AMBER Archive (2004)

Subject: Re: AMBER: recovering total force array

From: Dave S Walker (dswalker_at_darkwing.uoregon.edu)
Date: Tue Oct 19 2004 - 14:04:16 CDT

Hi all,

I had spent some time a few months ago trying to recover the total force
array from sander while running in parallel, and never succeeded. I've
recently taken another crack at doing this, so I hope to make my
predicament a little clearer:

In runmd.f, I had been trying to replace the velocity array with the force
array when making the call to CORPAC(). The forces (and perhaps the
velocities and coordinates) in runmd.f I suspect are distributed among the
nodes and I had been trying to use a reduce statement to recover all the
forces before they're written over in Step 3. Doing this has failed to
reproduce the array I get with 1 cpu. Comparing the information contained
in 2 cpus separately, before reducing, told me the operation works fine.
I also noticed that reduce statements are not used in the coordinate and
velocity archives to retrieve those parameters, which served as a hint
that I was on the wrong track.

Being able to do this in runmd.f seems like a convenient place because I
could use the input file parameter 'ntwv' to govern when it is stored, but
I've been unable to get it to work there. Is there another subroutine
that works as conveniently or would another mpi command do the trick? The
code is pretty extensive, so any feedback or suggestions would be greatly
appreciated. Thanks again for your patience.

Dave
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu