Good morning everyone. It seems that my first merge of a branch didn't go
terribly well, sorry about that again.
Please no-one else invest any more time in this than you have to in order
to get your own thing working. If you have a problem or concern with the
code that I've introduced then let me know and I will fix it for you.
Given the circumstances working on a Saturday is not a big problem for me,
although I have to sleep at some point.
>>Also, there appear to have been a large number of changes from this in the
>>pmemd branch without it being clear if protections were put in place for
>>the GPU - is EMIL supported in the GPU code? - there doesn't appear to be
>>a check to quit if not.
Yes, it works with cuda but there is an increase in code size and a
smaller (proportionally) but still measurable one in resident memory
footprint, I was going to show you the figures at the meeting to discover
if that matters. If so then its no problem to throw a warning when the two
are configured together.
>>Plus I am assuming someone tested the serial and parallel performance to
>>make sure it was not impacted?
Yes, execution speed is unaffected (unless you actually set emil_do_calc=1,
of course). Inside the main loop of mdrun the only extra code is inside a
couple of if-thens which the branch prediction probably gets right anyway,
so its not surprising that there is no penalty. Graphs at the meeting.
>>I have changed the emil build scripts. I added a custom Makefile, and it
>>conforms now to the rest of Amber (except for MTK++ and the 3rd-party
tools
>>that still need to be configured).
Thanks for that. Its saturday morning now where I am, I was expecting to
have to do it myself. With the use of autotools.... I really thought that
was what I was supposed to do, and it seemed to be working. I suppose I
should have asked.
Josh
On 11 January 2014 05:27, Jason Swails <jason.swails.gmail.com> wrote:
> On Fri, Jan 10, 2014 at 9:20 PM, David A Case <case.biomaps.rutgers.edu
> >wrote:
>
> > On Fri, Jan 10, 2014, Ross Walker wrote:
> > >
> > > 1) What is EMIL?
> > > 2) Was it tested before committing?
> >
> > Can you post instructions on how to get a branch tested by CruiseControl?
> > The emil code required automake (should not have); it worked on a number
> > of machines, all of which happened to have useful versions of autotools,
> > but having the ability to run Cruise Control on a branch (as you or
> > someone has been doing with gpu-tachyon) would probably help catch such
> > problems.
>
>
> I have changed the emil build scripts. I added a custom Makefile, and it
> conforms now to the rest of Amber (except for MTK++ and the 3rd-party tools
> that still need to be configured).
>
> It appears to work, but only builds the static library. Based on the
> EMILLIB variable set in the config.h file by configure2, though, that
> appears to be the one used in the sander and pmemd builds.
>
> I'm still working with this a little bit, but it appears to be a bit more
> stable now.
>
> All the best,
> Jason
>
> --
> Jason M. Swails
> BioMaPS,
> Rutgers University
> Postdoctoral Researcher
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat Jan 11 2014 - 00:00:03 PST
