Re: Platform-dependent LLL reduction (qflll)

[Bill Allombert <Bill.Allombert@math.u-bordeaux.fr>]

On Mon, May 10, 2021 at 06:22:45PM +0000, mancha wrote:
> Hi.
> 
> While helping someone with their knapsack-crypto project on freenode's
> ##math channel, we discovered that LLL-reduced bases generated by 32-bit
> and 64-bit versions of PARI/GP 2.13.1 don't always match. Is this
> expected/anticipated behavior?

Yes. This is due to our implementation of fplll with PARI t_REAL.
The initial working precision is 32bit on 32bit and 64bit on 64bit,
so we get different result. In libpari it is possible to force
64bit in both case.

Note that in the GIT master branch we have an improved fplll
implementation that uses C doubles and is usually faster.
Unfortunately it is even more likely to give different results
on different platforms.

Cheers,
Bill.