[Smeagol-discuss] Fw: Re: fw:parallel compilation problem

ambavale sagar sagarambavale at yahoo.co.in
Mon Jan 19 17:56:28 GMT 2009


Hi Ivan and all,
I told  in my previous mail that transport calculation of example is running well with 4 nodes. But when i tried with more nodes (8 nodes) it gives segmentation fault at random places. I changed my arch.make file as suggested i.e put mpif90 and removed "/" (mpif90 is in path). Few more things are changed as I searched in siesta forum though problem persists. Please have a look at my arch.make file given below and suggest if any changes should be made..
Thanks a lot.

my new arch.make file is :

..SUFFIXES:
..SUFFIXES: .f .F .o .a .f90 .F90

SIESTA_ARCH=intel10

SOURCE_DIR=/home/phy/para-smgl/smeagol.1.0b/
EXEC = smglpara

FPP=
FPP_OUTPUT= 
#FC=/opt/MPI/lam-7.1.4/lam/bin/mpif90
FC=mpif90
FC_ASIS=$(FC)
RANLIB=ranlib

SYS=nag

SP_KIND=4
DP_KIND=8
KINDS=$(SP_KIND)
 $(DP_KIND)

#FFLAGS=-g
#FFLAGS=  -O2 -mtune=pentium4 -zero
FFLAGS=  -O1
FPPFLAGS= -DMPI -DFC_HAVE_FLUSH -DFC_HAVE_ABORT -DGRID_DP 
LDFLAGS= -Vaxlib 
TRANSPORTFLAGS= -O1

ARFLAGS_EXTRA=

FCFLAGS_fixed_f=
FCFLAGS_free_f90=
FPPFLAGS_fixed_F=
FPPFLAGS_free_F90=

COMP_LIBS=linalg.a
#COMP_LIBS=dc_lapack.a liblapack.a libblas.a 

NETCDF_LIBS=-lnetcdf
NETCDF_INTERFACE=libnetcdf_f90.a
DEFS_CDF=-DCDF

#LIBS=$(SCALAPACK_LIBS) $(BLACS_LIBS) $(LAPACK_LIBS) $(BLAS_LIBS) $(NETCDF_LIBS)

#SIESTA needs an F90 interface to MPI
#This will give you SIESTA's own implementation
#If your compiler vendor offers an alternative, you may change
#to it here.
MPI_INTERFACE=libmpi_f90.a
MPI_INCLUDE=/opt/MPI/lam-7.1.4/lam/include/
#MPI_LIBS= -L/opt/MPI/lam-7.1.4/lam/lib -llamf77mpi -lmpi -llam -lnsl
DEFS_MPI=-DMPI

DEFS= $(DEFS_CDF)
 $(DEFS_MPI)

GUIDE=/opt/intel/mkl/10.0.1.014/lib/em64t/libguide.a
#BLAS_LIBS=libblas.a
#LAPACK_LIBS=dc_lapack.a liblapack.a
#BLACS_LIBS=/opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_blacs_ilp64.a
#SCALAPACK_LIBS=/opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_scalapack_ilp64.a  

#this also gives segmentation fault)
#SCALAPACK_LIBS= /opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_scalapack.a
#BLACS_LIBS= /opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_blacs_ilp64.a \
 #      /opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_em64t.a  /opt/intel/mkl/10.0.1.014/lib/em64t/libiomp5.a

#LAPACK=/opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_lapack.a
#BLAS=/opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_blas95.a
#LIBS_PATH=-L/opt/intel/mkl/10.0.1.014/lib/em64t
#LIBS= $(LIBS_PATH) $(NETCDF_LIBS) $(SCALAPACK_LIBS) $(BLACS_LIBS) $(GUIDE) $(BLACS) $(BLAS) \
#      $(LAPACK)
 $(OTHER_LIBS) /home/phy/siesta-2.0.1-para/Src/MPI/Interfaces.o  -lpthread -lguide

#LIBS= $(LIBS_PATH) $(NETCDF_LIBS) $(SCALAPACK_LIBS) $(BLACS_LIBS) $(GUIDE) $(LAPACK)  $(BLAS) \
 #     $(OTHER_LIBS)  -lpthread -lguide

#sagar
MKLPATH=/opt/intel/cmkl/10.0.1.014/lib/em64t
LIBS=-L$(MKLPATH) -lmkl_scalapack_lp64 -lmkl_blacs_lp64 -lmkl_intel_lp64 -Wl,--start-group -lmkl_sequential -lmkl_core -Wl,--end-group -lpthread

#Dependency rules are created by autoconf according to whether
#discrete preprocessing is necessary or not.
..F.o:
    $(FC) -c $(FFLAGS) $(INCFLAGS) $(FPPFLAGS) $(FPPFLAGS_fixed_F)  $< 
..F90.o:
    $(FC) -c $(FFLAGS) $(INCFLAGS) $(FPPFLAGS) $(FPPFLAGS_free_F90) $< 
..f.o:
    $(FC) -c $(FFLAGS) $(INCFLAGS) $(FCFLAGS_fixed_f)  $<
..f90.o:
    $(FC) -c $(FFLAGS) $(INCFLAGS)
 $(FCFLAGS_free_f90)  $<



--- On Sun, 18/1/09, Ivan Rungger <runggeri at tcd.ie> wrote:
From: Ivan Rungger <runggeri at tcd.ie>
Subject: Re: [Smeagol-discuss] fw:parallel compilation problem
To: sagarambavale at yahoo.co.in
Cc: "smeagol-post" <smeagol-discuss at lists.tchpc.tcd.ie>
Date: Sunday, 18 January, 2009, 11:12 PM

Hello Sagar,

>I tried leads calculation with 1 processor and then
> transport calculation with 4 processors. It ran smoothly.

This sounds good. The problem with compiling siesta is probably that you
specify FC with slashes ("/") in it, which will not work with the
Makefile as it is. Try to change FC=/opt/MPI/lam-7.1.4/lam/bin/mpif90
with FC=mpif90 (and make sure that the directory 
/opt/MPI/lam-7.1.4/lam/bin/ is the first in your PATH
 environment
variable), that should work.

Regarding the problem of siesta2.0 running a set of serial jobs instead
of one parallel, I don't know what can be the origin.

Anyway the good thing is that smeagol runs in parallel.

Cheers,

 Ivan



ambavale sagar wrote:
> Dear Ivan,
> I tried leads calculation with 1 processor and then transport
> calculation with 4 processors. It ran smoothly.
>  Then as per your advice i tried to compile siesta 1.3f1p (without
> smeagol patch) with same arch.make file, but it couldn't be compiled.
> In the very beginning it gave the error like:
>
> ==> Incorporating information about present compilation (compiler and
> flags)
> sed: -e expression #1, char 11: unknown option to`s'
> make: *** [version] error 1
>
> Then I tried compilation of siesta2.0 which compiled without error but
> for what
 ever number of nodes i give in run command , it runs
> serially. It neither gives the error message nor stops. With some
> minutely changed arch.make file it runs same process on all nodes and
> writes same output 4 times (for 4 nodes).
>
> Please help me about this. I am completely clueless..
>
> Regards
> Sagar
>
>
>
> --- On *Thu, 15/1/09, Ivan Rungger /<runggeri at tcd.ie>/* wrote:
>
>     From: Ivan Rungger <runggeri at tcd.ie>
>     Subject: Re: [Smeagol-discuss] fw:parallel compilation problem
>     To: sagarambavale at yahoo.co.in, "smeagol-post"
>     <smeagol-discuss at lists.tchpc.tcd.ie>
>     Date: Thursday, 15 January, 2009, 4:42 PM
>
>     Hello Sagar,
>
>        the leads calculation in smeagol can only be done on 1 processor.
Try
>     running the leads calculation with 1 processor, and then
 the smeagol
>     transport calculation in the /examples/gamma directory with more
processors.
>
>     However from your output is seems that the code is crashing rather at
>     the beginning of the calculation, in which case it would be a siesta
>     problem. Try to compile the bare siesta code (without the smeagol
>     patch), and see if that
>      runs for the siesta examples.
>
>     Cheers,
>
>      Ivan
>
>     ambavale sagar wrote:
>     > Hi,
>     > I tried many ways. But can not make it run in parallel. Can you
please
>     > look at the problem and my arch.make file.. I am sorry to post
the
>     > same mail again.
>     >
>     > Sagar
>     >
>     > --- On *Wed, 7/1/09, ambavale sagar
/<sagarambavale at yahoo.co.in>/*
>     wrote:
>     >
>     >    
 From: ambavale sagar <sagarambavale at yahoo.co.in>
>     >     Subject: parallel compilation problem
>     >     To: "smeagol-post"
>     <smeagol-discuss at lists.tchpc.tcd.ie>
>     >     Date: Wednesday, 7 January, 2009, 11:37 AM
>     >
>     >     Dear All and Alex,
>     >
>     >     I compiled parallel smeagol on Xeon dual processor quad core
>     >     workstation with lam-mpi and intel-mkl. the program compiled
>     >     without giving any error message. Now when I tried to run the
>     >     /examples/gamma/leads/Au.fdf on
>      2 or 4 nodes using command mpirun
>     >     -np 4 ~/para-smgl/smeagol.1.0b/Src/smglpara
<Au.fdf>Au.out
>     >
>     >     it gives me the "forrtl: severe (174): SIGSEGV,
segmentation
>     fault
>     >     occurred" error and stops at particular
 point :
>     >
>     >
>     >     SIESTA 1.3..7  -- [1.3f1 release] (04 Feb 2004)
>     >     Architecture  : intel-parallel-sagar
>     >     Compiler flags: 
>     >     PARALLEL version
>     >     NetCDF-capable
>     >
>     >     * Running on    4 nodes in parallel
>     >     >> Start of run:   7-JAN-2009  10:44:33
>     >
>     >
>     >     .....
>     >
>     >     ......
>     >
>     >     siesta: overlap: 12.1666 5.4652 2
>     >     siesta: overlap: rmaxh veclen direction
>     >     siesta: overlap: 12.1666 7.7290 3
>     >
>     >     superc: Internal auxiliary supercell: 5 x 5 x 3 = 75
>     >     superc: Number of atoms, orbitals, and projectors: 150 150
600
>     >      
>     > 
>         * Maximum
 dynamic memory allocated = 2 MB
>     >
>     >     siesta: ===============================
>     >       Begin MD step = 1
>     >       ===============================
>     >
>     >     superc: Internal auxiliary supercell: 5 x 5 x 3 = 75
>     >     superc: Number of atoms, orbitals, and projectors: 150 150
600
>     >
>     >     **********************
>     >
>     >     But when i use single node in a command i.e. 
>     >
>     >     mpirun -np 1 ~/para-smgl/smeagol.1.0b/Src/smglpara
>     <Au.fdf>Au.out
>     >
>     >     it runs successfully and ends normally. 
>     >
>     >     Where might be a mistake in compilation?
>     >
>     >     Below is my arch.make file:
>     >
>     >     ****************
>     >
>     >    
 SIESTA_ARCH=intel-parallel-sagar
>     >     #
>     >     FC=/opt/MPI/lam-7.1.4/lam/bin/mpif90
>     >     FC_ASIS=$(FC)
>     >
>     >     #FFLAGS=  -O2
>     >     FFLAGS= -w -xP -O3 -mp1
>     >    
>      LDFLAGS= -Vaxlib
>     >     COMP_LIBS=
>     >     TRANSPORTFLAGS= -c -tpp7 -O3
>     >
>     >     SOURCE_DIR=/home/phy/para-smgl/smeagol.1..0b/
>     >     EXEC = smglpara
>     >
>     >     NETCDF_LIBS=-lnetcdf
>     >     NETCDF_INTERFACE=libnetcdf_f90..a
>     >     DEFS_CDF=-DCDF
>     >
>     >     MPI_INTERFACE=libmpi_f90.a
>     >     MPI_INCLUDE=/opt/MPI/lam-7.1.4/lam/include/
>     >     DEFS_MPI=-DMPI
>     >
>     >     GUIDE=/opt/intel/mkl/10.0.1.014/lib/em64t/libguide.a
>     >    
 LAPACK=/opt/intel/mkl/10.0.1..014/lib/em64t/libmkl_lapack.a
>     >     BLAS=/opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_blas95.a
>     >     #sagar
>     >     BLACS=
/opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_blacs_ilp64.a 
>     >     /opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_em64t.a 
>     >     /opt/intel/mkl/10.0.1.014/lib/em64t/libiomp5.a
>     >
>     >     SCALAPACK_LIBS=
/opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_scalapack.a
>     >
>     >    
>      LIBS_PATH=-L/opt/intel/mkl/10.0.1.014/lib/em64t
>     >
>     >     LIBS= $(LIBS_PATH) $(NETCDF_LIBS) $(SCALAPACK_LIBS)
$(BLACS_LIBS)
>     >     $(BLAS) $(LAPACK_LIBS) $(BLAS_LIBS)
>     >     /home/guest/sagar/tran/smeagol.1.0b/Src/MPI/Interfaces.o   
>     >     $(BLACS) $(OTHER_LIBS) -lpthread
>     >
>     >     #
>     >    
 RANLIB=echo
>     >     SYS=bsd
>     >     DEFS= $(DEFS_CDF) $(DEFS_MPI)
>     >     #DEFS= $(DEFS_MPI)
>     >
>     >     #
>     >     .F.o:
>     >         $(FC) -c $(FFLAGS)  $(DEFS) $<
>     >     .f.o:
>     >         $(FC) -c $(FFLAGS)   $<
>     >     .F90.o:
>     >         $(FC) -c $(FFLAGS)  $(DEFS) $<
>     >     .f90.o:
>     >         $(FC) -c $(FFLAGS)   $<
>     >                    
>     >     #
>     >     ***************************************
>     >
>     >
>     >
>     >     Thanks
>     >
>     >     Sagar Ambavale
>     >
>     >     Research student..
>     >
>     >     The M.S.
>      University of Baroda
>     >
>     >     India
>     >
>  
   >
>     >
>     >    
>    
------------------------------------------------------------------------
>     >     Connect with friends all over the world. Get Yahoo! India
>     >     Messenger.
>     >    
>    
<http://in.rd.yahoo.com/tagline_messenger_1/*http://in.messenger.yahoo.com/?wm=n/>
>     >
>     >
>     >
------------------------------------------------------------------------
>     > Add more friends to your messenger and enjoy! Invite them now..
>     >
>    
<http://in.rd.yahoo.com/tagline_messenger_6/*http://messenger.yahoo.com/invite/>
>     >
>     >
------------------------------------------------------------------------
>     >
>     > _______________________________________________
>     > Smeagol-discuss mailing list
>     >
 Smeagol-discuss at lists.tchpc.tcd.ie
>     > http://lists.tchpc.tcd.ie/listinfo/smeagol-discuss
>
>
>     --
>      
>     =================================================
>     Ivan Rungger,
>
>     School of Physics and CRANN,  
>     Trinity College Dublin,  
>     Dublin 2,  IRELAND  
>     Phone: +353-1-8968454  
>     Email: runggeri at tcd.ie
>     =================================================
>
>
>               
>
>
> ------------------------------------------------------------------------
> Add more friends to your messenger and enjoy! Invite them now.
>
<http://in.rd.yahoo.com/tagline_messenger_6/*http://messenger.yahoo.com/invite/>



-- 
=================================================
Ivan Rungger,

School of Physics and CRANN,  
Trinity College Dublin,  
Dublin 2, 
 IRELAND  
Phone: +353-1-8968454  
Email: runggeri at tcd.ie
=================================================




       Add more friends to your messenger and enjoy!  Invite them now.


      Add more friends to your messenger and enjoy! Go to http://messenger.yahoo.com/invite/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.tchpc.tcd.ie/pipermail/smeagol-discuss/attachments/20090119/f3c66d80/attachment-0001.html 


More information about the Smeagol-discuss mailing list