I definitely want to use BLAS when it is appropriate. The reason is that GOTO BLAS has been open sourced, under a BSD style license. So it is now trivial to install an extremely high performance BLAS (takes ~ 90s to install on my machine). Rather than use bits and pieces of GOTO, I think it is better if we just make it a dependency.
Of course, one BLAS is as good as another, so Sage may not choose to use GOTO as such, but that shouldn't affect FLINT. Bill. On Jan 17, 1:12 pm, Fredrik Johansson <fredrik.johans...@gmail.com> wrote: > On Mon, Jan 17, 2011 at 12:13 PM, Bill Hart <goodwillh...@googlemail.com> > wrote: > > But I think Fredrik had some figures which show we beat Sage. > > Yes, even for small primes. Multiplying two 1000x1000 matrices mod p = > 17 takes 2.27 seconds in Sage and 1.05 seconds with FLINT 2's > nmod_mat's. The corresponding time in FLINT 2 for a 50-bit prime is > 1.6 seconds, and 2.6 seconds for a full 64-bit prime (these would > currently be extremely slow in Sage). This is just on my computer; > results elsewhere may be entirely different. > > We can probably improve nmod_mat performance a bit more. Ultimately, > BLAS will likely be faster for moduli in the appropriate range, > however. > > Fredrik -- 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
