<html lang="en">
<head>
<title>MPI Data Distribution - FFTW 3.3.3</title>
<meta http-equiv="Content-Type" content="text/html">
<meta name="description" content="FFTW 3.3.3">
<meta name="generator" content="makeinfo 4.13">
<link title="Top" rel="start" href="index.html#Top">
<link rel="up" href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" title="Distributed-memory FFTW with MPI">
<link rel="prev" href="2d-MPI-example.html#g_t2d-MPI-example" title="2d MPI example">
<link rel="next" href="Multi_002ddimensional-MPI-DFTs-of-Real-Data.html#Multi_002ddimensional-MPI-DFTs-of-Real-Data" title="Multi-dimensional MPI DFTs of Real Data">
<link href="http://www.gnu.org/software/texinfo/" rel="generator-home" title="Texinfo Homepage">
<!--
This manual is for FFTW
(version 3.3.3, 25 November 2012).
Copyright (C) 2003 Matteo Frigo.
Copyright (C) 2003 Massachusetts Institute of Technology.
Permission is granted to make and distribute verbatim copies of
this manual provided the copyright notice and this permission
notice are preserved on all copies.
Permission is granted to copy and distribute modified versions of
this manual under the conditions for verbatim copying, provided
that the entire resulting derived work is distributed under the
terms of a permission notice identical to this one.
Permission is granted to copy and distribute translations of this
manual into another language, under the above conditions for
modified versions, except that this permission notice may be
stated in a translation approved by the Free Software Foundation.
-->
<meta http-equiv="Content-Style-Type" content="text/css">
<style type="text/css"><!--
pre.display { font-family:inherit }
pre.format { font-family:inherit }
pre.smalldisplay { font-family:inherit; font-size:smaller }
pre.smallformat { font-family:inherit; font-size:smaller }
pre.smallexample { font-size:smaller }
pre.smalllisp { font-size:smaller }
span.sc { font-variant:small-caps }
span.roman { font-family:serif; font-weight:normal; }
span.sansserif { font-family:sans-serif; font-weight:normal; }
--></style>
</head>
<body>
<div class="node">
<a name="MPI-Data-Distribution"></a>
<p>
Next: <a rel="next" accesskey="n" href="Multi_002ddimensional-MPI-DFTs-of-Real-Data.html#Multi_002ddimensional-MPI-DFTs-of-Real-Data">Multi-dimensional MPI DFTs of Real Data</a>,
Previous: <a rel="previous" accesskey="p" href="2d-MPI-example.html#g_t2d-MPI-example">2d MPI example</a>,
Up: <a rel="up" accesskey="u" href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI">Distributed-memory FFTW with MPI</a>
<hr>
</div>
<h3 class="section">6.4 MPI Data Distribution</h3>
<p><a name="index-data-distribution-368"></a>
The most important concept to understand in using FFTW's MPI interface
is the data distribution. With a serial or multithreaded FFT, all of
the inputs and outputs are stored as a single contiguous chunk of
memory. With a distributed-memory FFT, the inputs and outputs are
broken into disjoint blocks, one per process.
<p>In particular, FFTW uses a <em>1d block distribution</em> of the data,
distributed along the <em>first dimension</em>. For example, if you
want to perform a 100 × 200 complex DFT, distributed over 4
processes, each process will get a 25 × 200 slice of the data.
That is, process 0 will get rows 0 through 24, process 1 will get rows
25 through 49, process 2 will get rows 50 through 74, and process 3
will get rows 75 through 99. If you take the same array but
distribute it over 3 processes, then it is not evenly divisible so the
different processes will have unequal chunks. FFTW's default choice
in this case is to assign 34 rows to processes 0 and 1, and 32 rows to
process 2.
<a name="index-block-distribution-369"></a>
<p>FFTW provides several ‘<samp><span class="samp">fftw_mpi_local_size</span></samp>’ routines that you can
call to find out what portion of an array is stored on the current
process. In most cases, you should use the default block sizes picked
by FFTW, but it is also possible to specify your own block size. For
example, with a 100 × 200 array on three processes, you can
tell FFTW to use a block size of 40, which would assign 40 rows to
processes 0 and 1, and 20 rows to process 2. FFTW's default is to
divide the data equally among the processes if possible, and as best
it can otherwise. The rows are always assigned in “rank order,”
i.e. process 0 gets the first block of rows, then process 1, and so
on. (You can change this by using <code>MPI_Comm_split</code> to create a
new communicator with re-ordered processes.) However, you should
always call the ‘<samp><span class="samp">fftw_mpi_local_size</span></samp>’ routines, if possible,
rather than trying to predict FFTW's distribution choices.
<p>In particular, it is critical that you allocate the storage size that
is returned by ‘<samp><span class="samp">fftw_mpi_local_size</span></samp>’, which is <em>not</em>
necessarily the size of the local slice of the array. The reason is
that intermediate steps of FFTW's algorithms involve transposing the
array and redistributing the data, so at these intermediate steps FFTW
may require more local storage space (albeit always proportional to
the total size divided by the number of processes). The
‘<samp><span class="samp">fftw_mpi_local_size</span></samp>’ functions know how much storage is required
for these intermediate steps and tell you the correct amount to
allocate.
<ul class="menu">
<li><a accesskey="1" href="Basic-and-advanced-distribution-interfaces.html#Basic-and-advanced-distribution-interfaces">Basic and advanced distribution interfaces</a>
<li><a accesskey="2" href="Load-balancing.html#Load-balancing">Load balancing</a>
<li><a accesskey="3" href="Transposed-distributions.html#Transposed-distributions">Transposed distributions</a>
<li><a accesskey="4" href="One_002ddimensional-distributions.html#One_002ddimensional-distributions">One-dimensional distributions</a>
</ul>
</body></html>
