[viff-devel] PRSS zero- and double-sharings

[Martin Geisler]

Martin Geisler <mg at daimi.au.dk> writes:

> Strangely the time for preprocessing has not improved... It stayed
> at an average time of about *20 ms* for a multiplication triple both
> before and after the change -- I don't understand that :-(

I do now! :-)

It turned out that the preprocessing was still done with the old code.
Using the new PRSS-based code brings the time down to *5 ms* pr
triple, so it cuts the time by a factor of four.

Thanks to Mikkel for making me think about why there were matrix
multiplications in the profile information...

-- 
Martin Geisler