Thank you for your suggestion. My original compilation of parallel
Amber10 used a high level of optimization (-O3 -qstrict). After
getting the wrong result for rem_hybrid I switched to a lower level
optimization (-O) and still got wrong results for this test. Now,
following your suggestion, I turned off optimization completely
(compiled with -O0) and rem_hybrid yielded correct results!
Given that with -O3 -qstrict the other tests passed (including
other REM tests), I presume that I can use this level optimization
to compile most of the code, but I am not sure which files are
generating wrong results for hybrid REM when compiled with optmization.
Any suggestions which files I need to compile with optimization
turned off?
Thank you so much again,
Shan-Ho
On Wed, 11 Feb 2009, Carlos Simmerling wrote:
> I don't really have any suggestions as I don't have access to an AIX machine
> and it seems to pass on other machines. perhaps you could try turning off
> optimization and see if that helps? if you're not planning hybrid REMD,
> maybe this isn't an issue.
>
>
>>>
>> If I turn off bounds checking, on the Linux cluster (using pgf90),
>> all serial and parallel Amber10 tests passed. However, on the AIX
>> cluster if I turn off bounds checking all serial tests pass, but
>> rem_hybrid runs until completion, and generates incorrect results.
>> The potential energy turned out to be around -1450 instead of -615.
>>
>> Any suggestions how I can troubleshoot what is going wrong in the
>> rem_hybrid test?
>>
>> Thank you so much! Shan-Ho
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Feb 13 2009 - 01:13:17 PST
