3
<title>FFTW MPI Performance Tips - FFTW 3.2alpha3</title>
4
<meta http-equiv="Content-Type" content="text/html">
5
<meta name="description" content="FFTW 3.2alpha3">
6
<meta name="generator" content="makeinfo 4.8">
7
<link title="Top" rel="start" href="index.html#Top">
8
<link rel="up" href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" title="Distributed-memory FFTW with MPI">
9
<link rel="prev" href="Avoiding-MPI-Deadlocks.html#Avoiding-MPI-Deadlocks" title="Avoiding MPI Deadlocks">
10
<link rel="next" href="Combining-MPI-and-Threads.html#Combining-MPI-and-Threads" title="Combining MPI and Threads">
11
<link href="http://www.gnu.org/software/texinfo/" rel="generator-home" title="Texinfo Homepage">
13
This manual is for FFTW
14
(version 3.2alpha3, 14 August 2007).
16
Copyright (C) 2003 Matteo Frigo.
18
Copyright (C) 2003 Massachusetts Institute of Technology.
20
Permission is granted to make and distribute verbatim copies of
21
this manual provided the copyright notice and this permission
22
notice are preserved on all copies.
24
Permission is granted to copy and distribute modified versions of
25
this manual under the conditions for verbatim copying, provided
26
that the entire resulting derived work is distributed under the
27
terms of a permission notice identical to this one.
29
Permission is granted to copy and distribute translations of this
30
manual into another language, under the above conditions for
31
modified versions, except that this permission notice may be
32
stated in a translation approved by the Free Software Foundation.
34
<meta http-equiv="Content-Style-Type" content="text/css">
35
<style type="text/css"><!--
36
pre.display { font-family:inherit }
37
pre.format { font-family:inherit }
38
pre.smalldisplay { font-family:inherit; font-size:smaller }
39
pre.smallformat { font-family:inherit; font-size:smaller }
40
pre.smallexample { font-size:smaller }
41
pre.smalllisp { font-size:smaller }
42
span.sc { font-variant:small-caps }
43
span.roman { font-family:serif; font-weight:normal; }
44
span.sansserif { font-family:sans-serif; font-weight:normal; }
50
<a name="FFTW-MPI-Performance-Tips"></a>
51
Next: <a rel="next" accesskey="n" href="Combining-MPI-and-Threads.html#Combining-MPI-and-Threads">Combining MPI and Threads</a>,
52
Previous: <a rel="previous" accesskey="p" href="Avoiding-MPI-Deadlocks.html#Avoiding-MPI-Deadlocks">Avoiding MPI Deadlocks</a>,
53
Up: <a rel="up" accesskey="u" href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI">Distributed-memory FFTW with MPI</a>
57
<h3 class="section">6.10 FFTW MPI Performance Tips</h3>
59
<p>In this section, we collect a few tips on getting the best performance
60
out of FFTW's MPI transforms.
62
<p>First, because of the 1d block distribution, FFTW's parallelization is
63
currently limited by the size of the first dimension.
64
(Multidimensional block distributions may be supported by a future
65
version.) More generally, you should ideally arrange the dimensions so
66
that FFTW can divide them equally among the processes. See <a href="Load-balancing.html#Load-balancing">Load balancing</a>.
67
<a name="index-block-distribution-394"></a><a name="index-load-balancing-395"></a>
68
Second, if it is not too inconvenient, you should consider working
69
with transposed output for multidimensional plans, as this saves a
70
considerable amount of communications. See <a href="Transposed-distributions.html#Transposed-distributions">Transposed distributions</a>.
71
<a name="index-transpose-396"></a>
72
Third, the fastest choices are generally either an in-place transform
73
or an out-of-place transform with the <code>FFTW_DESTROY_INPUT</code> flag
74
(which allows the input array to be used as scratch space). In-place
75
is especially beneficial if the amount of data per process is large.
76
<a name="index-FFTW_005fDESTROY_005fINPUT-397"></a>
77
Fourth, if you have multiple arrays to transform at once, rather than
78
calling FFTW's MPI transforms several times it usually seems to be
79
faster to interleave the data and use the advanced interface. (This
80
groups the communications together instead of requiring separate
81
messages for each transform.)