Re: [eigen] Bugs on bugzilla

[Matthew Brett <matthew.brett@xxxxxxxxx>]

Hi,

On Fri, Jun 6, 2014 at 4:29 PM, Christoph Hertzberg
<chtz@xxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> On 04.06.2014 19:34, Matthew Brett wrote:
>>
>>  From the far distance - maybe it's worth considering trying to make
>> Eigen a default option for open-source optimized blas / lapack
>> libraries?  The whole world of numerical open-source packages is
>> really suffering at the moment for lack of reliable and easy-to-build
>> options on Windows [1], and on OSX [2].  If you think Eigen might be
>> competitive as a general blas / lapack library (cf ATLAS, MKL etc),
>> and it was possible to build on Windows, then my guess is you'd have a
>> lot of interest and support from us numerical packages people...
>
>
> I'm not the expert on this, but I think Eigen's BLAS implementation is
> almost feature complete (very few Fortran functions left, which could easily
> be implemented in C/C++ if someone cared). And we are definitely at least
> competitive with other BLAS/LAPACKs
> (http://eigen.tuxfamily.org/index.php?title=Benchmark -- these require an
> update, but cf recent comparisons with Blaze on the mailing list and in
> bugzilla)
>
> For LAPACK I think we only have very few things implemented natively (I only
> found cholesky, lu and eigenvalues)
>
> The relevant bugzilla entries are:
> http://eigen.tuxfamily.org/bz/show_bug.cgi?id=61
> http://eigen.tuxfamily.org/bz/show_bug.cgi?id=62
>
>
> Assuming we finished those, what would be the next steps to become a
> "default option"? I.e., adaptions to our build-system, how/where to publish
> and announce, ...

I think the only needful thing after that would be a cblas interface
to the Fortran blas routines:

http://forum.kde.org/viewtopic.php?t=120377

I am no expert either, but there is a cblas interface linked in the
thread above, which might be useful.

I am a Python / Numpy person, but I am sure you could get strong
interest from the R / Octave / Julia guys and gals too.

Cheers,

Matthew