Toom testing, and unbalanced recursive calls in toom22.

[Niels Möller]

Torbjorn Granlund <tg at gmplib.org> writes:

> Do you agree that if that if we need toom22 to allow further unbalanced
> operands, we should not call mul_basecase for known-suboptimal operands,
> but instead call toom32 for these operands?

Yes.

When I proposed to simply use mul_basecase, that was on the theory
than the difference between basecase and the best algorithm would
likely be very small over the range of practical interest. (And then I
don't care if basecase is highly sub-optimal for untypical sizes used
by the toom22 testcases).

A related question is, should we call toom22 for known-suboptimal
operands? We need to figure out where the border between toom22 and
toom32 is (and hopefully, also rule out use of e.g., toom42 for the
recursive call),

Regards,
/Niels