The MKL-based NumPy is around 4 times faster than the Accelerate Framework, and is faster than Atlas as well (2-3x).
It should be possible to incorporate MKL with the NumPy/SciPy in Macports. Intel allows free download of its MKL libraries for non-commercial use: http://software.intel.com/en-us/articles/non-commercial-software-download/ And it also provides instructions on how to build NumPy/SciPy with the MKL libraries: http://software.intel.com/en-us/articles/numpy-scipy-with-mkl/ Do you think MKL can be included as a variant? -Kamesh On 19 April 2012 04:56, Adam Mercer <[email protected]> wrote: > On Tue, Apr 17, 2012 at 14:17, Frank Schima > <[email protected]> wrote: > > py-numpy and py-scipy have atlas variants which use the atlas port > instead > > of the Accelerate framework. The atlas port does tune to your processor. > > Have you tried that? I have Cc'ed Adam, the maintainer of these ports, > if he > > could comment on their performance with the atlas variant? Maybe the > Intel > > MKL libraries could be added as a variant? > > Personally I don't use the atlas variant as I've found it causes a lot > of build problems. I find the Accelerate framework fast enough for my > needs, but my codes aren't that processor intensive. If someone wanted > to add an MKL variant I'd be willing to accept it providing they were > willing to support it. > > Cheers > > Adam > _______________________________________________ > macports-users mailing list > [email protected] > http://lists.macosforge.org/mailman/listinfo.cgi/macports-users >
_______________________________________________ macports-users mailing list [email protected] http://lists.macosforge.org/mailman/listinfo.cgi/macports-users
