Re: Rethinking octave_idx_type

[John W. Eaton]

On 11/26/2016 03:50 PM, Mike Miller wrote:

> On 64-bit systems, octave_idx_type would now be a 64-bit value, and
> additional logic would have to be put in place now to ensure that we
> don't pass 64-bit array sizes into BLAS and other Fortran libraries,
> unless the user builds Octave with --enable-64.

Yes, and pointers on these systems are already 64-bits.  Mostly we are 
talking about changing loop counters and array indices to be 64-bits 
instead of 32, not generating arrays of index values that are now twice 
as large.  So I don't think there is much of a memory issue for dense 
arrays.

The sparse matrix implementation could also switch to using 64-bit 
arrays by default for the row and column index vectors and that would 
require more memory.  But fairly often we also see complaints about not 
being able to store very large sparse matrices, so I think this would be 
a good change.

When using 32-bit BLAS on 64-bit systems we would now have to check all 
array dimensions to ensure that they do not overflow 32-bit signed 
integers.  Integer work arrays would remain 32-bit integers as well.

> Is it possible today to build Octave on a 32-bit system with
> --enable-64?

No.  This proposal doesn't affect 32-bit systems.

jwe