Mercurial > hg > sv-dependency-builds
diff src/fftw-3.3.3/doc/html/FFTW-MPI-Fortran-Interface.html @ 10:37bf6b4a2645
Add FFTW3
| author | Chris Cannam |
|---|---|
| date | Wed, 20 Mar 2013 15:35:50 +0000 |
| parents | |
| children |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/fftw-3.3.3/doc/html/FFTW-MPI-Fortran-Interface.html Wed Mar 20 15:35:50 2013 +0000 @@ -0,0 +1,189 @@ +<html lang="en"> +<head> +<title>FFTW MPI Fortran Interface - 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="FFTW-MPI-Reference.html#FFTW-MPI-Reference" title="FFTW MPI Reference"> +<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="FFTW-MPI-Fortran-Interface"></a> +<p> +Previous: <a rel="previous" accesskey="p" href="FFTW-MPI-Reference.html#FFTW-MPI-Reference">FFTW MPI Reference</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">6.13 FFTW MPI Fortran Interface</h3> + +<p><a name="index-Fortran-interface-494"></a> +<a name="index-iso_005fc_005fbinding-495"></a>The FFTW MPI interface is callable from modern Fortran compilers +supporting the Fortran 2003 <code>iso_c_binding</code> standard for calling +C functions. As described in <a href="Calling-FFTW-from-Modern-Fortran.html#Calling-FFTW-from-Modern-Fortran">Calling FFTW from Modern Fortran</a>, +this means that you can directly call FFTW's C interface from Fortran +with only minor changes in syntax. There are, however, a few things +specific to the MPI interface to keep in mind: + + <ul> +<li>Instead of including <code>fftw3.f03</code> as in <a href="Overview-of-Fortran-interface.html#Overview-of-Fortran-interface">Overview of Fortran interface</a>, you should <code>include 'fftw3-mpi.f03'</code> (after +<code>use, intrinsic :: iso_c_binding</code> as before). The +<code>fftw3-mpi.f03</code> file includes <code>fftw3.f03</code>, so you should +<em>not</em> <code>include</code> them both yourself. (You will also want to +include the MPI header file, usually via <code>include 'mpif.h'</code> or +similar, although though this is not needed by <code>fftw3-mpi.f03</code> +<i>per se</i>.) (To use the ‘<samp><span class="samp">fftwl_</span></samp>’ <code>long double</code> extended-precision routines in supporting compilers, you should include <code>fftw3f-mpi.f03</code> in <em>addition</em> to <code>fftw3-mpi.f03</code>. See <a href="Extended-and-quadruple-precision-in-Fortran.html#Extended-and-quadruple-precision-in-Fortran">Extended and quadruple precision in Fortran</a>.) + + <li>Because of the different storage conventions between C and Fortran, +you reverse the order of your array dimensions when passing them to +FFTW (see <a href="Reversing-array-dimensions.html#Reversing-array-dimensions">Reversing array dimensions</a>). This is merely a +difference in notation and incurs no performance overhead. However, +it means that, whereas in C the <em>first</em> dimension is distributed, +in Fortran the <em>last</em> dimension of your array is distributed. + + <li><a name="index-MPI-communicator-496"></a>In Fortran, communicators are stored as <code>integer</code> types; there is +no <code>MPI_Comm</code> type, nor is there any way to access a C +<code>MPI_Comm</code>. Fortunately, this is taken care of for you by the +FFTW Fortran interface: whenever the C interface expects an +<code>MPI_Comm</code> type, you should pass the Fortran communicator as an +<code>integer</code>.<a rel="footnote" href="#fn-1" name="fnd-1"><sup>1</sup></a> + + <li>Because you need to call the ‘<samp><span class="samp">local_size</span></samp>’ function to find out +how much space to allocate, and this may be <em>larger</em> than the +local portion of the array (see <a href="MPI-Data-Distribution.html#MPI-Data-Distribution">MPI Data Distribution</a>), you should +<em>always</em> allocate your arrays dynamically using FFTW's allocation +routines as described in <a href="Allocating-aligned-memory-in-Fortran.html#Allocating-aligned-memory-in-Fortran">Allocating aligned memory in Fortran</a>. +(Coincidentally, this also provides the best performance by +guaranteeding proper data alignment.) + + <li>Because all sizes in the MPI FFTW interface are declared as +<code>ptrdiff_t</code> in C, you should use <code>integer(C_INTPTR_T)</code> in +Fortran (see <a href="FFTW-Fortran-type-reference.html#FFTW-Fortran-type-reference">FFTW Fortran type reference</a>). + + <li><a name="index-fftw_005fexecute_005fdft-497"></a><a name="index-fftw_005fmpi_005fexecute_005fdft-498"></a><a name="index-new_002darray-execution-499"></a>In Fortran, because of the language semantics, we generally recommend +using the new-array execute functions for all plans, even in the +common case where you are executing the plan on the same arrays for +which the plan was created (see <a href="Plan-execution-in-Fortran.html#Plan-execution-in-Fortran">Plan execution in Fortran</a>). +However, note that in the MPI interface these functions are changed: +<code>fftw_execute_dft</code> becomes <code>fftw_mpi_execute_dft</code>, +etcetera. See <a href="Using-MPI-Plans.html#Using-MPI-Plans">Using MPI Plans</a>. + + </ul> + + <p>For example, here is a Fortran code snippet to perform a distributed +L × M complex DFT in-place. (This assumes you have already +initialized MPI with <code>MPI_init</code> and have also performed +<code>call fftw_mpi_init</code>.) + +<pre class="example"> use, intrinsic :: iso_c_binding + include 'fftw3-mpi.f03' + integer(C_INTPTR_T), parameter :: L = ... + integer(C_INTPTR_T), parameter :: M = ... + type(C_PTR) :: plan, cdata + complex(C_DOUBLE_COMPLEX), pointer :: data(:,:) + integer(C_INTPTR_T) :: i, j, alloc_local, local_M, local_j_offset + + ! <span class="roman">get local data size and allocate (note dimension reversal)</span> + alloc_local = fftw_mpi_local_size_2d(M, L, MPI_COMM_WORLD, & + local_M, local_j_offset) + cdata = fftw_alloc_complex(alloc_local) + call c_f_pointer(cdata, data, [L,local_M]) + + ! <span class="roman">create MPI plan for in-place forward DFT (note dimension reversal)</span> + plan = fftw_mpi_plan_dft_2d(M, L, data, data, MPI_COMM_WORLD, & + FFTW_FORWARD, FFTW_MEASURE) + + ! <span class="roman">initialize data to some function</span> my_function(i,j) + do j = 1, local_M + do i = 1, L + data(i, j) = my_function(i, j + local_j_offset) + end do + end do + + ! <span class="roman">compute transform (as many times as desired)</span> + call fftw_mpi_execute_dft(plan, data, data) + + call fftw_destroy_plan(plan) + call fftw_free(cdata) +</pre> + <p>Note that when we called <code>fftw_mpi_local_size_2d</code> and +<code>fftw_mpi_plan_dft_2d</code> with the dimensions in reversed order, +since a L × M Fortran array is viewed by FFTW in C as a +M × L array. This means that the array was distributed over +the <code>M</code> dimension, the local portion of which is a +L × local_M array in Fortran. (You must <em>not</em> use an +<code>allocate</code> statement to allocate an L × local_M array, +however; you must allocate <code>alloc_local</code> complex numbers, which +may be greater than <code>L * local_M</code>, in order to reserve space for +intermediate steps of the transform.) Finally, we mention that +because C's array indices are zero-based, the <code>local_j_offset</code> +argument can conveniently be interpreted as an offset in the 1-based +<code>j</code> index (rather than as a starting index as in C). + + <p>If instead you had used the <code>ior(FFTW_MEASURE, +FFTW_MPI_TRANSPOSED_OUT)</code> flag, the output of the transform would be a +transposed M × local_L array, associated with the <em>same</em> +<code>cdata</code> allocation (since the transform is in-place), and which +you could declare with: + +<pre class="example"> complex(C_DOUBLE_COMPLEX), pointer :: tdata(:,:) + ... + call c_f_pointer(cdata, tdata, [M,local_L]) +</pre> + <p>where <code>local_L</code> would have been obtained by changing the +<code>fftw_mpi_local_size_2d</code> call to: + +<pre class="example"> alloc_local = fftw_mpi_local_size_2d_transposed(M, L, MPI_COMM_WORLD, & + local_M, local_j_offset, local_L, local_i_offset) +</pre> + <div class="footnote"> +<hr> +<h4>Footnotes</h4><p class="footnote"><small>[<a name="fn-1" href="#fnd-1">1</a>]</small> Technically, this is because you aren't +actually calling the C functions directly. You are calling wrapper +functions that translate the communicator with <code>MPI_Comm_f2c</code> +before calling the ordinary C interface. This is all done +transparently, however, since the <code>fftw3-mpi.f03</code> interface file +renames the wrappers so that they are called in Fortran with the same +names as the C interface functions.</p> + + <hr></div> + + </body></html> +