VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Fri May 25 2012 - 14:34:03 CDT

Axel,
  Indeed, it may make sense to make some changes to this and other
measure commands that similarly require some values that can only
be computed by traversing the entire trajectory. Another goal I have
that's somewhat related is to make changes to the design of the
measure command APIs to better support GPU acceleration. I have
running prototype GPU-accelerated implementations of all of the VMD measure
commands, but as you might guess, the peak speedups for all of the
simple ones are deflated somewhat by the host-GPU memory transfer
times, it is essentially the same issue as in the case of doing
out-of-core I/O on the host -- you want to minimize the number of
copies and the number of passes over a trajectory, so in a sense what
I'm after is an API that allows the back-end code to stream the trajectory
through the CPU and/or GPUs, doing as many calculations at once using
the memory resident frame(s) as possible. I would prefer that the user
wouldn't have to do things like pass in the results from previous
calculations for themself, but avoiding this requires a much more
sophisticated "batching" version of the measure command APIs. If you
or others have thoughts about this, I'd love to hear them. Ideally
we'd find a good solution that works well for out-of-core processing
and that supports both traditional multi-core CPU-based processing
as well as schemes that will work for GPUs and other accelerators
that have their own separate memory systems.

Cheers,
  John Stone
  vmd_at_ks.uiuc.edu

On Tue, May 22, 2012 at 07:53:22PM -0400, Axel Kohlmeyer wrote:
> On Tue, May 22, 2012 at 7:02 PM, John Stone <johns_at_ks.uiuc.edu> wrote:
> > Hi,
> >  You are correct that the existing implementation requires all of
> > the frames to be loaded.  There isn't any way to work around this except
> > to write your own analysis script that works one frame at a time.  One
>
> john,
>
> how about adding an option to feed a set of
> averaged positions to the measure rmsf command?
>
> then this could be done in a two pass bigdcd procedure.
> and for file formats that your out-of-core stuff won't
> support (initially or ever).
>
> just an idea...
> axel.
>
> > of the features I am working on for a future version of VMD is out-of-core
> > trajectory access, which would allow you to do this and other analyses
> > in a manner similar to the way BigDCD works, but it will support all of
> > the features that need all of the referenced frames to be loaded as well.
> >
> > Cheers,
> >  John Stone
> >  vmd_at_ks.uiuc.edu
> >
> > On Mon, May 21, 2012 at 10:39:45AM -0400, Raul Araya wrote:
> >>    Dear VMD users
> >>
> >>    Is there a way to use the measure rmsf command within the framework of
> >>    bigdcd?
> >>    I ask this because I know that the rmsf command will use avpos to
> >>    calculate the average position of each atom...to do this I imagine that
> >>    vmd requires to whole or at least the desired part of the trajectory to be
> >>    loaded in the memory in order to be able to compute the average position.
> >>    Thus, I think this will not work using bigdcd unless one includes in the
> >>    measuring script some way to store the average position. How can this be
> >>    accomplished?
> >>
> >>    Thanks...
> >>
> >>    Raul Araya Secchi
> >>    B.Sc Molecular Biotechnology.
> >>    Molecular Biotechnology Engineer.
> >>    PhD Student (Biotechnology Program. UNAB, Chile)
> >>    Computational Biology Lab (DLab)
> >>    Center for Mathematical Modeling (CMM)
> >>    Facultad de Ciencias Fisicas y Matematicas.
> >>    Universidad de Chile.
> >
> > --
> > NIH Resource for Macromolecular Modeling and Bioinformatics
> > Beckman Institute for Advanced Science and Technology
> > University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
> > http://www.ks.uiuc.edu/~johns/           Phone: 217-244-3349
> > http://www.ks.uiuc.edu/Research/vmd/       Fax: 217-244-6078
>
>
>
> --
> Dr. Axel Kohlmeyer
> akohlmey_at_gmail.com  http://goo.gl/1wk0
>
> College of Science and Technology
> Temple University, Philadelphia PA, USA. 

-- 
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
http://www.ks.uiuc.edu/~johns/           Phone: 217-244-3349
http://www.ks.uiuc.edu/Research/vmd/       Fax: 217-244-6078