Is anyone working on making octave use the full precision of 64-bit
machines?
Currently, compiling octave on a x86_64 produces a 64-bit executable.
But, because of the definition of double, this gives no precision gain
for floating point over 32-bit machines.
With gfortran, I can use -fdefault-real-8 and -fdefault-integer-8 and
create an application which uses more of the machine's precision.
So I gather that you are looking for more than 64 bit floating point precision.
Tatsuro's link on the other hand points to the effort of using 64 bit integer indexing, which would allow larger arrays but not higher precision.
The IEEE standard defines 128 bit floating point numbers, it is also called quad-float. I do not think that it is fully supported by gfortran. I assume so, since there must be a reason that I cannot find BLAS and LAPACK that features quad-float operations.
Without those I think Octave cannot do much to bring useful quad-float support to the users.
Maybe it is just adding the new function signatures and recompiling BLAS/LAPACK, somehow though I doubt that.
-Matyas