On Thu, Feb 28, 2013 at 12:41 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Excellent work Jason. :-)
>
> Can you verify that you tested existing (no EP) runs as well like the JAC
> NVE benchmark and FACTOR IX NVE benchmark. I know that a lot of the fixed
> array sizes in PMEMD Bob had tuned to get maximum cache reuse so making
> them arbitrarily bigger could hurt performance - unless they are only made
> bigger in the EP cases such that only EP performance is affected.
> A quick check on a modern CPU machine would verify if this is a concern or
> not - the last time it was probably 'tweaked' was the pentium 4 days.
>
It's a temporary array in a setup routine that is deallocated after the
setup is done, so there won't be any performance implications. I haven't
done any benchmarks (I don't have anywhere to run them easily), so if
someone wants to check this, they're welcome to.
Of course this is only done for EP runs, so this change is in a code path
that is never traversed in the existing benchmarks, so those won't be
informative, anyway.
I suppose you could say it definitely got 'slower', but since it now
produces an eventual result as opposed to an immediate segfault, it's
probably a permissible slowdown ;).
--Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Feb 28 2013 - 10:00:04 PST
