[SIESTA-L] How to optimize the bulk with SIESTA

[Hsin-Yen Chen]

Dear Sir,
I like to optimize the BaTiO3 bulk in Cublic phase.
However, the "MD.TypeOfRun  CG" was not converged with "MD.NumCGsteps 500".
Furthermore, I checked that the output file to find the message,
"siesta: System type = molecule". Did it be right?
My object is to verify the BaTiO3 phonon properties and investigate the other
perovskite ferroelectics.

Would someone provide the demo input to verify the BaTiO3 phonon properties?

Thanks!


 Hsin-Yen Chen

 Computing Centre, Academia Sinica

 P.O. Box 1-8, Nankang, Taipei 11529
 TAIWAN
 TEL: 886-2-27899247
 FAX: 886-2-27836444
 E-mail: [EMAIL PROTECTED]

Re: [SIESTA-L] Core dump when using more than 2 atom types and program compiled with IFC 8.0

[Eduardo Anglada]

On Wednesday 12 May 2004 22:40, you wrote:

Hi,

It's a problem with the compiler. The only version of ifc 8 which seems to
work  is the 0.0.46 (available at premier.intel.com , not at the standard
location of the compiler):

[EMAIL PROTECTED]:~> ifc -V
ifc: warning: The Intel Fortran driver is now named ifort.  You can suppress
this message with '-quiet'
Intel(R) Fortran Compiler for 32-bit applications, Version 8.0   Build
20040412Z Package ID: l_fc_pc_8.0.046
Copyright (C) 1985-2004 Intel Corporation.  All rights reserved.
FOR NON-COMMERCIAL USE ONLY

ifort: Command line error: no files specified; for help type "ifort -help"

Also don't use the -static flag. If you want a "portable" binary use the
following flag:

-static-libcxa

The resulting binary will run on any system which has the same glibc as the
one used for compiling.

Regards,

Eduardo

Re: [SIESTA-L] Siesta on Linux -- Is it possible?

[Lucas Fernandez Seivane]

Hi to all

It is completely posible, in fact the Oviedo group uses it a lot on a 12
PCs cluster (Pentium 4, ethernet100). As
a lot of people has noted here, you have to take care to compile all the
libraries properly (BLAS, LAPACK, SCALAPACK).

As other people has previously told, if you really want to get top-notch
performance, you will probably have to switch from ethernet to myrinet or
similar (Infiniband, SCI...). With E100 or similar, you only get about a
50% advantage using various computers.

Best wishes

On Wed, 12 May 2004, Juan Manuel wrote:


Neubauer, Paul R. wrote:

>I received no answers to my last set of questions, so I'll cut down the 
questions to a single, simple question.
>
>Is there anyone, anywhere in the world, who has successfully compiled siesta 
to run on Intel architecture Linux with MPI?
>
>If this proves to be possible, I will inquire further about their environment, 
but for the moment, I will happily settle for confirmation that it is possible at 
all.
>
>Thanks very much,
>Paul
>
>--
>Paul Neubauer
>Sr. Systems Consultant
>Universiaty Computing Services
>Ball State University
>Muncie, IN 47306 USA
>(765)285-1936
>[EMAIL PROTECTED]
>
>
>
It possible.

I have 2 pc (P4-2800) whith Gigabit ethernet link.
Whith debian woody (an OpenMosix system).

IFC Version 7.1   Build 20040309Z (VERY IMPORTAN, early versi?on not
compile siesta)
( the opt "-tpp7 -xW" if for P4 optimitation)
LAPACK -> ver: 3.0
 OPTS = -u -FI -O2 -mp -tpp7 -xW -Vaxlib -static
 NOOPT= -u -FI -O0 -mp -Vaxlib -static
ATLAS -> ver: 3.6.0
 F77FLAGS = -xW -O2 -mp1 -static
 CCFLAG0  = -xW -O3 -mp1 -static
MPI -> mpich-1.2.5.2 -f90flags="-tpp7 -xW -O2 -mp1 -static"
BLACS -> blacks 1.1
 mpif90
 F77NO_OPTFLAGS = -u -FI -O2 -mp -tpp7 -xW -Vaxlib -static
 mpicc
 CCFLAGS= -tpp7 -xW -mp -O2 -static
SCALAPACK -> scalapack 1.7
 mpif90
 F77FLAGS  = -u -FI -w90 -w95 -cm -O2 -mp -tpp7 -xW -Vaxlib -static
 mpicc
 CCLOADFLAGS   = -O2 -mp -tpp7 -xW -static

arch.make :  (not all)

SIESTA_ARCH=intel-atlas-mpich
.
FC=mpif90
#
FFLAGS= -tpp7 -xW -O2 -static
FFLAGS_DEBUG= -g
LDFLAGS=-Vaxlib
LDFLAGS=
COMP_LIBS=
RANLIB=ranlib
#
FREE_F90=-FR
#

MPI_INTERFACE=libmpi_f90.a
MPI_INCLUDE=/usr/local/intel/compiler71/mpich-1.2.5.2-icc/include
MPI_LIBS= /usr/local/intel/compiler71/mpich-1.2.5.2-icc/lib/libmpich.a
BLACS_LIBS=/usr/local/intel/compiler71/blacs-1.1/blacsF77init.a
/usr/local/intel/compiler71/blacs-1.1/blacs.a
SCALAPACK_LIBS=/usr/local/intel/compiler71/scalapack-1.7/libscalapack.a
DEFS_MPI=-DMPI
#
LAPACK=-L/usr/local/intel/compiler71/lapack-3.0 -llapack
ATLAS=-L/usr/local/intel/compiler71/atlas-3.6.0/icc -llapack -lf77blas
-latlas -lsvml
#
LIBS=$(NETCDF_LIBS) $(SCALAPACK_LIBS) $(BLACS_LIBS) $(MPI_LIBS)
$(LAPACK) $(ATLAS)
SYS=bsd
DEFS= $(DEFS_CDF) $(DEFS_MPI)
.


The ratio 1-2 cpus is 1.66 (2cpus is 66% faster)