AMBER Archive (2009)

Subject: Re: [AMBER] Error with calculation of the potential energy

From: Robert Duke (rduke_at_email.unc.edu)
Date: Mon Apr 27 2009 - 14:26:06 CDT


Okay, I have to assume you have modified the makefiles and heaven knows what
else as you have included new manifest constants in the config.h? I really
can't be much help, but I would strongly suggest, first of all, 1) don't
touch the source or makefiles, 2) start conservative and build the
uniprocessor version, 3) start conservative with F90_OPT_DFLT set to
$(F90_OPT_LO). I would then bump up the optimization until I got into
trouble. If I really wanted to run this stuff, I would read up on the
performance manifest constants in the pmemd README and try using some for a
chip most like the sparc (or what I actually do when I have access to a
machine is build 12 different ways for uni and then for mpi). I think other
folks do have some experience with sun on the list, so they may be able to
help. I don't think sun machines are particularly impressive in these
applications; someone please feel free to correct me if I am wrong.
Regards - Bob

----- Original Message -----
From: "Per Jr. Greisen" <pgreisen_at_gmail.com>
To: "AMBER Mailing List" <amber_at_ambermd.org>
Sent: Monday, April 27, 2009 2:57 PM
Subject: Re: [AMBER] Error with calculation of the potential energy

Hi,

The system is

SunOS newton 5.10 Generic_137111-01 sun4u sparc SUNW,Sun-Fire-15000

and I am using the fortran compiler from sun. I have written the
following config.h file for sunOS

MATH_DEFINES =
MATH_LIBS =
FFT_DEFINES = -DPUBFFT
FFT_INCLUDE =
FFT_LIBS =
NETCDF_HOME =
NETCDF_DEFINES =
NETCDF_MOD =
NETCDF_LIBS =
#DIRFRC_DEFINES = -DDIRFRC_EFS -DDIRFRC_NOVEC
CPP = /lib/cpp
NFLAGS=-xlic_lib=sunperf
##CPPFLAGS = -P
#F90_DEFINES = -DFFTLOADBAL_2PROC
PREPROCFLAGS = -DMPI -I/opt/SUNWhpc/include

F90 = mpf90
#MODULE_SUFFIX = mod
#F90FLAGS = -c -auto
#F90_OPT_DBG = -g -traceback
F90_OPT_LO = -c -O1 $NFLAGS
F90_OPT_MED = -c -O2 $NFLAGS
#F90_OPT_HI = -c -fast $NFLAGS
F90_OPT_HI = -g -c -fast -ftrap=%none -xlic_lib=sunperf
#F90_OPT_HI = -c -fast -xlic_lib=sunperf
F90_OPT_DFLT = $(F90_OPT_HI)

OPT_LO = "mpf90 -c -O1 $NFLAGS"
OPT_MED = "mpf90 -c -O2 $NFLAGS"
#OPT_HI = "mpf90 -c -fast $NFLAGS"
OPT_HI = "mpf90 -c -fast -xlic_lib=sunperf"
OPT_EW_PAIRLIST = "$OPT_MED"

CC = mpcc
CFLAGS = -fast $NFLAGS

LOAD = mpf90
LOADFLAGS =
LOADLIBS = -g -ftrap=%none -L/opt/SS12/SUNWspro/lib/ -xlic_lib=sunperf
-lmpi -fast
#LOADLIBS = -g -L/opt/SS12/SUNWspro/lib/ -xlic_lib=sunperf -lmpi -fast
#LOADLIBS = "-L/opt/SS12/SUNWspro/lib/ -xlic_lib=sunperf -lmvec -fast"

Any suggestions to fix the problem appreciated - thanks in advance

With kind regards
Per

On Mon, Apr 27, 2009 at 8:25 PM, Robert Duke <rduke_at_email.unc.edu> wrote:
> What is the chip? The compiler? The config file? This is definitely a
> config or compiler or unsupported s/w or h/w problem. - Bob Duke
> ----- Original Message ----- From: "Per Jr. Greisen" <pgreisen_at_gmail.com>
> To: "AMBER Mailing List" <amber_at_ambermd.org>
> Sent: Monday, April 27, 2009 1:11 PM
> Subject: Re: [AMBER] Error with calculation of the potential energy
>
>
> Hi,
>
> When I try to test the pmemd it seems that it has an error with
> calculating the potential energy - I put out the output from some of
> the tests.
>
> export TESTsander='../../exe/pmemd'; cd dhfr && ./Run.dhfr
>
> which returns some of the following in the TEST_FAILURES.diff
>
> 78,79c78,79
> < NSTEP = 1 TIME(PS) = 510.051 TEMP(K) = 301.29 PRESS = 0.
> < Etot = -57751.8286 EKtot = 14110.5335 EPtot =
> -71862.3622
> ---
>>
>> NSTEP = 1 TIME(PS) = 510.051 TEMP(K) = 301.65 PRESS = 0.
>> Etot = 3706.9950 EKtot = 14127.5125 EPtot =
>> -10420.5175
>
> 81,82c81,82
> < 1-4 NB = 545.9440 1-4 EEL = 6666.3920 VDWAALS =
> 8109.3892
> < EELEC = -89881.6441 EHBOND = 0. RESTRAINT = 0.
> ---
>>
>> 1-4 NB = 545.9440 1-4 EEL = 6666.3920 VDWAALS -727.9808
>> EELEC = -19602.4294 EHBOND = 0. RESTRAINT = 0.
>
> 84,88c84,88
>
> so it seems that it has not been compiled correctly; here it is
> compiled with ftrap=%none - if I on the other hand compile remove
> that compiling option the pmemd ends with FPE - division by zero. If I
> write the following part out in pairs_calc_vec.i by modifying the
> code:
>
> ! Do the Coulomb part of the direct sum using erfc spline table:
>
> delr = sqrt(delr2)
> write (0,*) delr
>
> delr is set to zero - how to fix this and why should one compile with
> ftrap=%none?
>
> Thanks in advance
>
> Best regards
>
> Per
>
>
>
>
>
>
>
>
>
>
> On Mon, Apr 27, 2009 at 1:08 PM, Carlos Simmerling
> <carlos.simmerling_at_gmail.com> wrote:
>>
>> please give more info- you said serial, parallel and pmemd, but only give
>> 2
>> examples so we don't know where to look for the problem. also give the
>> whole
>> section from the energy, not just the totals, so we can see which
>> components
>> differ. finally, did all of the test cases pass?
>>
>> On Mon, Apr 27, 2009 at 7:03 AM, Per Jr. Greisen <pgreisen_at_gmail.com>
>> wrote:
>>
>>> Hi all,
>>>
>>> I have compiled a serial, parallel, and pmemd of amber but I get some
>>> strange results when I try to run the test examples it seems that the
>>> potential energy is calculated wrongly in the different test cases
>>>
>>> < NSTEP = 1 TIME(PS) = 5420.002 TEMP(K) = 301.21
>>> PRESS = -5.8
>>> < Etot = -608.3990 EKtot = 1063.0604 EPtot =
>>> -1671.4594
>>> ---
>>> > NSTEP = 1 TIME(PS) = 5420.002 TEMP(K) = 301.03 PRESS =
>>> -961.1
>>> > Etot = 1257.0312 EKtot = 1062.3997 EPtot =
>>> 194.6315
>>> 93,95c93,95
>>>
>>> Any suggestions on how to fix this? Thanks in advance
>>>
>>> With kind regards
>>>
>>> Per
>>>
>>> _______________________________________________
>>> 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
>
>
>
> _______________________________________________
> 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