tg at gmplib.org
Thu Dec 8 12:35:46 CET 2011
bodrato at mail.dm.unipi.it writes:
> Would you consider migrating this to mpn at some point, or is there a
> reason to keep it at mpz?
It is possible to migrate to mpn, but not very easy, because one should
estimate the length of every partial result. There are a lot of
MPZ_REALLOC in the current code...
It should also be possible to have a far better memory management and to
reduce the global footprint of the computation.
I see. The new log table I committed a few months back might help
in computing Stirling's formula:
log(n!) ~= 0.5*log(2*pi) + (0.5+$n)*log($n) - $n
But the important thing now is to get the basic improvements.
Meanwhile, we have a regression for a handful of hosts:
(Note that gmplib.org/devel/tm-date.html is bright red; there is a
regression related to fat binaries too, for which I am responsible.)
More information about the gmp-devel