Hello,
I hope it's o.k, to email this address directly. I did post a query on
the forum a little while back, but haven't had any responses yet.
I'm working on a fortran code that needs to be compiled with 8-byte
integers (integer*8) as otherwise we can't address all of the memory
that we need.
Unfortunately, this seems to prevent my calling BLACS and ScaLAPACK from
my code. I've tried various combinations of building the software stack
(BLAS, LAPACK, BLACS and ScaLAPACK) with 8-byte Fortran integers, but it
always falls over - presumably because I'm building on top of a 4-byte
integer MPI implementation, which I'm not in a position to change (this
is on a Power5+ system running Suse Linux).
I've tried building BLACS with i4 and then compiling the BLACS tests
with i8, but both xCbtest and xFbtest fail with the error:
BLACS WARNING 'Cannot set BLACS system context, can only BLACS_GET'
from {-1,-1}, pnum=0, Contxt=-1, on line 14 of file 'blacs_set_.c'.
BLACS WARNING 'Cannot set BLACS system context, can only BLACS_GET'
from {-1,-1}, pnum=1, Contxt=-1, on line 14 of file 'blacs_set_.c'.
BLACS ERROR 'Illegal grid (0 x 4), #procs=4'
BLACS WARNING 'Cannot set BLACS system context, can only BLACS_GET'
from {-1,-1}, pnum=2, Contxt=-1, on line 14 of file 'blacs_set_.c'.
BLACS ERROR 'Illegal grid (0 x 4), #procs=4'
from {-1,-1}, pnum=2, Contxt=-1, on line -1 of file
'BLACS_GRIDINIT/BLACS_GRIDMAP'.
BLACS WARNING 'Cannot set BLACS system context, can only BLACS_GET'
BLACS ERROR 'Illegal grid (0 x 4), #procs=4'
from {-1,-1}, pnum=0, Contxt=-1, on line -1 of file
'BLACS_GRIDINIT/BLACS_GRIDMAP'.
from {-1,-1}, pnum=1, Contxt=-1, on line -1 of file
'BLACS_GRIDINIT/BLACS_GRIDMAP'.
from {-1,-1}, pnum=3, Contxt=-1, on line 14 of file 'blacs_set_.c'.
BLACS ERROR 'Illegal grid (0 x 4), #procs=4'
from {-1,-1}, pnum=3, Contxt=-1, on line -1 of file
'BLACS_GRIDINIT/BLACS_GRIDMAP'.
ERROR: 0031-250 task 2: Terminated
ERROR: 0031-250 task 1: Terminated
ERROR: 0031-250 task 0: Terminated
ERROR: 0031-250 task 3: Terminated
Is there any way to work around this? Presumably this could be made to
work if BLACS could do the translation from i8->i4 before calling MPI,
but I've no idea how much work would be involved in doing this.
It seems that some vendors have i8 versions of BLACS available (although
I'm not sure if these build on top of i4 or i8 MPI), so presumably at
least some of the code required to get this to work has been written at
some point.
Is there any source for this sort of build available in the public
domain, or is there any other way to get this working?
I'd be very grateful for any suggestions.
Best wishes,
Jens
--
===================================================================
Jens Thomas, email: j.m.h.thomas@Domain.Removed
STFC Daresbury Lab, tel: +44-1925-603849
Warrington, fax: +44-1925-603634
WA4 4AD, UK. http: http://www.cse.scitech.ac.uk
===================================================================
|