Hi Ludo, Thank you for your precision.
On Fri, 15 Feb 2019 at 18:09, Ludovic Courtès <ludovic.cour...@inria.fr> wrote: > > zimoun <zimon.touto...@gmail.com> skribis: > > > Well, is it still accurate ? > > Is Guix binary still slower than other ? > > The comment links to > <http://fftw.org/fftw3_doc/Installation-on-Unix.html>, which reads: > > Enable various SIMD instruction sets. […] FFTW will try to detect at > runtime whether the CPU supports these extensions. That is, you can > compile with --enable-avx and the code will still run on a CPU without > AVX support. If I understand well, the claims p.45 of the EasyBuild presentation [1] should not true any more. Because there is a factor 2+ with their benchmarks. [1] https://users.ugent.be/~kehoste/eum18/eum18_easybuild_past_present_future_20180130.pdf > So I believe the numbers one gets with the ‘fftw’ package in Guix are > the best one can get because FFTW does the right thing of using the > right version of its hot functions at run time. (That said, you can > redo the benchs, and if this is not the case, it’s a bug! :-)) A benchmark from "our side" should be informative. :-) But I am not sure that I would like to dive in EasyBuild stuff... ;-) Thank you again ! All the best, simon
