An Analysis of FFTW and FFTE Performance
One of the most frequently used algorithms in engineering and scientific applications is Fast Fourier Transform (FFT). Its open source implementation (Fastest Fourier Transform of the West, FFTW) is widely used, mainly due to its excellent performance, comparable to the vendor-supplied libraries. On the other hand, even if not yet in a fully production state, FFTE (Fastest Fourier Transform of the East) keeps up with FFTW, and outperforms it for very large transform sizes. Here we present results of the performance and scalability tests of FFTW and FFTE libraries. Comparison is done using different compilers and parallelization approaches on CURIE and JUGENE supercomputers.
KeywordsFFT MPI OpenMP Hybrid parallelism
Unable to display preview. Download preview PDF.
- 1.Cooley–Tukey FFT algorithm, http://en.wikipedia.org/wiki/Cooley-Tukey_FFT_algorithm
- 2.CURIE Supercomputer, http://www-hpc.cea.fr/en/complexe/tgcc-curie.htm
- 3.JUGENE Supercomputer, http://www.fz-juelich.de/jsc/jugene
- 4.PRACE Home Page, http://www.prace-ri.eu/
- 5.FFTW Home Page, http://www.fftw.org/
- 6.FFTE: A Fast Fourier Transform Package, http://www.ffte.jp/