Allocation for toom33

[Niels Möller]

Torbjorn Granlund <tg at gmplib.org> writes:

> I checked the measured data to see if toom33 is anytime considered the
> best function and s + t >= n + 5 does not hold.  Here are the points for
> a core i7:
>
> [...]

One other observation: They all have large s and small t. Is there any
input sizes where it's optimal to choose a toom varaiant that results
in a small s?

Does it ever happen that s <= an - M? If it does, how small can it be?
Can it ever happen that s < t?

Regards,
/Niels