annotate src/fftw-3.3.8/doc/html/FFTW-MPI-Wisdom.html @ 169:223a55898ab9 tip default

Add null config files
author Chris Cannam <cannam@all-day-breakfast.com>
date Mon, 02 Mar 2020 14:03:47 +0000
parents bd3cc4d1df30
children
rev   line source
cannam@167 1 <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
cannam@167 2 <html>
cannam@167 3 <!-- This manual is for FFTW
cannam@167 4 (version 3.3.8, 24 May 2018).
cannam@167 5
cannam@167 6 Copyright (C) 2003 Matteo Frigo.
cannam@167 7
cannam@167 8 Copyright (C) 2003 Massachusetts Institute of Technology.
cannam@167 9
cannam@167 10 Permission is granted to make and distribute verbatim copies of this
cannam@167 11 manual provided the copyright notice and this permission notice are
cannam@167 12 preserved on all copies.
cannam@167 13
cannam@167 14 Permission is granted to copy and distribute modified versions of this
cannam@167 15 manual under the conditions for verbatim copying, provided that the
cannam@167 16 entire resulting derived work is distributed under the terms of a
cannam@167 17 permission notice identical to this one.
cannam@167 18
cannam@167 19 Permission is granted to copy and distribute translations of this manual
cannam@167 20 into another language, under the above conditions for modified versions,
cannam@167 21 except that this permission notice may be stated in a translation
cannam@167 22 approved by the Free Software Foundation. -->
cannam@167 23 <!-- Created by GNU Texinfo 6.3, http://www.gnu.org/software/texinfo/ -->
cannam@167 24 <head>
cannam@167 25 <title>FFTW 3.3.8: FFTW MPI Wisdom</title>
cannam@167 26
cannam@167 27 <meta name="description" content="FFTW 3.3.8: FFTW MPI Wisdom">
cannam@167 28 <meta name="keywords" content="FFTW 3.3.8: FFTW MPI Wisdom">
cannam@167 29 <meta name="resource-type" content="document">
cannam@167 30 <meta name="distribution" content="global">
cannam@167 31 <meta name="Generator" content="makeinfo">
cannam@167 32 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
cannam@167 33 <link href="index.html#Top" rel="start" title="Top">
cannam@167 34 <link href="Concept-Index.html#Concept-Index" rel="index" title="Concept Index">
cannam@167 35 <link href="index.html#SEC_Contents" rel="contents" title="Table of Contents">
cannam@167 36 <link href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" rel="up" title="Distributed-memory FFTW with MPI">
cannam@167 37 <link href="Avoiding-MPI-Deadlocks.html#Avoiding-MPI-Deadlocks" rel="next" title="Avoiding MPI Deadlocks">
cannam@167 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@167 39 <style type="text/css">
cannam@167 40 <!--
cannam@167 41 a.summary-letter {text-decoration: none}
cannam@167 42 blockquote.indentedblock {margin-right: 0em}
cannam@167 43 blockquote.smallindentedblock {margin-right: 0em; font-size: smaller}
cannam@167 44 blockquote.smallquotation {font-size: smaller}
cannam@167 45 div.display {margin-left: 3.2em}
cannam@167 46 div.example {margin-left: 3.2em}
cannam@167 47 div.lisp {margin-left: 3.2em}
cannam@167 48 div.smalldisplay {margin-left: 3.2em}
cannam@167 49 div.smallexample {margin-left: 3.2em}
cannam@167 50 div.smalllisp {margin-left: 3.2em}
cannam@167 51 kbd {font-style: oblique}
cannam@167 52 pre.display {font-family: inherit}
cannam@167 53 pre.format {font-family: inherit}
cannam@167 54 pre.menu-comment {font-family: serif}
cannam@167 55 pre.menu-preformatted {font-family: serif}
cannam@167 56 pre.smalldisplay {font-family: inherit; font-size: smaller}
cannam@167 57 pre.smallexample {font-size: smaller}
cannam@167 58 pre.smallformat {font-family: inherit; font-size: smaller}
cannam@167 59 pre.smalllisp {font-size: smaller}
cannam@167 60 span.nolinebreak {white-space: nowrap}
cannam@167 61 span.roman {font-family: initial; font-weight: normal}
cannam@167 62 span.sansserif {font-family: sans-serif; font-weight: normal}
cannam@167 63 ul.no-bullet {list-style: none}
cannam@167 64 -->
cannam@167 65 </style>
cannam@167 66
cannam@167 67
cannam@167 68 </head>
cannam@167 69
cannam@167 70 <body lang="en">
cannam@167 71 <a name="FFTW-MPI-Wisdom"></a>
cannam@167 72 <div class="header">
cannam@167 73 <p>
cannam@167 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>
cannam@167 75 </div>
cannam@167 76 <hr>
cannam@167 77 <a name="FFTW-MPI-Wisdom-1"></a>
cannam@167 78 <h3 class="section">6.8 FFTW MPI Wisdom</h3>
cannam@167 79 <a name="index-wisdom-2"></a>
cannam@167 80 <a name="index-saving-plans-to-disk-2"></a>
cannam@167 81
cannam@167 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
cannam@167 83 be used to save MPI plans as well as to save uniprocessor plans.
cannam@167 84 However, for MPI there are several unavoidable complications.
cannam@167 85 </p>
cannam@167 86 <a name="index-MPI-I_002fO-1"></a>
cannam@167 87 <p>First, the MPI standard does not guarantee that every process can
cannam@167 88 perform file I/O (at least, not using C stdio routines)&mdash;in general,
cannam@167 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
cannam@167 90 want to export the wisdom from a single process to a file, we must
cannam@167 91 first export the wisdom to a string, then send it to process 0, then
cannam@167 92 write it to a file.
cannam@167 93 </p>
cannam@167 94 <p>Second, in principle we may want to have separate wisdom for every
cannam@167 95 process, since in general the processes may run on different hardware
cannam@167 96 even for a single MPI program. However, in practice FFTW&rsquo;s MPI code
cannam@167 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
cannam@167 98 for every process. Thus, we need a mechanism to synchronize the wisdom.
cannam@167 99 </p>
cannam@167 100 <p>To address both of these problems, FFTW provides the following two
cannam@167 101 functions:
cannam@167 102 </p>
cannam@167 103 <div class="example">
cannam@167 104 <pre class="example">void fftw_mpi_broadcast_wisdom(MPI_Comm comm);
cannam@167 105 void fftw_mpi_gather_wisdom(MPI_Comm comm);
cannam@167 106 </pre></div>
cannam@167 107 <a name="index-fftw_005fmpi_005fgather_005fwisdom"></a>
cannam@167 108 <a name="index-fftw_005fmpi_005fbroadcast_005fwisdom"></a>
cannam@167 109
cannam@167 110 <p>Given a communicator <code>comm</code>, <code>fftw_mpi_broadcast_wisdom</code>
cannam@167 111 will broadcast the wisdom from process 0 to all other processes.
cannam@167 112 Conversely, <code>fftw_mpi_gather_wisdom</code> will collect wisdom from all
cannam@167 113 processes onto process 0. (If the plans created for the same problem
cannam@167 114 by different processes are not the same, <code>fftw_mpi_gather_wisdom</code>
cannam@167 115 will arbitrarily choose one of the plans.) Both of these functions
cannam@167 116 may result in suboptimal plans for different processes if the
cannam@167 117 processes are running on non-identical hardware. Both of these
cannam@167 118 functions are <em>collective</em> calls, which means that they must be
cannam@167 119 executed by all processes in the communicator.
cannam@167 120 <a name="index-collective-function-1"></a>
cannam@167 121 </p>
cannam@167 122
cannam@167 123 <p>So, for example, a typical code snippet to import wisdom from a file
cannam@167 124 and use it on all processes would be:
cannam@167 125 </p>
cannam@167 126 <div class="example">
cannam@167 127 <pre class="example">{
cannam@167 128 int rank;
cannam@167 129
cannam@167 130 fftw_mpi_init();
cannam@167 131 MPI_Comm_rank(MPI_COMM_WORLD, &amp;rank);
cannam@167 132 if (rank == 0) fftw_import_wisdom_from_filename(&quot;mywisdom&quot;);
cannam@167 133 fftw_mpi_broadcast_wisdom(MPI_COMM_WORLD);
cannam@167 134 }
cannam@167 135 </pre></div>
cannam@167 136
cannam@167 137 <p>(Note that we must call <code>fftw_mpi_init</code> before importing any
cannam@167 138 wisdom that might contain MPI plans.) Similarly, a typical code
cannam@167 139 snippet to export wisdom from all processes to a file is:
cannam@167 140 <a name="index-fftw_005fmpi_005finit-2"></a>
cannam@167 141 </p>
cannam@167 142 <div class="example">
cannam@167 143 <pre class="example">{
cannam@167 144 int rank;
cannam@167 145
cannam@167 146 fftw_mpi_gather_wisdom(MPI_COMM_WORLD);
cannam@167 147 MPI_Comm_rank(MPI_COMM_WORLD, &amp;rank);
cannam@167 148 if (rank == 0) fftw_export_wisdom_to_filename(&quot;mywisdom&quot;);
cannam@167 149 }
cannam@167 150 </pre></div>
cannam@167 151
cannam@167 152 <div class="footnote">
cannam@167 153 <hr>
cannam@167 154 <h4 class="footnotes-heading">Footnotes</h4>
cannam@167 155
cannam@167 156 <h3><a name="FOOT7" href="#DOCF7">(7)</a></h3>
cannam@167 157 <p>In fact,
cannam@167 158 even this assumption is not technically guaranteed by the standard,
cannam@167 159 although it seems to be universal in actual MPI implementations and is
cannam@167 160 widely assumed by MPI-using software. Technically, you need to query
cannam@167 161 the <code>MPI_IO</code> attribute of <code>MPI_COMM_WORLD</code> with
cannam@167 162 <code>MPI_Attr_get</code>. If this attribute is <code>MPI_PROC_NULL</code>, no
cannam@167 163 I/O is possible. If it is <code>MPI_ANY_SOURCE</code>, any process can
cannam@167 164 perform I/O. Otherwise, it is the rank of a process that can perform
cannam@167 165 I/O ... but since it is not guaranteed to yield the <em>same</em> rank
cannam@167 166 on all processes, you have to do an <code>MPI_Allreduce</code> of some kind
cannam@167 167 if you want all processes to agree about which is going to do I/O.
cannam@167 168 And even then, the standard only guarantees that this process can
cannam@167 169 perform output, but not input. See e.g. <cite>Parallel Programming
cannam@167 170 with MPI</cite> by P. S. Pacheco, section 8.1.3. Needless to say, in our
cannam@167 171 experience virtually no MPI programmers worry about this.</p>
cannam@167 172 </div>
cannam@167 173 <hr>
cannam@167 174 <div class="header">
cannam@167 175 <p>
cannam@167 176 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>
cannam@167 177 </div>
cannam@167 178
cannam@167 179
cannam@167 180
cannam@167 181 </body>
cannam@167 182 </html>