Re: Building octave with large matrices support

[Jaroslav Hajek]

On Wed, Aug 26, 2009 at 10:22 PM, John W. Eaton<address@hidden> wrote:
> On 25-Aug-2009, Jaroslav Hajek wrote:
>
> | Yes, that is essential. Octave's index type (octave_idx_type) must
> | match the Fortran default integer, otherwise you'll see failures (even
> | segfaults). Also, there's currently no way to use a mixture (say,
> | 32-bit BLAS with 64-bit SuiteSparse and 64-bit Octave) - that would be
> | straightforward but big and tedious change (you'd need to specify a
> | separate index type for each library and let configure sort it all
> | out).
>
> Yes, it might be possible, but I don't have much interest in
> introducing this extra complexity.
>
> I would like to improve the configure script to ensure that the
> libraries have the correct integer size when Octave is configured with
> the --enable-64 option.  As it is now, there are no checks, so it is
> easy to build a version that won't really work properly.
>

BLAS and LAPACK are already checked against the Fortran conventions,
including integer return values (ISAMAX).
I think we should at least check that the Fortran integer matches
octave_idx_type, which is an assumption made throughout Octave
sources.

-- 
RNDr. Jaroslav Hajek
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz