cannam@127: <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
cannam@127: <html>
cannam@127: <!-- This manual is for FFTW
cannam@127: (version 3.3.5, 30 July 2016).
cannam@127:
cannam@127: Copyright (C) 2003 Matteo Frigo.
cannam@127:
cannam@127: Copyright (C) 2003 Massachusetts Institute of Technology.
cannam@127:
cannam@127: Permission is granted to make and distribute verbatim copies of this
cannam@127: manual provided the copyright notice and this permission notice are
cannam@127: preserved on all copies.
cannam@127:
cannam@127: Permission is granted to copy and distribute modified versions of this
cannam@127: manual under the conditions for verbatim copying, provided that the
cannam@127: entire resulting derived work is distributed under the terms of a
cannam@127: permission notice identical to this one.
cannam@127:
cannam@127: Permission is granted to copy and distribute translations of this manual
cannam@127: into another language, under the above conditions for modified versions,
cannam@127: except that this permission notice may be stated in a translation
cannam@127: approved by the Free Software Foundation. -->
cannam@127: <!-- Created by GNU Texinfo 5.2, http://www.gnu.org/software/texinfo/ -->
cannam@127: <head>
cannam@127: <title>FFTW 3.3.5: Combining MPI and Threads</title>
cannam@127:
cannam@127: <meta name="description" content="FFTW 3.3.5: Combining MPI and Threads">
cannam@127: <meta name="keywords" content="FFTW 3.3.5: Combining MPI and Threads">
cannam@127: <meta name="resource-type" content="document">
cannam@127: <meta name="distribution" content="global">
cannam@127: <meta name="Generator" content="makeinfo">
cannam@127: <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
cannam@127: <link href="index.html#Top" rel="start" title="Top">
cannam@127: <link href="Concept-Index.html#Concept-Index" rel="index" title="Concept Index">
cannam@127: <link href="index.html#SEC_Contents" rel="contents" title="Table of Contents">
cannam@127: <link href="Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI" rel="up" title="Distributed-memory FFTW with MPI">
cannam@127: <link href="FFTW-MPI-Reference.html#FFTW-MPI-Reference" rel="next" title="FFTW MPI Reference">
cannam@127: <link href="FFTW-MPI-Performance-Tips.html#FFTW-MPI-Performance-Tips" rel="prev" title="FFTW MPI Performance Tips">
cannam@127: <style type="text/css">
cannam@127: <!--
cannam@127: a.summary-letter {text-decoration: none}
cannam@127: blockquote.smallquotation {font-size: smaller}
cannam@127: div.display {margin-left: 3.2em}
cannam@127: div.example {margin-left: 3.2em}
cannam@127: div.indentedblock {margin-left: 3.2em}
cannam@127: div.lisp {margin-left: 3.2em}
cannam@127: div.smalldisplay {margin-left: 3.2em}
cannam@127: div.smallexample {margin-left: 3.2em}
cannam@127: div.smallindentedblock {margin-left: 3.2em; font-size: smaller}
cannam@127: div.smalllisp {margin-left: 3.2em}
cannam@127: kbd {font-style:oblique}
cannam@127: pre.display {font-family: inherit}
cannam@127: pre.format {font-family: inherit}
cannam@127: pre.menu-comment {font-family: serif}
cannam@127: pre.menu-preformatted {font-family: serif}
cannam@127: pre.smalldisplay {font-family: inherit; font-size: smaller}
cannam@127: pre.smallexample {font-size: smaller}
cannam@127: pre.smallformat {font-family: inherit; font-size: smaller}
cannam@127: pre.smalllisp {font-size: smaller}
cannam@127: span.nocodebreak {white-space:nowrap}
cannam@127: span.nolinebreak {white-space:nowrap}
cannam@127: span.roman {font-family:serif; font-weight:normal}
cannam@127: span.sansserif {font-family:sans-serif; font-weight:normal}
cannam@127: ul.no-bullet {list-style: none}
cannam@127: -->
cannam@127: </style>
cannam@127:
cannam@127:
cannam@127: </head>
cannam@127:
cannam@127: <body lang="en" bgcolor="#FFFFFF" text="#000000" link="#0000FF" vlink="#800080" alink="#FF0000">
cannam@127: <a name="Combining-MPI-and-Threads"></a>
cannam@127: <div class="header">
cannam@127: <p>
cannam@127: Next: <a href="FFTW-MPI-Reference.html#FFTW-MPI-Reference" accesskey="n" rel="next">FFTW MPI Reference</a>, Previous: <a href="FFTW-MPI-Performance-Tips.html#FFTW-MPI-Performance-Tips" accesskey="p" rel="prev">FFTW MPI Performance Tips</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> [<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@127: </div>
cannam@127: <hr>
cannam@127: <a name="Combining-MPI-and-Threads-1"></a>
cannam@127: <h3 class="section">6.11 Combining MPI and Threads</h3>
cannam@127: <a name="index-threads-2"></a>
cannam@127:
cannam@127: <p>In certain cases, it may be advantageous to combine MPI
cannam@127: (distributed-memory) and threads (shared-memory) parallelization.
cannam@127: FFTW supports this, with certain caveats. For example, if you have a
cannam@127: cluster of 4-processor shared-memory nodes, you may want to use
cannam@127: threads within the nodes and MPI between the nodes, instead of MPI for
cannam@127: all parallelization.
cannam@127: </p>
cannam@127: <p>In particular, it is possible to seamlessly combine the MPI FFTW
cannam@127: routines with the multi-threaded FFTW routines (see <a href="Multi_002dthreaded-FFTW.html#Multi_002dthreaded-FFTW">Multi-threaded FFTW</a>). However, some care must be taken in the initialization code,
cannam@127: which should look something like this:
cannam@127: </p>
cannam@127: <div class="example">
cannam@127: <pre class="example">int threads_ok;
cannam@127:
cannam@127: int main(int argc, char **argv)
cannam@127: {
cannam@127: int provided;
cannam@127: MPI_Init_thread(&argc, &argv, MPI_THREAD_FUNNELED, &provided);
cannam@127: threads_ok = provided >= MPI_THREAD_FUNNELED;
cannam@127:
cannam@127: if (threads_ok) threads_ok = fftw_init_threads();
cannam@127: fftw_mpi_init();
cannam@127:
cannam@127: ...
cannam@127: if (threads_ok) fftw_plan_with_nthreads(...);
cannam@127: ...
cannam@127:
cannam@127: MPI_Finalize();
cannam@127: }
cannam@127: </pre></div>
cannam@127: <a name="index-fftw_005fmpi_005finit-3"></a>
cannam@127: <a name="index-fftw_005finit_005fthreads-2"></a>
cannam@127: <a name="index-fftw_005fplan_005fwith_005fnthreads-1"></a>
cannam@127:
cannam@127: <p>First, note that instead of calling <code>MPI_Init</code>, you should call
cannam@127: <code>MPI_Init_threads</code>, which is the initialization routine defined
cannam@127: by the MPI-2 standard to indicate to MPI that your program will be
cannam@127: multithreaded. We pass <code>MPI_THREAD_FUNNELED</code>, which indicates
cannam@127: that we will only call MPI routines from the main thread. (FFTW will
cannam@127: launch additional threads internally, but the extra threads will not
cannam@127: call MPI code.) (You may also pass <code>MPI_THREAD_SERIALIZED</code> or
cannam@127: <code>MPI_THREAD_MULTIPLE</code>, which requests additional multithreading
cannam@127: support from the MPI implementation, but this is not required by
cannam@127: FFTW.) The <code>provided</code> parameter returns what level of threads
cannam@127: support is actually supported by your MPI implementation; this
cannam@127: <em>must</em> be at least <code>MPI_THREAD_FUNNELED</code> if you want to call
cannam@127: the FFTW threads routines, so we define a global variable
cannam@127: <code>threads_ok</code> to record this. You should only call
cannam@127: <code>fftw_init_threads</code> or <code>fftw_plan_with_nthreads</code> if
cannam@127: <code>threads_ok</code> is true. For more information on thread safety in
cannam@127: MPI, see the
cannam@127: <a href="http://www.mpi-forum.org/docs/mpi-20-html/node162.htm">MPI and
cannam@127: Threads</a> section of the MPI-2 standard.
cannam@127: <a name="index-thread-safety-2"></a>
cannam@127: </p>
cannam@127:
cannam@127: <p>Second, we must call <code>fftw_init_threads</code> <em>before</em>
cannam@127: <code>fftw_mpi_init</code>. This is critical for technical reasons having
cannam@127: to do with how FFTW initializes its list of algorithms.
cannam@127: </p>
cannam@127: <p>Then, if you call <code>fftw_plan_with_nthreads(N)</code>, <em>every</em> MPI
cannam@127: process will launch (up to) <code>N</code> threads to parallelize its transforms.
cannam@127: </p>
cannam@127: <p>For example, in the hypothetical cluster of 4-processor nodes, you
cannam@127: might wish to launch only a single MPI process per node, and then call
cannam@127: <code>fftw_plan_with_nthreads(4)</code> on each process to use all
cannam@127: processors in the nodes.
cannam@127: </p>
cannam@127: <p>This may or may not be faster than simply using as many MPI processes
cannam@127: as you have processors, however. On the one hand, using threads
cannam@127: within a node eliminates the need for explicit message passing within
cannam@127: the node. On the other hand, FFTW’s transpose routines are not
cannam@127: multi-threaded, and this means that the communications that do take
cannam@127: place will not benefit from parallelization within the node.
cannam@127: Moreover, many MPI implementations already have optimizations to
cannam@127: exploit shared memory when it is available, so adding the
cannam@127: multithreaded FFTW on top of this may be superfluous.
cannam@127: <a name="index-transpose-4"></a>
cannam@127: </p>
cannam@127: <hr>
cannam@127: <div class="header">
cannam@127: <p>
cannam@127: Next: <a href="FFTW-MPI-Reference.html#FFTW-MPI-Reference" accesskey="n" rel="next">FFTW MPI Reference</a>, Previous: <a href="FFTW-MPI-Performance-Tips.html#FFTW-MPI-Performance-Tips" accesskey="p" rel="prev">FFTW MPI Performance Tips</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> [<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@127: </div>
cannam@127:
cannam@127:
cannam@127:
cannam@127: </body>
cannam@127: </html>