Re: sizemax() and 64-bit systems

[Mike Miller]

On Mon, 13 Aug 2012, Martin Helm wrote:

> Am 13.08.2012 07:59, schrieb Mike Miller:
> > Is there another way to get the sizemax() functionality?  I'm trying
> > to find out for these machines what is the largest number of elements
> > per matrix.  Is there a way to figure it out from octave_config_info?
>
> Just look for USE_64_BIT_IDX_T in octave_config_info, if that is set to 
> false the limit will be roughly 2e9 (a bit less than 2^31).

Thanks.  That is "false" on all of my machines.  Here's a command for 
Linux users that will show this info (from the shell command line):

echo "octave_config_info" | octave -qfH | grep IDX
  USE_64_BIT_IDX_T = false

Now for the next big question -- what is the trick to compiling Octave so 
that it will allow more elements per matrix?

By the way, I'm having the same problem with R.  It is hard-coded to limit 
the number of elements to INT_MAX, which is 2^31-1, but why aren't we 
using uint_max instead?

grep INT_MAX /usr/include/limits.h
#  define INT_MIN       (-INT_MAX - 1)
#  define INT_MAX       2147483647
#  define UINT_MAX      4294967295U

It just seems like indexing would be done with uint instead of with int 
because element indexes are always positive integers, aren't they?  I 
think maybe the problem is in FORTRAN.

It sounds like changing to 64-bit indexing is a big deal for R, but it 
also sounds like Octave has it down to a compile-time option.  Oh, just 
hit the jackpot:

http://www.gnu.org/software/octave/doc/interpreter/Compiling-Octave-with-64_002dbit-Indexing.html

Now I know what needs to be done.  It is doable, much more so in Octave 
than in R, but it is also a lot more work than just compiling Octave 
better because all the numerical libraries need to be compiled for 64-bit.

This leads to yet another question -- are these libraries already compiled 
for 64-bit in any Linux repositories?  Is there a way to test whether mine 
are already 64-bit?

Thanks again, very much, for the excellent information.

Mike

--
Michael B. Miller, Ph.D.
Minnesota Center for Twin and Family Research
Department of Psychology
University of Minnesota