nisse at lysator.liu.se
Thu Aug 4 21:13:08 CEST 2011
Torbjorn Granlund <tg at gmplib.org> writes:
> Given the limited use of the code, I was thinking of computing with many
> extra bits, simply checking that no final value is close to a tie case,
> and abort if that happens.
Sounds reasonable to me, but it would be nice to define rigorously what
thresholds should be used to define the "tie cases".
> The code should also generate floor(B*log(b)/log(2)). Then we can get
> rid of all floating point arithmetic in GMP, including very slow
> floating point divides.
Could be done essentially with the same algorithm, but tabulating the
roots two; there's no need to recompute them for each base. Right?
And then it should also be reasonbly wasy to go from one table to the
other (just have to think a little bit about rounding directions), so
this may be a faster way to compute the log(2)/log(b) table as well.
Niels Möller. PGP-encrypted email is preferred. Keyid C0B98E26.
Internet email is subject to wholesale government surveillance.
More information about the gmp-devel