Hi!

On 12 Jul., 23:35, Burcin Erocal <bur...@erocal.org> wrote:
> On Tue, 12 Jul 2011 14:12:18 -0700 (PDT)
> At SD16, Clement and I looked into wrapping linbox directly. The speeds
> were promising, but we never took the time to fix pickling, etc.


I, for one, would appreciate fast matrix multiplication over finite
fields!

However, there is one complication: In my current project (computing
Ext-Algebras of basic algebras), I definitely need finite non-prime
fields, and the current implementation in Sage for that case is
definitely too slow. I just did an example of multiplying two 500x500
matrices over GF(125). Sage took 11.24 seconds, MeatAxe (with some
improvements that I implemented yesterday) took 0.2 seconds.

If LinBox really can not "officially" deal with non-prime fields, I
should probably stick with MeatAxe then (even though, in the small
version, it only supports fields of size <256).

My current MeatAxe wrapper (in my optional cohomology spkg) is based
on an old MeatAxe version (2.2.4). Perhaps I could take the most
recent version (2.4.13), modify my wrapper accordingly, and contribute
it to Sage. The MeatAxe licence is good enough for Sage, isn't it? But
is MeatAxe eligible to become a standard package? Currently, there
only is an experimental meataxe-2.4.3.spkg.

Correct me if the following is nonsense:
- The MeatAxe sources should go into an spkg. Question: How should I
deal with my modifications to the original MeatAxe sources? Do I
remember correctly that the spkg should contain the unmodified
sources, together with some automatically applied patches that are
stored in a sub-directory patches/?
- The spkg should create a library mtx.a, that should go into, say,
SAGE_ROOT/local/lib/mtx-2.4.13/, so that my wrapper can easily link
against it.

Best regards,
Simon

-- 
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

Reply via email to