The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Hi,
In fact most compilers do not allow allocating twice the same array.
Unfortunately, as you told, some bugs are system dependent and are
triggered only on some system/compiler couples. Would you mind telling
us which of the automated test triggered the bug first ?
regards

PMA

On 3/26/07, Masao ARAI <ARAI.Masao@nims.go.jp> wrote:
> Thank you for your reply.
>
> On SR-11000, the error occurs on the tests supplied in
> the abinit source tree.
>
> The problem depends on systems or compilers.
> In some (most?) systems, the problem is just ignored (with
> possible memory leaks). But, unfortunately, compiler by Hitachi is
> too strict. It does not allow us to allocate same array twice.
>
> Best regards,
> Masao Arai
>
> > Hi,
> >
> > First, Yes this mailing list is appropriate for any type of comment,
> > questions, remarks ... Yet, there is also a developer@abinit.org
> > mailing list.
> > It is very nice from you to come with this bug report + solution.
> > Thank you very much. We would be very interesting in having an input
> > file enabling us to reproduce this bug. This way it may become an
> > automatic test and prevent the bug from popping up anytime in the
> > future. It would also help us confirm the bug (although your code
> > analysis looks very clear).
> >
> > best regards
> > PMA
> >
> > On 3/26/07, Masao ARAI <ARAI.Masao@nims.go.jp> wrote:
> >> Dear all,
> >>
> >> When I tested abinit-5.3.3 on Hitachi SR-11000, I got following error
> >> from
> >> abinip.
> >>
> >> KCHF503R allocatable array is already allocated.
> >> 0x0000000100008194 _hf_err+0x554
> >> 0x00000001000998bc _hf_alloc+0x2ac
> >> 0x0000000100454c54 fresid+0x854
> >>
> >> This message indicates that some arrays are allocated multiple times.
> >> I checked "15common/fresid.F90".
> >>
> >> At line 241, two arrays "rhor" and "work" are allocated.
> >>
> >> allocate(rhor(nfftot,dtset%nspden),work(nfftot,dtset%nspden))
> >>
> >> These arrays are deallocated at line 679 only when
> >> (mpi_enreg%mode_para=='b').
> >>
> >> +677 if(mpi_enreg%mode_para=='b') then
> >> +678 call
> >> pre_scatter(work_slice,work,ngfft(1),ngfft(2),ngfft(3),mpi_enreg,'scatter')
> >>
> >> +679 deallocate(rhor,work)
> >> +680 else
> >> +681 work_slice(:,:)=work(:,:)
> >> +682 end if
> >>
> >> So, if (mpi_enreg%mode_para != 'b"), the fresid will try to allocate the
> >> same arrays each time
> >> it is called.
> >>
> >> If the deallocate is moved after if-endif block,
> >>
> >> if(mpi_enreg%mode_para=='b') then
> >> call
> >> pre_scatter(work_slice,work,ngfft(1),ngfft(2),ngfft(3),mpi_enreg,'scatter')
> >>
> >> else
> >> work_slice(:,:)=work(:,:)
> >> end if
> >> deallocate(rhor,work)
> >>
> >> the abinip seems to work fine.
> >>
> >> Best regards,
> >> Masao Arai
> >>
> >> P.S.
> >> Is this ML appropriate for sending these reports? Or, should I send to
> >> developer ML?
> >>
> >> --
> >> Masao Arai
> >> CMSC, NIMS, JAPAN
> >>
> >
> >


--
Pierre-Matthieu Anglade