https://snicdocs.nsc.liu.se/w/index.php?title=Parallel_FFTs_in_Molsim&feed=atom&action=historyParallel FFTs in Molsim - Revision history2024-03-28T09:38:16ZRevision history for this page on the wikiMediaWiki 1.31.10https://snicdocs.nsc.liu.se/w/index.php?title=Parallel_FFTs_in_Molsim&diff=4442&oldid=prevJoachim Hein (LUNARC): Created page with "{{project info |description=Performance improvement of the SPME in Molsim |fields=Computational chemistry |financing=SNIC |active=yes |start date=2012-04 |end date= }} == Overvi..."2012-09-14T16:38:12Z<p>Created page with "{{project info |description=Performance improvement of the SPME in Molsim |fields=Computational chemistry |financing=SNIC |active=yes |start date=2012-04 |end date= }} == Overvi..."</p>
<p><b>New page</b></p><div>{{project info<br />
|description=Performance improvement of the SPME in Molsim<br />
|fields=Computational chemistry<br />
|financing=SNIC<br />
|active=yes<br />
|start date=2012-04<br />
|end date=<br />
}}<br />
<br />
== Overview ==<br />
[[Molsim]] deploys an SMPE (smooth particle mesh Ewald) algorithm to evaluate the long range interactions. When deploying a larger number of processors the original implementation of the FFT proved to be a performance bottle neck. The project takes a two stage approach to the problem:<br />
* Replacing the originally deployed FFT library (compiled [[Fortran]]-source) with the high performance library [[FFTW]]<br />
* Upgrading the original serial transformation with an parallel transformation using [[MPI]]<br />
<br />
The replacement of the FFT library with FFTW has been accomplished in July 2012. When comparing the performance to the originally deployed FFT library, FFTW turned out to be 4.5 times faster.<br />
<br />
The parallisation of the Fourier transformation routines is ongoing.<br />
<br />
== Members ==<br />
{{#ask: [[Category:Person]] [[project::{{PAGENAME}}]]<br />
|?centre<br />
|?role<br />
|?field<br />
}}</div>Joachim Hein (LUNARC)