Peter -
Okay, it is not actually in the amber8/test tests, it is in the
amber8/benchmarks tests (hb, to be exact). This is part of my standard
pmemd testing, along with about 20 others, so that is how I am certain ntr=1
basically works. I am still fighting to get completely correctly executing
code for pmemd from pgi. It can probably be done; an i/o problem earlier is
probably a 32/64 bit problem (we have a suggestion to use 64 bit compiles on
the web site; on the utah cluster I am working on this seems to be a bad
idea...). It looks like some f90 features like cpu_time may be a little
loosely defined, but I will have to play with it some more. I will
definitely have sample config.h's for pmemd for pathscale by early next
week, if not sooner. I hope to know more about the pgi issues in the next
week also.
Regards - Bob
----- Original Message -----
From: "Peter Varnai" <pv232.cam.ac.uk>
To: <amber.scripps.edu>
Sent: Thursday, February 24, 2005 5:59 PM
Subject: Re: AMBER: pmemd + pgi 5.1 + ntr = 1
> Dear Bob,
>
> Thanks for the answer -
>
>> ntr=1 should work (its in the stand pmemd test suite). There are various
>
> well the test.pmemd does not seem to include input with ntr=1 option.
>
>> aspects of pgi compilers + pmemd I am less certain about. I have pgi 5.2
>> basically working on cray opteron boxes, but have not tested everything
>> yet,
>> and there are a lot of cray mods involved. I was actually trying to get
>> 5.2
>> running today on a linux opteron cluster, and it was a total nightmare in
>> regard to being sure that pgi is getting the right libraries.
>> Ultimately, I
>
> Well, I compiled both sander and pmemd without any problem, all tests
> passed.
>
>> this more, but in the meantime I might suggest trying a pathscale
>> evaluation
>> copy (www.pathscale.com). It is known to produce faster code. I will
>> have a
>
> would you have a config.h for pathscale?
>
> Thanks,
> Peter
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Thu Feb 24 2005 - 23:53:00 PST
