[gmx-developers] Question for those contributing matrix / array operations and types

[David van der Spoel]

Den 2019-07-25 kl. 16:45, skrev Eric Irrgang:
> Hi Devs,
> 
> I have thought for a while that a lot of GROMACS matrix and vectorized 
> data/math would be simpler, faster, better documented, and more portable 
> if it were based on Eigen http://eigen.tuxfamily.org/index.php but I 
> can't find a reference to it in the last [six 
> years](https://redmine.gromacs.org/issues/1017#note-10) and various 
> contributors on this list have been spending effort on related issues.
> 
> I don't know whether Eigen has been ruled out or whether it just doesn't 
> meet the needs of the developers. Could you please share your thoughts, 
> experiences, or reasons for not using Eigen?
I agree that a complete modern library is to be preferred over having a 
subset of the arkane blas library, even though blas likely is somewhat 
faster. In addition, for some of my gromacs related (but not core 
gromacs) stuff I need a singukar value decomposition algorithm that is 
not available in the subset, which means that I still am dependent on 
external blas libraries. Other open source chemistry software such as 
openbabel uses eigen as well.

There may be technical drawbacks that I am not aware of obviously.
> 
> I am asking from the two perspectives of (1) no-copy data interchange 
> across the GROMACS API boundary, and (2) simplifying and normalizing 
> arithmetic code for GROMACS and GROMACS-adjacent software (i.e. if I 
> want to help people write code that interoperates cleanly with Eigen and 
> GROMACS, what level of integration is appropriate?).
> 
> Relevant Redmine issues:
> https://redmine.gromacs.org/issues/1017
> https://redmine.gromacs.org/issues/2035
> https://redmine.gromacs.org/issues/2281
> https://redmine.gromacs.org/issues/2834
> https://redmine.gromacs.org/issues/2976
> https://redmine.gromacs.org/issues/2993
> 
> Also relevant: https://www.mozilla.org/en-US/MPL/2.0/FAQ/
> 
> Thank you for your perspective,
> Eric
> 


-- 
David van der Spoel, Ph.D., Professor of Biology
Head of Department, Cell & Molecular Biology, Uppsala University.
Box 596, SE-75124 Uppsala, Sweden. Phone: +46184714205.
http://www.icm.uu.se