TL/DR version: no, I haven’t
I remember repeated squaring¹ of an integer matrix
where I had wanted to use extended for not losing
the exact precision. Don’t know for sure the results
mostly exceeded 2^63 but I think so.
Those matrices have not been all that “large” however
so maybe this is not an example of what you asked for.
Dividing by a power of two at some steps
would lose me some precision so that’s
not an option here.
Maybe someone knows of a better way?
For rationals, I played with Hilbert matrices
(for academic reasons only) – again, not all
that large.
I don’t remember truly large arrays that didn’t
allow for some loss of precision other than
“rounding” to zero or infinity – except some
instances of toying around maybe, but never
for often-called functions in application.
¹ as part of an n-th power function, but it was
mostly used with some power of two anyway
Am 18.08.22 um 08:34 schrieb Elijah Stone:
Another strange question, this time to do with extended precision numbers.
Have you ever wanted good performance out of a large array of
extended-precision integers (or, for that matter, rationals), where most
of the integers involved would not have fit in a machine integer
(magnitude approx. 2^63)?
(I recognise that this is probably not an ideal place to ask, given that
j's extended integers have not historically been even remotely
performant. But most things that aren't j have poor support for arrays,
sooo~)
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
--
----------------------
mail written using NEO
neo-layout.org
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
