[Wien] lapw1(c)_mpi crash in the initialization procedure

[Peter Blaha]

> Obviously, it was not a good idea to send a large attachment. But as
> I think, you do not really need the attachment here the same mail
> without attachment:

I approved manually this larger mail and it will come through some time.

The blacs_get line is in the code, so this cannot explain the problems.
(VERSION_05 should be identical to VERSION_03 for these mpi stuff.)

When comparing hns.F and modules.F in VERSION_02 (before the optimization 
of hns.F) with the current version, I see that these allocation statements 
have been changed in hns.F from:
      allocate( A1R(HSROWS, NSL2*3), A1I(HSROWS, NSL2*3) )
 to
#ifdef Parallel
      allocate( A1R(lld, NSL2), A1I(lld, NSL2) )
 
In modules.F, however, the routine par_syr2k  has not been changed.
While this is fine for the REAL version since their it works like:
          allocate( ABCLM1(LDHS, size(A1R,2)) )
....
!_REAL          CALL COPYTOABCLM(TYPE,LINDEX,NV,NLO,A1R,ABCLM1)

in the complex version there is a step involving   HSROWS elements:
!_COMPLEX          allocate( A(HSROWS, size(A1R,2)) )
....
!_COMPLEX          A(1:HSROWS,1:LINDEX)=DCMPLX(A1R(1:HSROWS,1:LINDEX), &
!_COMPLEX                                     -A1I(1:HSROWS,1:LINDEX))
!_COMPLEX          CALL COPYTOABCLM(TYPE,LINDEX,NV,NLO,A,ABCLM1)
...

My guess would be to change all the HSROWS in this routine to LDHS.

As mentioned before, I cannot test this at the moment.


> I at least have an idea. There was a correction to the original
> version I had generated in connection with my optimizations of HNS. 
> The modules.F file should contain the line
>    
>        call blacs_get(0,0,ictxtallr)
> 
> just before the line
> 
>        call blacs_gridinit(ICTXTALLR,'R',npe,1)
> 
> This is not necessary on the IBM machines where I developed the
> code, but it is the correct way and necessary on all other archi-
> tectures. This should have been corrected in the official version 
> of Wien2k_03 as an update. Nevertheless, I send you my version of
> modules.F as attachment. All these remarks concern for the moment
> only Version 03, I have no experience with version 05 up to now.
> 
> Can you check, whether this may have caused the error you see? 
> 
> Best regards,
> Renate Dohmen
> 
> ------------------
> Dr. Renate Dohmen
> Computing Centre Garching of the MPG
> Boltzmannstr. 2
> D-85748 Garching
> Germany
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> 


                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------