AMBER Archive (2002)

Subject: Re: compiling amber on a linux box using DMEM-ALLOC

From: Scott Brozell (sbrozell_at_scripps.edu)
Date: Fri Sep 06 2002 - 19:16:40 CDT

Hi,

Ok, rt_polymerase is a big calculation. One can make order of magnitude
estimates of the necessary sizes by reading bench.rt_polymerase.out.save,
keeping in mind that 8 processors were used.
This sander/sizes.h works with that benchmark on a Xeon box running
RedHat7.3 with amber7 compiled by g77:

c
c ------------start sizes.h-----------------------------------------------
c
      integer MAXINT,MAXPR,MAXREA,MAXHOL,MAXDUP
c
c --- following five lines give "master" sizes for sander. These
c generally are what need to be changed when resizing the code.
c
c --- If the sizes above are too small, the program will complain.
c Here are some typcial values:
c
c Use Parameter Typical Values for Large Systems
c -------------------------------------------------------------------------
c floating pt MAXREA 25 * Natom (more for pol., min. gb)
c integers MAXINT 12 * Natom (more for pol.)
c holerith MAXHOL 6 * Natom (more for pol.)
c integers MAXPR cutoff-dependent, less for parallel systems
c dihedral dup MAXDUP data dependent -- generally 0 when
c LEaP is used, a few hundred for old
c prmtops for moderate sized systems.
c Program will complain if this is too small
c -------------------------------------------------------------------------
c
c Note: if you are carrying out NMR-based refinement, you may also
c need to set some sizes in the file "nmr.h".
c
c Note further: if -DMEM_ALLOC is set in the MACHINE file, MAXREA, MAXINT,
c MAXPR and MAXHOL are not used; these sizes are determined by the code,
c and allocated at run-time.
c
c --- "standard" (pretty-big) sizes: should work for test cases
c
#ifndef MEM_ALLOC
      parameter (MAXREA=20000000)
      parameter (MAXINT=20000000)
      parameter (MAXHOL=1000000)
      parameter (MAXPR=30000000)
#endif
      parameter (MAXDUP=1000)
c
c --- allocate a "stack" space for temporary real variables:
c (size depends on the problem: the maximum value used is reported
c at the end of a calculation)
c
c (When MEM_ALLOC is set at compilation time, MAX_RSTACK and MAX_ISTACK
c just give the default values; these can be overridden in the
c &cntrl namelist using "lastrst" and "lastist".)
c
      integer MAX_RSTACK,MAX_ISTACK,MAX_STACK_PTRS,MAX_HEAP_PTRS
      parameter (MAX_RSTACK=10000000)
      parameter (MAX_ISTACK=10000000)
      parameter (MAX_STACK_PTRS=100)
      parameter (MAX_HEAP_PTRS=100)
c
c ------------end sizes.h-----------------------------------------------
c

For a 1 processor job the executable uses 252M.
MAXREA and MAXINT could be decreased;
from the output:

| Memory Use Allocated Used
| Real 20000000 10163811
| Hollerith 1000000 889171
| Integer 20000000 10528792
 
...

| Local SIZE OF NONBOND LIST = 28907091
| TOTAL SIZE OF NONBOND LIST = 28907091

scott

On Fri, 6 Sep 2002, ramon kleber da rocha wrote:

> On Fri, 6 Sep 2002, Scott Brozell wrote:
>
> > Hello,
> >
> > MEM_ALLOC requires the dynamic memory allocation methods of
> > Fortran 90. Hence, it is not compatible with g77.
> > Either the Intel or Portland Group fortran 90 compilers
> > for Linux will support MEM_ALLOC.
> > Alternatively, it is straightforward to modify
> > sander/sizes.h
> > usually
> > parameter (MAXPR=8400000)
> > is big enough.
>
> Dear Scott,
> Thank you for your reply. I am looking for a compiler that support
> memory allocation methods in order to install amber7. Do you know if any
> linux distribution provide that compiler? Besides, I have changed MAXPR to
> 10800000, but I got another error message when I try to run the benchmark
> for rt_polymerase. Thats the message:
>
> Real Memory requirement of: 9740309 exceeds MAXREA of 4000000
>
> Hollerith Memory requirement of: 889171 exceeds MAXHOL of 800000
>
> Static Integer Memory requirement of:10397638 exceeds MAXINT of
> 4000000
>
> Max Nonbonded Pairs:10800000
>
>
> Do you know something about that?
> Best regards,
>
>
> ________________________________________________________________________________
> Ramon Kleber da Rocha, MSc. VOICE +55-31-3499-5765
> e-mail rkrocha_at_dedalus.lcc.ufmg.br FAX +55-31-3499-5700
>
> Laboratorio de QSAR e Modelagem Molecular
> Nucleo de Estudos em Quimica Medicinal - NEQUIM - http://www.qui.ufmg.br/~nequim
> Departamento de Quimica - Universidade Federal de Minas Gerais - (DQ/ICEx/UFMG)
> ---
> The World is my Fatherland, Science is my Religion. Christiaan Huygens 1629-1695