Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.3/doc/html/An-improved-replacement-for-MPI_005fAlltoall.html @ 10:37bf6b4a2645

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add FFTW3
author Chris Cannam
date Wed, 20 Mar 2013 15:35:50 +0000
parents
children
rev   line source
Chris@10 1 <html lang="en">
Chris@10 2 <head>
Chris@10 3 <title>An improved replacement for MPI_Alltoall - FFTW 3.3.3</title>
Chris@10 4 <meta http-equiv="Content-Type" content="text/html">
Chris@10 5 <meta name="description" content="FFTW 3.3.3">
Chris@10 6 <meta name="generator" content="makeinfo 4.13">
Chris@10 7 <link title="Top" rel="start" href="index.html#Top">
Chris@10 8 <link rel="up" href="FFTW-MPI-Transposes.html#FFTW-MPI-Transposes" title="FFTW MPI Transposes">
Chris@10 9 <link rel="prev" href="Advanced-distributed_002dtranspose-interface.html#Advanced-distributed_002dtranspose-interface" title="Advanced distributed-transpose interface">
Chris@10 10 <link href="http://www.gnu.org/software/texinfo/" rel="generator-home" title="Texinfo Homepage">
Chris@10 11 <!--
Chris@10 12 This manual is for FFTW
Chris@10 13 (version 3.3.3, 25 November 2012).
Chris@10 14
Chris@10 15 Copyright (C) 2003 Matteo Frigo.
Chris@10 16
Chris@10 17 Copyright (C) 2003 Massachusetts Institute of Technology.
Chris@10 18
Chris@10 19 Permission is granted to make and distribute verbatim copies of
Chris@10 20 this manual provided the copyright notice and this permission
Chris@10 21 notice are preserved on all copies.
Chris@10 22
Chris@10 23 Permission is granted to copy and distribute modified versions of
Chris@10 24 this manual under the conditions for verbatim copying, provided
Chris@10 25 that the entire resulting derived work is distributed under the
Chris@10 26 terms of a permission notice identical to this one.
Chris@10 27
Chris@10 28 Permission is granted to copy and distribute translations of this
Chris@10 29 manual into another language, under the above conditions for
Chris@10 30 modified versions, except that this permission notice may be
Chris@10 31 stated in a translation approved by the Free Software Foundation.
Chris@10 32 -->
Chris@10 33 <meta http-equiv="Content-Style-Type" content="text/css">
Chris@10 34 <style type="text/css"><!--
Chris@10 35 pre.display { font-family:inherit }
Chris@10 36 pre.format { font-family:inherit }
Chris@10 37 pre.smalldisplay { font-family:inherit; font-size:smaller }
Chris@10 38 pre.smallformat { font-family:inherit; font-size:smaller }
Chris@10 39 pre.smallexample { font-size:smaller }
Chris@10 40 pre.smalllisp { font-size:smaller }
Chris@10 41 span.sc { font-variant:small-caps }
Chris@10 42 span.roman { font-family:serif; font-weight:normal; }
Chris@10 43 span.sansserif { font-family:sans-serif; font-weight:normal; }
Chris@10 44 --></style>
Chris@10 45 </head>
Chris@10 46 <body>
Chris@10 47 <div class="node">
Chris@10 48 <a name="An-improved-replacement-for-MPI_Alltoall"></a>
Chris@10 49 <a name="An-improved-replacement-for-MPI_005fAlltoall"></a>
Chris@10 50 <p>
Chris@10 51 Previous:&nbsp;<a rel="previous" accesskey="p" href="Advanced-distributed_002dtranspose-interface.html#Advanced-distributed_002dtranspose-interface">Advanced distributed-transpose interface</a>,
Chris@10 52 Up:&nbsp;<a rel="up" accesskey="u" href="FFTW-MPI-Transposes.html#FFTW-MPI-Transposes">FFTW MPI Transposes</a>
Chris@10 53 <hr>
Chris@10 54 </div>
Chris@10 55
Chris@10 56 <h4 class="subsection">6.7.3 An improved replacement for MPI_Alltoall</h4>
Chris@10 57
Chris@10 58 <p>We close this section by noting that FFTW's MPI transpose routines can
Chris@10 59 be thought of as a generalization for the <code>MPI_Alltoall</code> function
Chris@10 60 (albeit only for floating-point types), and in some circumstances can
Chris@10 61 function as an improved replacement.
Chris@10 62 <a name="index-MPI_005fAlltoall-406"></a>
Chris@10 63
Chris@10 64 <p><code>MPI_Alltoall</code> is defined by the MPI standard as:
Chris@10 65
Chris@10 66 <pre class="example"> int MPI_Alltoall(void *sendbuf, int sendcount, MPI_Datatype sendtype,
Chris@10 67 void *recvbuf, int recvcnt, MPI_Datatype recvtype,
Chris@10 68 MPI_Comm comm);
Chris@10 69 </pre>
Chris@10 70 <p>In particular, for <code>double*</code> arrays <code>in</code> and <code>out</code>,
Chris@10 71 consider the call:
Chris@10 72
Chris@10 73 <pre class="example"> MPI_Alltoall(in, howmany, MPI_DOUBLE, out, howmany MPI_DOUBLE, comm);
Chris@10 74 </pre>
Chris@10 75 <p>This is completely equivalent to:
Chris@10 76
Chris@10 77 <pre class="example"> MPI_Comm_size(comm, &amp;P);
Chris@10 78 plan = fftw_mpi_plan_many_transpose(P, P, howmany, 1, 1, in, out, comm, FFTW_ESTIMATE);
Chris@10 79 fftw_execute(plan);
Chris@10 80 fftw_destroy_plan(plan);
Chris@10 81 </pre>
Chris@10 82 <p>That is, computing a P&nbsp;&times;&nbsp;P transpose on <code>P</code> processes,
Chris@10 83 with a block size of 1, is just a standard all-to-all communication.
Chris@10 84
Chris@10 85 <p>However, using the FFTW routine instead of <code>MPI_Alltoall</code> may
Chris@10 86 have certain advantages. First of all, FFTW's routine can operate
Chris@10 87 in-place (<code>in == out</code>) whereas <code>MPI_Alltoall</code> can only
Chris@10 88 operate out-of-place.
Chris@10 89 <a name="index-in_002dplace-407"></a>
Chris@10 90
Chris@10 91 <p>Second, even for out-of-place plans, FFTW's routine may be faster,
Chris@10 92 especially if you need to perform the all-to-all communication many
Chris@10 93 times and can afford to use <code>FFTW_MEASURE</code> or
Chris@10 94 <code>FFTW_PATIENT</code>. It should certainly be no slower, not including
Chris@10 95 the time to create the plan, since one of the possible algorithms that
Chris@10 96 FFTW uses for an out-of-place transpose <em>is</em> simply to call
Chris@10 97 <code>MPI_Alltoall</code>. However, FFTW also considers several other
Chris@10 98 possible algorithms that, depending on your MPI implementation and
Chris@10 99 your hardware, may be faster.
Chris@10 100 <a name="index-FFTW_005fMEASURE-408"></a><a name="index-FFTW_005fPATIENT-409"></a>
Chris@10 101 <!-- -->
Chris@10 102
Chris@10 103 </body></html>
Chris@10 104