Mercurial > hg > sv-dependency-builds
diff src/fftw-3.3.3/doc/html/MPI-Plan-Creation.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/MPI-Plan-Creation.html Wed Mar 20 15:35:50 2013 +0000 @@ -0,0 +1,253 @@ +<html lang="en"> +<head> +<title>MPI Plan Creation - 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="FFTW-MPI-Reference.html#FFTW-MPI-Reference" title="FFTW MPI Reference"> +<link rel="prev" href="MPI-Data-Distribution-Functions.html#MPI-Data-Distribution-Functions" title="MPI Data Distribution Functions"> +<link rel="next" href="MPI-Wisdom-Communication.html#MPI-Wisdom-Communication" title="MPI Wisdom Communication"> +<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-Plan-Creation"></a> +<p> +Next: <a rel="next" accesskey="n" href="MPI-Wisdom-Communication.html#MPI-Wisdom-Communication">MPI Wisdom Communication</a>, +Previous: <a rel="previous" accesskey="p" href="MPI-Data-Distribution-Functions.html#MPI-Data-Distribution-Functions">MPI Data Distribution Functions</a>, +Up: <a rel="up" accesskey="u" href="FFTW-MPI-Reference.html#FFTW-MPI-Reference">FFTW MPI Reference</a> +<hr> +</div> + +<h4 class="subsection">6.12.5 MPI Plan Creation</h4> + +<h5 class="subsubheading">Complex-data MPI DFTs</h5> + +<p>Plans for complex-data DFTs (see <a href="2d-MPI-example.html#g_t2d-MPI-example">2d MPI example</a>) are created by: + + <p><a name="index-fftw_005fmpi_005fplan_005fdft_005f1d-461"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005f2d-462"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005f3d-463"></a><a name="index-fftw_005fmpi_005fplan_005fdft-464"></a><a name="index-fftw_005fmpi_005fplan_005fmany_005fdft-465"></a> +<pre class="example"> fftw_plan fftw_mpi_plan_dft_1d(ptrdiff_t n0, fftw_complex *in, fftw_complex *out, + MPI_Comm comm, int sign, unsigned flags); + fftw_plan fftw_mpi_plan_dft_2d(ptrdiff_t n0, ptrdiff_t n1, + fftw_complex *in, fftw_complex *out, + MPI_Comm comm, int sign, unsigned flags); + fftw_plan fftw_mpi_plan_dft_3d(ptrdiff_t n0, ptrdiff_t n1, ptrdiff_t n2, + fftw_complex *in, fftw_complex *out, + MPI_Comm comm, int sign, unsigned flags); + fftw_plan fftw_mpi_plan_dft(int rnk, const ptrdiff_t *n, + fftw_complex *in, fftw_complex *out, + MPI_Comm comm, int sign, unsigned flags); + fftw_plan fftw_mpi_plan_many_dft(int rnk, const ptrdiff_t *n, + ptrdiff_t howmany, ptrdiff_t block, ptrdiff_t tblock, + fftw_complex *in, fftw_complex *out, + MPI_Comm comm, int sign, unsigned flags); +</pre> + <p><a name="index-MPI-communicator-466"></a><a name="index-collective-function-467"></a>These are similar to their serial counterparts (see <a href="Complex-DFTs.html#Complex-DFTs">Complex DFTs</a>) +in specifying the dimensions, sign, and flags of the transform. The +<code>comm</code> argument gives an MPI communicator that specifies the set +of processes to participate in the transform; plan creation is a +collective function that must be called for all processes in the +communicator. The <code>in</code> and <code>out</code> pointers refer only to a +portion of the overall transform data (see <a href="MPI-Data-Distribution.html#MPI-Data-Distribution">MPI Data Distribution</a>) +as specified by the ‘<samp><span class="samp">local_size</span></samp>’ functions in the previous +section. Unless <code>flags</code> contains <code>FFTW_ESTIMATE</code>, these +arrays are overwritten during plan creation as for the serial +interface. For multi-dimensional transforms, any dimensions <code>> +1</code> are supported; for one-dimensional transforms, only composite +(non-prime) <code>n0</code> are currently supported (unlike the serial +FFTW). Requesting an unsupported transform size will yield a +<code>NULL</code> plan. (As in the serial interface, highly composite sizes +generally yield the best performance.) + + <p><a name="index-advanced-interface-468"></a><a name="index-FFTW_005fMPI_005fDEFAULT_005fBLOCK-469"></a><a name="index-stride-470"></a>The advanced-interface <code>fftw_mpi_plan_many_dft</code> additionally +allows you to specify the block sizes for the first dimension +(<code>block</code>) of the n<sub>0</sub> × n<sub>1</sub> × n<sub>2</sub> × … × n<sub>d-1</sub> input data and the first dimension +(<code>tblock</code>) of the n<sub>1</sub> × n<sub>0</sub> × n<sub>2</sub> ×…× n<sub>d-1</sub> transposed data (at intermediate +steps of the transform, and for the output if +<code>FFTW_TRANSPOSED_OUT</code> is specified in <code>flags</code>). These must +be the same block sizes as were passed to the corresponding +‘<samp><span class="samp">local_size</span></samp>’ function; you can pass <code>FFTW_MPI_DEFAULT_BLOCK</code> +to use FFTW's default block size as in the basic interface. Also, the +<code>howmany</code> parameter specifies that the transform is of contiguous +<code>howmany</code>-tuples rather than individual complex numbers; this +corresponds to the same parameter in the serial advanced interface +(see <a href="Advanced-Complex-DFTs.html#Advanced-Complex-DFTs">Advanced Complex DFTs</a>) with <code>stride = howmany</code> and +<code>dist = 1</code>. + +<h5 class="subsubheading">MPI flags</h5> + +<p>The <code>flags</code> can be any of those for the serial FFTW +(see <a href="Planner-Flags.html#Planner-Flags">Planner Flags</a>), and in addition may include one or more of +the following MPI-specific flags, which improve performance at the +cost of changing the output or input data formats. + + <ul> +<li><a name="index-FFTW_005fMPI_005fSCRAMBLED_005fOUT-471"></a><a name="index-FFTW_005fMPI_005fSCRAMBLED_005fIN-472"></a><code>FFTW_MPI_SCRAMBLED_OUT</code>, <code>FFTW_MPI_SCRAMBLED_IN</code>: valid for +1d transforms only, these flags indicate that the output/input of the +transform are in an undocumented “scrambled” order. A forward +<code>FFTW_MPI_SCRAMBLED_OUT</code> transform can be inverted by a backward +<code>FFTW_MPI_SCRAMBLED_IN</code> (times the usual 1/<i>N</i> normalization). +See <a href="One_002ddimensional-distributions.html#One_002ddimensional-distributions">One-dimensional distributions</a>. + + <li><a name="index-FFTW_005fMPI_005fTRANSPOSED_005fOUT-473"></a><a name="index-FFTW_005fMPI_005fTRANSPOSED_005fIN-474"></a><code>FFTW_MPI_TRANSPOSED_OUT</code>, <code>FFTW_MPI_TRANSPOSED_IN</code>: valid +for multidimensional (<code>rnk > 1</code>) transforms only, these flags +specify that the output or input of an n<sub>0</sub> × n<sub>1</sub> × n<sub>2</sub> × … × n<sub>d-1</sub> transform is +transposed to n<sub>1</sub> × n<sub>0</sub> × n<sub>2</sub> ×…× n<sub>d-1</sub>. See <a href="Transposed-distributions.html#Transposed-distributions">Transposed distributions</a>. + + </ul> + +<h5 class="subsubheading">Real-data MPI DFTs</h5> + +<p><a name="index-r2c-475"></a>Plans for real-input/output (r2c/c2r) DFTs (see <a href="Multi_002ddimensional-MPI-DFTs-of-Real-Data.html#Multi_002ddimensional-MPI-DFTs-of-Real-Data">Multi-dimensional MPI DFTs of Real Data</a>) are created by: + + <p><a name="index-fftw_005fmpi_005fplan_005fdft_005fr2c_005f2d-476"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005fr2c_005f2d-477"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005fr2c_005f3d-478"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005fr2c-479"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005fc2r_005f2d-480"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005fc2r_005f2d-481"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005fc2r_005f3d-482"></a><a name="index-fftw_005fmpi_005fplan_005fdft_005fc2r-483"></a> +<pre class="example"> fftw_plan fftw_mpi_plan_dft_r2c_2d(ptrdiff_t n0, ptrdiff_t n1, + double *in, fftw_complex *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_dft_r2c_2d(ptrdiff_t n0, ptrdiff_t n1, + double *in, fftw_complex *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_dft_r2c_3d(ptrdiff_t n0, ptrdiff_t n1, ptrdiff_t n2, + double *in, fftw_complex *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_dft_r2c(int rnk, const ptrdiff_t *n, + double *in, fftw_complex *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_dft_c2r_2d(ptrdiff_t n0, ptrdiff_t n1, + fftw_complex *in, double *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_dft_c2r_2d(ptrdiff_t n0, ptrdiff_t n1, + fftw_complex *in, double *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_dft_c2r_3d(ptrdiff_t n0, ptrdiff_t n1, ptrdiff_t n2, + fftw_complex *in, double *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_dft_c2r(int rnk, const ptrdiff_t *n, + fftw_complex *in, double *out, + MPI_Comm comm, unsigned flags); +</pre> + <p>Similar to the serial interface (see <a href="Real_002ddata-DFTs.html#Real_002ddata-DFTs">Real-data DFTs</a>), these +transform logically n<sub>0</sub> × n<sub>1</sub> × n<sub>2</sub> × … × n<sub>d-1</sub> real data to/from n<sub>0</sub> × n<sub>1</sub> × n<sub>2</sub> × … × (n<sub>d-1</sub>/2 + 1) complex +data, representing the non-redundant half of the conjugate-symmetry +output of a real-input DFT (see <a href="Multi_002ddimensional-Transforms.html#Multi_002ddimensional-Transforms">Multi-dimensional Transforms</a>). +However, the real array must be stored within a padded n<sub>0</sub> × n<sub>1</sub> × n<sub>2</sub> × … × [2 (n<sub>d-1</sub>/2 + 1)] + + <p>array (much like the in-place serial r2c transforms, but here for +out-of-place transforms as well). Currently, only multi-dimensional +(<code>rnk > 1</code>) r2c/c2r transforms are supported (requesting a plan +for <code>rnk = 1</code> will yield <code>NULL</code>). As explained above +(see <a href="Multi_002ddimensional-MPI-DFTs-of-Real-Data.html#Multi_002ddimensional-MPI-DFTs-of-Real-Data">Multi-dimensional MPI DFTs of Real Data</a>), the data +distribution of both the real and complex arrays is given by the +‘<samp><span class="samp">local_size</span></samp>’ function called for the dimensions of the +<em>complex</em> array. Similar to the other planning functions, the +input and output arrays are overwritten when the plan is created +except in <code>FFTW_ESTIMATE</code> mode. + + <p>As for the complex DFTs above, there is an advance interface that +allows you to manually specify block sizes and to transform contiguous +<code>howmany</code>-tuples of real/complex numbers: + + <p><a name="index-fftw_005fmpi_005fplan_005fmany_005fdft_005fr2c-484"></a><a name="index-fftw_005fmpi_005fplan_005fmany_005fdft_005fc2r-485"></a> +<pre class="example"> fftw_plan fftw_mpi_plan_many_dft_r2c + (int rnk, const ptrdiff_t *n, ptrdiff_t howmany, + ptrdiff_t iblock, ptrdiff_t oblock, + double *in, fftw_complex *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_many_dft_c2r + (int rnk, const ptrdiff_t *n, ptrdiff_t howmany, + ptrdiff_t iblock, ptrdiff_t oblock, + fftw_complex *in, double *out, + MPI_Comm comm, unsigned flags); +</pre> + <h5 class="subsubheading">MPI r2r transforms</h5> + +<p><a name="index-r2r-486"></a>There are corresponding plan-creation routines for r2r +transforms (see <a href="More-DFTs-of-Real-Data.html#More-DFTs-of-Real-Data">More DFTs of Real Data</a>), currently supporting +multidimensional (<code>rnk > 1</code>) transforms only (<code>rnk = 1</code> will +yield a <code>NULL</code> plan): + +<pre class="example"> fftw_plan fftw_mpi_plan_r2r_2d(ptrdiff_t n0, ptrdiff_t n1, + double *in, double *out, + MPI_Comm comm, + fftw_r2r_kind kind0, fftw_r2r_kind kind1, + unsigned flags); + fftw_plan fftw_mpi_plan_r2r_3d(ptrdiff_t n0, ptrdiff_t n1, ptrdiff_t n2, + double *in, double *out, + MPI_Comm comm, + fftw_r2r_kind kind0, fftw_r2r_kind kind1, fftw_r2r_kind kind2, + unsigned flags); + fftw_plan fftw_mpi_plan_r2r(int rnk, const ptrdiff_t *n, + double *in, double *out, + MPI_Comm comm, const fftw_r2r_kind *kind, + unsigned flags); + fftw_plan fftw_mpi_plan_many_r2r(int rnk, const ptrdiff_t *n, + ptrdiff_t iblock, ptrdiff_t oblock, + double *in, double *out, + MPI_Comm comm, const fftw_r2r_kind *kind, + unsigned flags); +</pre> + <p>The parameters are much the same as for the complex DFTs above, except +that the arrays are of real numbers (and hence the outputs of the +‘<samp><span class="samp">local_size</span></samp>’ data-distribution functions should be interpreted as +counts of real rather than complex numbers). Also, the <code>kind</code> +parameters specify the r2r kinds along each dimension as for the +serial interface (see <a href="Real_002dto_002dReal-Transform-Kinds.html#Real_002dto_002dReal-Transform-Kinds">Real-to-Real Transform Kinds</a>). See <a href="Other-Multi_002ddimensional-Real_002ddata-MPI-Transforms.html#Other-Multi_002ddimensional-Real_002ddata-MPI-Transforms">Other Multi-dimensional Real-data MPI Transforms</a>. + +<h5 class="subsubheading">MPI transposition</h5> + +<p><a name="index-transpose-487"></a> +FFTW also provides routines to plan a transpose of a distributed +<code>n0</code> by <code>n1</code> array of real numbers, or an array of +<code>howmany</code>-tuples of real numbers with specified block sizes +(see <a href="FFTW-MPI-Transposes.html#FFTW-MPI-Transposes">FFTW MPI Transposes</a>): + + <p><a name="index-fftw_005fmpi_005fplan_005ftranspose-488"></a><a name="index-fftw_005fmpi_005fplan_005fmany_005ftranspose-489"></a> +<pre class="example"> fftw_plan fftw_mpi_plan_transpose(ptrdiff_t n0, ptrdiff_t n1, + double *in, double *out, + MPI_Comm comm, unsigned flags); + fftw_plan fftw_mpi_plan_many_transpose + (ptrdiff_t n0, ptrdiff_t n1, ptrdiff_t howmany, + ptrdiff_t block0, ptrdiff_t block1, + double *in, double *out, MPI_Comm comm, unsigned flags); +</pre> + <p><a name="index-new_002darray-execution-490"></a><a name="index-fftw_005fmpi_005fexecute_005fr2r-491"></a>These plans are used with the <code>fftw_mpi_execute_r2r</code> new-array +execute function (see <a href="Using-MPI-Plans.html#Using-MPI-Plans">Using MPI Plans</a>), since they count as (rank +zero) r2r plans from FFTW's perspective. + + </body></html> +