Difference between revisions of "FFTW"

From SNIC Documentation
Jump to: navigation, search
 
(3 intermediate revisions by the same user not shown)
Line 2: Line 2:
 
|description=Freely available high performance library to perform fast Fourier transformations
 
|description=Freely available high performance library to perform fast Fourier transformations
 
|license=free
 
|license=free
|research areas=Computational science
+
|fields=Computational science; Performance library
 
}}
 
}}
 +
[http://www.fftw.org/ {{PAGENAME}}] is a {{#show: {{PAGENAME}} |?description}}.
  
FFTW (Fastest Fourier Transform in the West) is a high performance library to calculate discrete Fourier transformations.  The library is freely available and typically easy to install.
 
  
The software is self optimising, trying several algorithms at run time and choosing the best performing oneTo achieve its high  computational speed latest version supports the [[SSE]] and [[AVX]] instructions of the latest x86-processors.
+
== Overview ==
 +
FFTW (Fastest Fourier Transform in the West) is a high performance library to calculate discrete Fourier transformationsThe library is freely available and typically easy to install, if it is not done by the HPC centre you are using.
  
The library contains parallel versions using either of [[pthreads]], [[OpenMP]] or [[MPI]].  Besides the current version 3.x, there is the older version 2.x, which have different application interfaces.  It is not recommended to use version 2.x for new developments, however it is still widely used in legacy code.
+
The software is self optimising, trying several algorithms at run time and choosing the best performing one.  To achieve its high  computational speed latest version supports the [[SSE]] and [[AVX]] instructions of the latest x86-processors.  The library provides a native [[C]]-API.  In [[Fortran]]-2003 this is easily accessed using the C-bindings which are part of the language standard.  For older versions of Fortran the library also contains a Fortran interface.  To avoid data alignment issues, which could inhibit the performance, using the Fortran-2003 bindings to the C-API is encouraged.
 +
 
 +
The library contains parallel versions using either of [[pthreads]], [[OpenMP]] or [[MPI]].  Besides the current version 3.x, there is the older version 2.x, which has a different application interfaces.  It is not recommended to use version 2.x for new developments, however it is still widely used in legacy code.
  
 
== License ==
 
== License ==
 
{{show license}}
 
{{show license}}
 +
 +
GNU GPL version 2
  
 
== Availability ==
 
== Availability ==

Latest revision as of 16:59, 14 September 2012

FFTW is a Freely available high performance library to perform fast Fourier transformations.


Overview

FFTW (Fastest Fourier Transform in the West) is a high performance library to calculate discrete Fourier transformations. The library is freely available and typically easy to install, if it is not done by the HPC centre you are using.

The software is self optimising, trying several algorithms at run time and choosing the best performing one. To achieve its high computational speed latest version supports the SSE and AVX instructions of the latest x86-processors. The library provides a native C-API. In Fortran-2003 this is easily accessed using the C-bindings which are part of the language standard. For older versions of Fortran the library also contains a Fortran interface. To avoid data alignment issues, which could inhibit the performance, using the Fortran-2003 bindings to the C-API is encouraged.

The library contains parallel versions using either of pthreads, OpenMP or MPI. Besides the current version 3.x, there is the older version 2.x, which has a different application interfaces. It is not recommended to use version 2.x for new developments, however it is still widely used in legacy code.

License

License: Free.

GNU GPL version 2

Availability

ResourceCentreDescription
AbiskoHPC2Ncapability resource of 153 TFLOPS with full bisectional infiniband interconnect
AlarikLUNARCthroughput cluster resource of 40 TFLOPS
AuroraLUNARCthroughput/general purpose cluster resource
PlatonLUNARCthroughput cluster resource of 26 TFLOPS

Links