Hi, LinBox uses the BLAS for dense matrices mod p where p <= 2^23, hence it would benefit from advances in that direction. There is also some (experimental?) code for dealing with non-prime fields in LinBox. Furthermore, there is some code - AFAIK not in LinBox yet - for packing multiple primes into one double which would be relevant for very small primes. Combining this with the prime-slicing idea should be quite fast. On Nov 30, 2011 2:29 PM, "Dima Pasechnik" <dimp...@gmail.com> wrote:
> I might get blamed for making discouraging remarks, but let me play the > devil's advocate: > > I wonder if these kinds of speed-ups are to be beaten, soon, by > sufficiently fast hardware implementations of level 2 and 3 BLAS, coupled > with some crude use of their super-fast vector arithmetic. > Assuming most CPUs will get vector processing commands in the near future > (GPUs are already there), it seems to be more important to see how level 2 > and 3 BLAS can be efficiently called from Sage, without the numpy layer, > which is targeted towards numerics, and (IMHO) lacks the appropriate > interfaces. > > -- > To post to this group, send an email to sage-devel@googlegroups.com > To unsubscribe from this group, send an email to > sage-devel+unsubscr...@googlegroups.com > For more options, visit this group at > http://groups.google.com/group/sage-devel > URL: http://www.sagemath.org > -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org