font-size: smaller}
|
cannam@127
|
60 span.nocodebreak {white-space:nowrap}
|
cannam@127
|
61 span.nolinebreak {white-space:nowrap}
|
cannam@127
|
62 span.roman {font-family:serif; font-weight:normal}
|
cannam@127
|
63 span.sansserif {font-family:sans-serif; font-weight:normal}
|
cannam@127
|
64 ul.no-bullet {list-style: none}
|
cannam@127
|
65 -->
|
cannam@127
|
66 </style>
|
cannam@127
|
67
|
cannam@127
|
68
|
cannam@127
|
69 </head>
|
cannam@127
|
70
|
cannam@127
|
71 <body lang="en" bgcolor="#FFFFFF" text="#000000" link="#0000FF" vlink="#800080" alink="#FF0000">
|
cannam@127
|
72 <a name="FFTW-MPI-Wisdom"></a>
|
cannam@127
|
73 <div class="header">
|
cannam@127
|
74 <p>
|
cannam@127
|
75 Next: <a href="Avoiding-MPI-Deadlocks.html#Avoiding-MPI-Deadlocks" accesskey="n" rel="next">Avoiding MPI Deadlocks</a>, Previous: <a href="FFTW-MPI-Transposes.html#FFTW-MPI-Transposes" accesskey="p" rel="prev">FFTW MPI Transposes</a>, Up: <a href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" accesskey="u" rel="up">Distributed-memory FFTW with MPI</a> [<a href="index.html#SEC_Contents" title="Table of contents" rel="contents">Contents</a>][<a href="Concept-Index.html#Concept-Index" title="Index" rel="index">Index</a>]</p>
|
cannam@127
|
76 </div>
|
cannam@127
|
77 <hr>
|
cannam@127
|
78 <a name="FFTW-MPI-Wisdom-1"></a>
|
cannam@127
|
79 <h3 class="section">6.8 FFTW MPI Wisdom</h3>
|
cannam@127
|
80 <a name="index-wisdom-2"></a>
|
cannam@127
|
81 <a name="index-saving-plans-to-disk-2"></a>
|
cannam@127
|
82
|
cannam@127
|
83 <p>FFTW’s “wisdom” facility (see <a href="Words-of-Wisdom_002dSaving-Plans.html#Words-of-Wisdom_002dSaving-Plans">Words of Wisdom-Saving Plans</a>) can
|
cannam@127
|
84 be used to save MPI plans as well as to save uniprocessor plans.
|
cannam@127
|
85 However, for MPI there are several unavoidable complications.
|
cannam@127
|
86 </p>
|
cannam@127
|
87 <a name="index-MPI-I_002fO-1"></a>
|
cannam@127
|
88 <p>First, the MPI standard does not guarantee that every process can
|
cannam@127
|
89 perform file I/O (at least, not using C stdio routines)—in general,
|
cannam@127
|
90 we may only assume that process 0 is capable of I/O.<a name="DOCF7" href="#FOOT7"><sup>7</sup></a> So, if we
|
cannam@127
|
91 want to export the wisdom from a single process to a file, we must
|
cannam@127
|
92 first export the wisdom to a string, then send it to process 0, then
|
cannam@127
|
93 write it to a file.
|
cannam@127
|
94 </p>
|
cannam@127
|
95 <p>Second, in principle we may want to have separate wisdom for every
|
cannam@127
|
96 process, since in general the processes may run on different hardware
|
cannam@127
|
97 even for a single MPI program. However, in practice FFTW’s MPI code
|
cannam@127
|
98 is designed for the case of homogeneous hardware (see <a href="Load-balancing.html#Load-balancing">Load balancing</a>), and in this case it is convenient to use the same wisdom
|
cannam@127
|
99 for every process. Thus, we need a mechanism to synchronize the wisdom.
|
cannam@127
|
100 </p>
|
cannam@127
|
101 <p>To address both of these problems, FFTW provides the following two
|
cannam@127
|
102 functions:
|
cannam@127
|
103 </p>
|
cannam@127
|
104 <div class="example">
|
cannam@127
|
105 <pre class="example">void fftw_mpi_broadcast_wisdom(MPI_Comm comm);
|
cannam@127
|
106 void fftw_mpi_gather_wisdom(MPI_Comm comm);
|
cannam@127
|
107 </pre></div>
|
cannam@127
|
108 <a name="index-fftw_005fmpi_005fgather_005fwisdom"></a>
|
cannam@127
|
109 <a name="index-fftw_005fmpi_005fbroadcast_005fwisdom"></a>
|
cannam@127
|
110
|
cannam@127
|
111 <p>Given a communicator <code>comm</code>, <code>fftw_mpi_broadcast_wisdom</code>
|
cannam@127
|
112 will broadcast the wisdom from process 0 to all other processes.
|
cannam@127
|
113 Conversely, <code>fftw_mpi_gather_wisdom</code> will collect wisdom from all
|
cannam@127
|
114 processes onto process 0. (If the plans created for the same problem
|
cannam@127
|
115 by different processes are not the same, <code>fftw_mpi_gather_wisdom</code>
|
cannam@127
|
116 will arbitrarily choose one of the plans.) Both of these functions
|
cannam@127
|
117 may result in suboptimal plans for different processes if the
|
cannam@127
|
118 processes are running on non-identical hardware. Both of these
|
cannam@127
|
119 functions are <em>collective</em> calls, which means that they must be
|
cannam@127
|
120 executed by all processes in the communicator.
|
cannam@127
|
121 <a name="index-collective-function-1"></a>
|
cannam@127
|
122 </p>
|
cannam@127
|
123
|
cannam@127
|
124 <p>So, for example, a typical code snippet to import wisdom from a file
|
cannam@127
|
125 and use it on all processes would be:
|
cannam@127
|
126 </p>
|
cannam@127
|
127 <div class="example">
|
cannam@127
|
128 <pre class="example">{
|
cannam@127
|
129 int rank;
|
cannam@127
|
130
|
cannam@127
|
131 fftw_mpi_init();
|
cannam@127
|
132 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
|
cannam@127
|
133 if (rank == 0) fftw_import_wisdom_from_filename("mywisdom");
|
cannam@127
|
134 fftw_mpi_broadcast_wisdom(MPI_COMM_WORLD);
|
cannam@127
|
135 }
|
cannam@127
|
136 </pre></div>
|
cannam@127
|
137
|
cannam@127
|
138 <p>(Note that we must call <code>fftw_mpi_init</code> before importing any
|
cannam@127
|
139 wisdom that might contain MPI plans.) Similarly, a typical code
|
cannam@127
|
140 snippet to export wisdom from all processes to a file is:
|
cannam@127
|
141 <a name="index-fftw_005fmpi_005finit-2"></a>
|
cannam@127
|
142 </p>
|
cannam@127
|
143 <div class="example">
|
cannam@127
|
144 <pre class="example">{
|
cannam@127
|
145 int rank;
|
cannam@127
|
146
|
cannam@127
|
147 fftw_mpi_gather_wisdom(MPI_COMM_WORLD);
|
cannam@127
|
148 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
|
cannam@127
|
149 if (rank == 0) fftw_export_wisdom_to_filename("mywisdom");
|
cannam@127
|
150 }
|
cannam@127
|
151 </pre></div>
|
cannam@127
|
152
|
cannam@127
|
153 <div class="footnote">
|
cannam@127
|
154 <hr>
|
cannam@127
|
155 <h4 class="footnotes-heading">Footnotes</h4>
|
cannam@127
|
156
|
cannam@127
|
157 <h3><a name="FOOT7" href="#DOCF7">(7)</a></h3>
|
cannam@127
|
158 <p>In fact,
|
cannam@127
|
159 even this assumption is not technically guaranteed by the standard,
|
cannam@127
|
160 although it seems to be universal in actual MPI implementations and is
|
cannam@127
|
161 widely assumed by MPI-using software. Technically, you need to query
|
cannam@127
|
162 the <code>MPI_IO</code> attribute of <code>MPI_COMM_WORLD</code> with
|
cannam@127
|
163 <code>MPI_Attr_get</code>. If this attribute is <code>MPI_PROC_NULL</code>, no
|
cannam@127
|
164 I/O is possible. If it is <code>MPI_ANY_SOURCE</code>, any process can
|
cannam@127
|
165 perform I/O. Otherwise, it is the rank of a process that can perform
|
cannam@127
|
166 I/O ... but since it is not guaranteed to yield the <em>same</em> rank
|
cannam@127
|
167 on all processes, you have to do an <code>MPI_Allreduce</code> of some kind
|
cannam@127
|
168 if you want all processes to agree about which is going to do I/O.
|
cannam@127
|
169 And even then, the standard only guarantees that this process can
|
cannam@127
|
170 perform output, but not input. See e.g. <cite>Parallel Programming
|
cannam@127
|
171 with MPI</cite> by P. S. Pacheco, section 8.1.3. Needless to say, in our
|
cannam@127
|
172 experience virtually no MPI programmers worry about this.</p>
|
cannam@127
|
173 </div>
|
cannam@127
|
174 <hr>
|
cannam@127
|
175 <div class="header">
|
cannam@127
|
176 <p>
|
cannam@127
|
177 Next: <a href="Avoiding-MPI-Deadlocks.html#Avoiding-MPI-Deadlocks" accesskey="n" rel="next">Avoiding MPI Deadlocks</a>, Previous: <a href="FFTW-MPI-Transposes.html#FFTW-MPI-Transposes" accesskey="p" rel="prev">FFTW MPI Transposes</a>, Up: <a href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" accesskey="u" rel="up">Distributed-memory FFTW with MPI</a> [<a href="index.html#SEC_Contents" title="Table of contents" rel="contents">Contents</a>][<a href="Concept-Index.html#Concept-Index" title="Index" rel="index">Index</a>]</p>
|
cannam@127
|
178 </div>
|
cannam@127
|
179
|
cannam@127
|
180
|
cannam@127
|
181
|
cannam@127
|
182 </body>
|
cannam@127
|
183 </html>
|