[gmx-developers] available contribution: loop over pull code
hess at sbc.su.se
hess at sbc.su.se
Wed Sep 8 22:36:52 CEST 2010
I was already thinking about being able to specify multiple pull pairs.
You had asked about that before and it would indeed be a useful option
and not too complicated too implement.
Now you ask asking for several more things.
I don't know if different pull types would be of general use and that
would complicate the code and make it more bug-prone, especially for
checking what options do and don't work together. We've already had
several bugs in 4.0 in combinations of pull options.
Communication cost will be an issue at high parallelization.
For you this might (usually) not be an issue, but in Gromacs we would like
to keep everything compatible and fast, also at high parallelization.
Ideally users should be able to write plugins for things like complex
pulling. We will start working on modularization and making the code more
object oriented now.
Berk
> Alright. We do not have such code, so I guess that answers it in the
> negative. Thank you for providing clarity on this issue. Your
> preferred method makes it difficult to apply different pull-types or
> dimensions to different reference groups, but I understand that you
> would like to keep the code simple. As for speed, I tested that and
> the loss is negligible.
>
> Quoting hess at sbc.su.se:
>
>> Hi,
>>
>> I would not like a loop over the pull code.
>> That complicates things a lot and increases the communication cost.
>> I would simply want to add something like a reference group number
>> to each pull group.
>>
>> Berk
>>
>>> Dear Gromacs Developers:
>>>
>>> A colleague of mine, Pawel Pomorski working at Sharcnet Canada, has
>>> modified gromacs version 4.0.5 to put a loop around the pull code.
>>> This has made it possible to have multiple reference groups (each
>>> reference group will multiple pulled groups). I have found this very
>>> useful for my work and have thoroughly tested the code modifications.
>>>
>>> We would like to get this code added to the main gromacs code for the
>>> next release. I assume that Pawel would need to update the code
>>> modifications to fit into the git master.
>>>
>>> Before Pawel does this work, we would like to get some assurance that
>>> the modification will be accepted (or not accepted) into the main
>>> gromacs code. I already have a number of test cases that I used to
>>> validate the modified 4.0.5 code and can run these on the new code as
>>> well.
>>>
>>> The purpose of this email is to ensure that Pawel doesn't end up doing
>>> any work for naught. I can always continue to use the 4.0.5 version
>>> that I have, but I think that this pull code modification will be
>>> useful for many users.
>>>
>>> In gromacs version <3.3.3, a loop over the pull code was easily
>>> implemented in an hour or so. The situation was quite a bit more
>>> complex in version 4.
>>>
>>> Finally, Pawel has many responsibilities and may not be able to port
>>> his modifications immediately. If he does, is it important that he
>>> download the git master one night and submit his new code the next
>>> day? the next week? the next month?
>>>
>>> Thank you,
>>> Chris.
>>>
>>>
>>>
>>> --
>>> gmx-developers mailing list
>>> gmx-developers at gromacs.org
>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>> Please don't post (un)subscribe requests to the list. Use the
>>> www interface or send it to gmx-developers-request at gromacs.org.
>>>
>>
>> --
>> gmx-developers mailing list
>> gmx-developers at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>> Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-developers-request at gromacs.org.
>>
>
>
>
> --
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-developers-request at gromacs.org.
>
More information about the gromacs.org_gmx-developers
mailing list