Thank you.
If I remove the include "mpif.h" statement, it becomes
a simple serial code which runs into no problem.
I checked the mpif77 -showme, actually they are
showing something a little bit difference:
Version 7.0.2:
/opt/absoft/bin/f90 -I/opt/lammpi/include
-L/opt/lammpi/lib -llamf77mpi -lmpi -llam -lutil
-lpthread
Version 7.1.1:
/opt/absoft/bin/f90 -I/opt/lam711//include -lpthread
-L/opt/lam711//lib -llamf77mpi -lmpi -llam -lutil
-lpthread -lpthread -ldl
There is a -ldl which is not in the 7.0.2 version and
-lpthread duplicated. Will this cause any trouble?
I also tried a debugger, there is nothing I can found.
The 7.1.1 compiled one hung at the write statement. I
checked the values of the array, they all looked
right.
Could you please try the toy code I mentioned in the
original post on a 7.1.1 version?
Thank you very much.
Shi
--- Josh Hursey <jjhursey_at_[hidden]> wrote:
> humm... Doesn't seem that the sample piece of code
> is using any MPI
> calls so that eliminates the possibility of the
> problem lying in
> LAM/MPI libraries. So I would focus on making sure
> that mpif77 is
> configured in a similar manner as before. One easy
> way to do this is to
> "mpif77 -showme" on both installations and compare
> the results. You
> could also try to compile the code not using mpif77,
> but using your
> standard fortran compiler and see if that narrows
> down the problem.
>
> You could also attach to a debugger, and use that to
> see where it might
> be hanging.
>
> Hope that helps,
> Josh
>
>
> On Mar 4, 2005, at 3:44 PM, Shi Jin wrote:
>
> > Hi, I have some problem with LAM/MPI 7.1.1 in
> dumping
> > large arrays to a binary file while the same works
> > with 7.0.2 perfectly. To demenstrate the problem,
> try
> > this tiny toy code:
> > program largefile
> > implicit none
> > include "mpif.h"
> > integer,parameter::recordunit=11,N=256
> > real(8),allocatable::sepRecord(:)
> > allocate(sepRecord(2*N**3))
> > sepRecord=1d0
> >
> > Open (UNIT=recordunit, FILE="largefile.dat",
> > STATUS="UNKNOWN", &
> > form="unformatted",ACTION="WRITE")
> > write(unit=recordunit) sepRecord
> > close(recordunit)
> >
> > deallocate(sepRecord)
> > end program
> >
> > For the LAM/MPI 7.0.2, it writes to a 256MB file
> in
> > seconds while using versin 7.1.1, the code hangs
> and
> > takes all the CPU but nothing is written.
> >
> > Is this a bug in the new version?
> > Thanks.
> >
> > Shi
> >
> >
> >
> >
> > __________________________________
> > Celebrate Yahoo!'s 10th Birthday!
> > Yahoo! Netrospective: 100 Moments of the Web
> > http://birthday.yahoo.com/netrospective/
> > _______________________________________________
> > This list is archived at
> http://www.lam-mpi.org/MailArchives/lam/
> >
> ----
> Josh Hursey
> jjhursey_at_[hidden]
> http://www.lam-mpi.org/
>
> _______________________________________________
> This list is archived at
> http://www.lam-mpi.org/MailArchives/lam/
>
__________________________________
Celebrate Yahoo!'s 10th Birthday!
Yahoo! Netrospective: 100 Moments of the Web
http://birthday.yahoo.com/netrospective/
|
