So, I changed -axSTPW and -xP options in config.h to -axW and -xW,
repectively. And then pmemd.MPI works well. Apparently, the AMD
Opteron 6172 Processor does not support SSE beyond SSE2.
Thanks,
Yubo
On Tue, Jul 3, 2012 at 7:02 AM, Jason Swails <jason.swails.gmail.com> wrote:
> For future reference, this is caused by the "-xHost" flag that is scattered
> throughout the config.h. If you remove "-xHost" from config.h and
> recompile, this problem should go away.
>
> This problem is caused because -xHost applies all vectorization
> optimizations that the 'current' processor supports. (That is, whatever
> machine you are compiling on). This is a problem for heterogeneous
> clusters that have nodes with different CPUs present. Whenever you run a
> program on a node that has a CPU that does not support all of the
> instructions that the machine you compiled Amber on supported, this error
> will occur (that's why some programs may appear to work while others do
> not).
>
> Note that removing -xHost will hurt performance (by disabling
> vectorization). If you absolutely *must* have that 10-15% performance
> boost, you'll need to look into the Intel compiler documentation and
> manually turn on each vectorization optimization supported by every
> processor in your cluster.
>
> HTH,
> Jason
>
> On Tue, Jul 3, 2012 at 7:54 AM, David A Case <case.biomaps.rutgers.edu>wrote:
>
>> On Tue, Jul 03, 2012, Yubo Fan wrote:
>> >
>> > I recently compiled Amber 11 and 12 on our new Opteron server.
>> > Everything works fine except pmemd.MPI of Amber 11.
>>
>> "Obvious" workaround: don't use pmemd.MPI from Amber11. I don't think it
>> has
>> any functionality not present in the Amber12 version.
>>
>> If you are curious, edit the config.h file and look at the PMEMD_* flags,
>> especially PMEMD_FOPTFLAGS and PMEMD_COPTFLAGS. See what is different
>> between
>> Amber11 and Amber12; consider copying the flags from Amber12 to the Amber11
>> config.h file.
>>
>
>> ....dac
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Candidate
> 352-392-4032
> _______________________________________________
> 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 Thu Jul 05 2012 - 08:30:03 PDT
