Bill Allombert on Tue, 05 Sep 2017 22:43:03 +0200

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: Speed up RgX_mul, RgX_sqr, RgX_divrem over Z, Fp, FpXQ

On Tue, Aug 29, 2017 at 03:28:30PM +0200, Peter Bruin wrote:
> Hi Bill,
> One more thing: starting with commit 4a80b98, multiplication of FpX
> (which moved from RgX_mul/RgX_sqr to the new functions RgX_mul_FpX/
> RgX_sqr_FpX in that commit) removes the special case for Flx that was
> introduced in commit b770112.  Was this intentional?

Yes, sorry, I forgot to mention it.

>  It does make a
> noticeable difference, see below.  One way to fix this is as in the
> attached patch; alternatively, one could of course try to speed up
> FpX_mul/FpX_sqr.

Normally FpX functions are calling Flx function for small p but this
is not done for FpX_mul.

However relying on that for RgX functions is not entirely satisfying
because this leads to a double conversion RgX->FpX->Flx
Thanks to your patch a direct conversion RgX->Flx occurs.

But it makes sense for FpX_mul to call Flx_mul when possible
for consistency.