AMBER Archive (2009)

Subject: Re: [AMBER] Problem build pmemd

From: Robert Duke (rduke_at_email.unc.edu)
Date: Wed Mar 18 2009 - 08:53:17 CDT

Hi Giorgos,
I put some data structs in pmemd in automatic stack storage; this tends to
be a bit of a hassle in places where there are low hard stack limits, but
there are performance benefits and the memory use is actually not huge, so I
am reluctant to change this. I did put in code to bump up the soft limit a
while back, but if the hard limit is low, that has to be fixed by an admin.
Okay, so how much? Depends on 1) how many atoms, 2) how many processors
(more means less stack per processor), 3) the version of pmemd. I don't
have a firm algorithm. I would try 500MB; that really should be enough
unless you have some really strange setup conditions (I am not looking at
the code, but I would think really high fft grid density could cause grief,
or really large interpolation order in recip space; other than that it is
mostly a function of atom count; I actually don't understand right off why
you are having problems with 134MB, but I would have to see your problem to
have a clue.
Regards - Bob Duke

----- Original Message -----
From: "Giorgos Lamprinidis" <lamprinidis_at_pharm.uoa.gr>
To: "AMBER Mailing List" <amber_at_ambermd.org>
Sent: Wednesday, March 18, 2009 9:28 AM
Subject: Re: [AMBER] Problem build pmemd

> Dear members,
>
> I am working with Mr Georgiadis from the University of Athens, Greece.
>
> I am trying to run the MM-PBSA tutorial from
> http://ambermd.org/tutorials/advanced/tutorial3/
> On my linux machine (P4) it works fine.So now we are trying to run it on
> our supercomputer which is an HP.
> I run the pmemd Georgiadis just compiled but i get this error when i run
> the heat.in procedure:
>
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> STOP PMEMD Terminated Abnormally!
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> STOP PMEMD Terminated Abnormally!
> STOP PMEMD Terminated Abnormally!
> STOP PMEMD Terminated Abnormally!
> | WARNING: Stack usage limited by a hard resource limit of 134217728
> bytes!
> | If segment violations occur, get your sysadmin to increase the
> limit.
> MPI Application rank 0 exited before MPI_Init() with status 0
> STOP PMEMD Terminated Abnormally!
> Failed sending message: (Unable to connect to socket (Connection
> refused)).
>
> the error comes 8 times since i am trying to run on 8 CPUs
> My question is, how much i must increase the stack limit?
> when i run the tutorial on Linux i see that the memory used by sander is
> about 150MB.
>
> Thx for your help guys,.....
>
> Dr. George Lamprinidis
>
>
>
>
>
>
>> It worked
>> Thank you very much for your response and your detailed explanation !
>>
>> Yiannis Georgiadis
>>
>> Robert Duke wrote:
>>> This is a simple matter of name mangling conventions between c and
>>> fortran, and you have some combination of c and fortran compilers that
>>> are not agreeing with each other. To figure out what you need to do,
>>> try the following in the pmemd directory after the compilation has been
>>> done (been linking has not succeeded):
>>>
>>> nm pmemd.o | grep unlimit_stack
>>> nm pmemd_clib.o | grep unlimit_stack
>>>
>>> Look at the outputs - they probably differ. You need to get the c
>>> compiler to be producing stuff from pmemd_clib.c that looks like what
>>> fortran is expecting (which is what you see in pmemd.o, related to a
>>> call from fortran to the c routine).
>>>
>>> So at the top of the pmemd_clib.c source file there is a manifest
>>> constants hack that handles this. Depending on which of the defined
>>> constants is defined by CFLAGS in config.h, the names of the pmemd c
>>> library routines will be mangled differently. The current choices for
>>> the defined constants are CLINK_CAPS, NO_C_UNDERSCORE, DBL_C_UNDERSCORE,
>>> and nothing.
>>> So do this in config.h by including the manifest constant in the
>>> config.h file - for example, you might use:
>>> CFLAGS = -DNO_C_UNDERSCORE
>>>
>>> For some compilers it is possible to change the name mangling expected
>>> by the fortran compiler by including a fortran compiler flag instead.
>>> Determining this requires reading the compiler manuals. I don't have a
>>> clue what compilers you have on this platform, but this covers all
>>> possible avenues. Presuming that the mpi routines are linking properly,
>>> you want to approach this problem by using the CFLAGS defined constant
>>> (if mpi is not linking either, then you have to get fortran to
>>> recognize the mpi names or recompile the mpi libraries in such a way
>>> that the exported mpi_* names are recognized).
>>>
>>> Regards - Bob Duke
>>>
>>> ----- Original Message ----- From: "Yiannis Georgiadis"
>>> <giannis_at_cc.uoa.gr>
>>> To: "AMBER Mailing List" <amber_at_ambermd.org>
>>> Sent: Friday, March 13, 2009 8:14 PM
>>> Subject: [AMBER] Problem build pmemd
>>>
>>>
>>>>
>>>> Has anyone managed to build pmemd in hpux 11i platform ?
>>>> I am using hpf90 and gcc and I am getting at linking :
>>>>
>>>> /usr/ccs/bin/ld : Unsatisfied symbols:
>>>>
>>>> get_wall_time
>>>> unlimit_stack
>>>>
>>>> Yiannis Georgiadis
>>>> UoA Computer Center Sys Admin
>>>>
>>>>
>>>> _______________________________________________
>>>> AMBER mailing list
>>>> AMBER_at_ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>
>>>
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER_at_ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>> ------------------------------------------------------------------------
>>>
>>>
>>> No virus found in this incoming message.
>>> Checked by AVG - www.avg.com Version: 8.0.237 / Virus Database:
>>> 270.11.12/1998 - Release Date: 03/12/09 18:23:00
>>>
>>>
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER_at_ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>
> _______________________________________________
> AMBER mailing list
> AMBER_at_ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber