Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.8/doc/html/FFTW-MPI-Wisdom.html @ 82:d0c2a83c1364

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
rev   line source
Chris@82 1 <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
Chris@82 2 <html>
Chris@82 3 <!-- This manual is for FFTW
Chris@82 4 (version 3.3.8, 24 May 2018).
Chris@82 5
Chris@82 6 Copyright (C) 2003 Matteo Frigo.
Chris@82 7
Chris@82 8 Copyright (C) 2003 Massachusetts Institute of Technology.
Chris@82 9
Chris@82 10 Permission is granted to make and distribute verbatim copies of this
Chris@82 11 manual provided the copyright notice and this permission notice are
Chris@82 12 preserved on all copies.
Chris@82 13
Chris@82 14 Permission is granted to copy and distribute modified versions of this
Chris@82 15 manual under the conditions for verbatim copying, provided that the
Chris@82 16 entire resulting derived work is distributed under the terms of a
Chris@82 17 permission notice identical to this one.
Chris@82 18
Chris@82 19 Permission is granted to copy and distribute translations of this manual
Chris@82 20 into another language, under the above conditions for modified versions,
Chris@82 21 except that this permission notice may be stated in a translation
Chris@82 22 approved by the Free Software Foundation. -->
Chris@82 23 <!-- Created by GNU Texinfo 6.3, http://www.gnu.org/software/texinfo/ -->
Chris@82 24 <head>
Chris@82 25 <title>FFTW 3.3.8: FFTW MPI Wisdom</title>
Chris@82 26
Chris@82 27 <meta name="description" content="FFTW 3.3.8: FFTW MPI Wisdom">
Chris@82 28 <meta name="keywords" content="FFTW 3.3.8: FFTW MPI Wisdom">
Chris@82 29 <meta name="resource-type" content="document">
Chris@82 30 <meta name="distribution" content="global">
Chris@82 31 <meta name="Generator" content="makeinfo">
Chris@82 32 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
Chris@82 33 <link href="index.html#Top" rel="start" title="Top">
Chris@82 34 <link href="Concept-Index.html#Concept-Index" rel="index" title="Concept Index">
Chris@82 35 <link href="index.html#SEC_Contents" rel="contents" title="Table of Contents">
Chris@82 36 <link href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" rel="up" title="Distributed-memory FFTW with MPI">
Chris@82 37 <link href="Avoiding-MPI-Deadlocks.html#Avoiding-MPI-Deadlocks" rel="next" title="Avoiding MPI Deadlocks">
Chris@82 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">
Chris@82 39 <style type="text/css">
Chris@82 40 <!--
Chris@82 41 a.summary-letter {text-decoration: none}
Chris@82 42 blockquote.indentedblock {margin-right: 0em}
Chris@82 43 blockquote.smallindentedblock {margin-right: 0em; font-size: smaller}
Chris@82 44 blockquote.smallquotation {font-size: smaller}
Chris@82 45 div.display {margin-left: 3.2em}
Chris@82 46 div.example {margin-left: 3.2em}
Chris@82 47 div.lisp {margin-left: 3.2em}
Chris@82 48 div.smalldisplay {margin-left: 3.2em}
Chris@82 49 div.smallexample {margin-left: 3.2em}
Chris@82 50 div.smalllisp {margin-left: 3.2em}
Chris@82 51 kbd {font-style: oblique}
Chris@82 52 pre.display {font-family: inherit}
Chris@82 53 pre.format {font-family: inherit}
Chris@82 54 pre.menu-comment {font-family: serif}
Chris@82 55 pre.menu-preformatted {font-family: serif}
Chris@82 56 pre.smalldisplay {font-family: inherit; font-size: smaller}
Chris@82 57 pre.smallexample {font-size: smaller}
Chris@82 58 pre.smallformat {font-family: inherit; font-size: smaller}
Chris@82 59 pre.smalllisp {font-size: smaller}
Chris@82 60 span.nolinebreak {white-space: nowrap}
Chris@82 61 span.roman {font-family: initial; font-weight: normal}
Chris@82 62 span.sansserif {font-family: sans-serif; font-weight: normal}
Chris@82 63 ul.no-bullet {list-style: none}
Chris@82 64 -->
Chris@82 65 </style>
Chris@82 66
Chris@82 67
Chris@82 68 </head>
Chris@82 69
Chris@82 70 <body lang="en">
Chris@82 71 <a name="FFTW-MPI-Wisdom"></a>
Chris@82 72 <div class="header">
Chris@82 73 <p>
Chris@82 74 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> &nbsp; [<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>
Chris@82 75 </div>
Chris@82 76 <hr>
Chris@82 77 <a name="FFTW-MPI-Wisdom-1"></a>
Chris@82 78 <h3 class="section">6.8 FFTW MPI Wisdom</h3>
Chris@82 79 <a name="index-wisdom-2"></a>
Chris@82 80 <a name="index-saving-plans-to-disk-2"></a>
Chris@82 81
Chris@82 82 <p>FFTW&rsquo;s &ldquo;wisdom&rdquo; facility (see <a href="Words-of-Wisdom_002dSaving-Plans.html#Words-of-Wisdom_002dSaving-Plans">Words of Wisdom-Saving Plans</a>) can
Chris@82 83 be used to save MPI plans as well as to save uniprocessor plans.
Chris@82 84 However, for MPI there are several unavoidable complications.
Chris@82 85 </p>
Chris@82 86 <a name="index-MPI-I_002fO-1"></a>
Chris@82 87 <p>First, the MPI standard does not guarantee that every process can
Chris@82 88 perform file I/O (at least, not using C stdio routines)&mdash;in general,
Chris@82 89 we may only assume that process 0 is capable of I/O.<a name="DOCF7" href="#FOOT7"><sup>7</sup></a> So, if we
Chris@82 90 want to export the wisdom from a single process to a file, we must
Chris@82 91 first export the wisdom to a string, then send it to process 0, then
Chris@82 92 write it to a file.
Chris@82 93 </p>
Chris@82 94 <p>Second, in principle we may want to have separate wisdom for every
Chris@82 95 process, since in general the processes may run on different hardware
Chris@82 96 even for a single MPI program. However, in practice FFTW&rsquo;s MPI code
Chris@82 97 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
Chris@82 98 for every process. Thus, we need a mechanism to synchronize the wisdom.
Chris@82 99 </p>
Chris@82 100 <p>To address both of these problems, FFTW provides the following two
Chris@82 101 functions:
Chris@82 102 </p>
Chris@82 103 <div class="example">
Chris@82 104 <pre class="example">void fftw_mpi_broadcast_wisdom(MPI_Comm comm);
Chris@82 105 void fftw_mpi_gather_wisdom(MPI_Comm comm);
Chris@82 106 </pre></div>
Chris@82 107 <a name="index-fftw_005fmpi_005fgather_005fwisdom"></a>
Chris@82 108 <a name="index-fftw_005fmpi_005fbroadcast_005fwisdom"></a>
Chris@82 109
Chris@82 110 <p>Given a communicator <code>comm</code>, <code>fftw_mpi_broadcast_wisdom</code>
Chris@82 111 will broadcast the wisdom from process 0 to all other processes.
Chris@82 112 Conversely, <code>fftw_mpi_gather_wisdom</code> will collect wisdom from all
Chris@82 113 processes onto process 0. (If the plans created for the same problem
Chris@82 114 by different processes are not the same, <code>fftw_mpi_gather_wisdom</code>
Chris@82 115 will arbitrarily choose one of the plans.) Both of these functions
Chris@82 116 may result in suboptimal plans for different processes if the
Chris@82 117 processes are running on non-identical hardware. Both of these
Chris@82 118 functions are <em>collective</em> calls, which means that they must be
Chris@82 119 executed by all processes in the communicator.
Chris@82 120 <a name="index-collective-function-1"></a>
Chris@82 121 </p>
Chris@82 122
Chris@82 123 <p>So, for example, a typical code snippet to import wisdom from a file
Chris@82 124 and use it on all processes would be:
Chris@82 125 </p>
Chris@82 126 <div class="example">
Chris@82 127 <pre class="example">{
Chris@82 128 int rank;
Chris@82 129
Chris@82 130 fftw_mpi_init();
Chris@82 131 MPI_Comm_rank(MPI_COMM_WORLD, &amp;rank);
Chris@82 132 if (rank == 0) fftw_import_wisdom_from_filename(&quot;mywisdom&quot;);
Chris@82 133 fftw_mpi_broadcast_wisdom(MPI_COMM_WORLD);
Chris@82 134 }
Chris@82 135 </pre></div>
Chris@82 136
Chris@82 137 <p>(Note that we must call <code>fftw_mpi_init</code> before importing any
Chris@82 138 wisdom that might contain MPI plans.) Similarly, a typical code
Chris@82 139 snippet to export wisdom from all processes to a file is:
Chris@82 140 <a name="index-fftw_005fmpi_005finit-2"></a>
Chris@82 141 </p>
Chris@82 142 <div class="example">
Chris@82 143 <pre class="example">{
Chris@82 144 int rank;
Chris@82 145
Chris@82 146 fftw_mpi_gather_wisdom(MPI_COMM_WORLD);
Chris@82 147 MPI_Comm_rank(MPI_COMM_WORLD, &amp;rank);
Chris@82 148 if (rank == 0) fftw_export_wisdom_to_filename(&quot;mywisdom&quot;);
Chris@82 149 }
Chris@82 150 </pre></div>
Chris@82 151
Chris@82 152 <div class="footnote">
Chris@82 153 <hr>
Chris@82 154 <h4 class="footnotes-heading">Footnotes</h4>
Chris@82 155
Chris@82 156 <h3><a name="FOOT7" href="#DOCF7">(7)</a></h3>
Chris@82 157 <p>In fact,
Chris@82 158 even this assumption is not technically guaranteed by the standard,
Chris@82 159 although it seems to be universal in actual MPI implementations and is
Chris@82 160 widely assumed by MPI-using software. Technically, you need to query
Chris@82 161 the <code>MPI_IO</code> attribute of <code>MPI_COMM_WORLD</code> with
Chris@82 162 <code>MPI_Attr_get</code>. If this attribute is <code>MPI_PROC_NULL</code>, no
Chris@82 163 I/O is possible. If it is <code>MPI_ANY_SOURCE</code>, any process can
Chris@82 164 perform I/O. Otherwise, it is the rank of a process that can perform
Chris@82 165 I/O ... but since it is not guaranteed to yield the <em>same</em> rank
Chris@82 166 on all processes, you have to do an <code>MPI_Allreduce</code> of some kind
Chris@82 167 if you want all processes to agree about which is going to do I/O.
Chris@82 168 And even then, the standard only guarantees that this process can
Chris@82 169 perform output, but not input. See e.g. <cite>Parallel Programming
Chris@82 170 with MPI</cite> by P. S. Pacheco, section 8.1.3. Needless to say, in our
Chris@82 171 experience virtually no MPI programmers worry about this.</p>
Chris@82 172 </div>
Chris@82 173 <hr>
Chris@82 174 <div class="header">
Chris@82 175 <p>
Chris@82 176 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> &nbsp; [<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>
Chris@82 177 </div>
Chris@82 178
Chris@82 179
Chris@82 180
Chris@82 181 </body>
Chris@82 182 </html>