[Cado-nfs-discuss] Linalg: contexts size

Emmanuel Thomé Emmanuel.Thome at inria.fr
Mon Jun 29 17:23:46 CEST 2015


Hi Pierpaolo,

m=n=192 is not built in at the moment. The relevant code may be generated
from mpfq, but it's not trivial. I do it with
linalg/bwc/mpfq/Makefile.mpfq, but nobody beyond me has ever tried to run
this generation script, so what comes out of it is pretty unpredictable.

Furthermore, as Paul noticed, only 64 and 128 have ever been tested. The
fact that there's a bug with 256 is news to me (low priority though).

Last, for large block sizes the proper thing to do would be to provide
sse/avx/avx2 implementations. Currently it's not the case.

E.

On Mon, Jun 29, 2015 at 05:16:15PM +0200, Paul Zimmermann wrote:
>        Dear Pierpaolo,
> 
> > From: Pierpaolo Santucci <spierpa.net at gmail.com>
> > Date: Fri, 26 Jun 2015 19:37:03 +0200
> > 
> > Hello everyone!
> > 
> > Using Cado-NFS linear algebra, with m=n=192 and prime=2, I get the
> > following error:
> > 
> > Unsupported combination: group size = 192, p = 2
> > Unsupported combination: group size = 192, p = , 1 limbs
> > 2, 1 limbs
> > 
> > What are the supported combinations of m and n?
> > Can I use different sizes for right and left contexts?
> > 
> > Thanks a lot!
> > 
> > Pierpaolo Santucci
> 
> looking at the source code (linalg/bwc/mpfq/mpfq_vbase.c) it seems only 64,
> 128 and 256 are supported. But when trying n=256 it seems to fail with a
> different kind of error:
> 
> Cannot find the proper rebinder for data backend = u64k4 and matmul backend = bucket
> 
> All combinations of m,n in {64,128} seem to work on the c59 in README.
> {m,n} = {256,128} seems ok too.
> 
> {m,n} = {256,64} seems to loop forever in linalg.
> 
> 
> Best regards,
> Paul
> 
> _______________________________________________
> Cado-nfs-discuss mailing list
> Cado-nfs-discuss at lists.gforge.inria.fr
> http://lists.gforge.inria.fr/mailman/listinfo/cado-nfs-discuss


More information about the Cado-nfs-discuss mailing list