PS: On 18 Jul., 18:22, Simon King <simon.k...@uni-jena.de> wrote: > ... > 3rd test: Multiplication of two 2000x2000 matrices over GF(125). > Default multiplication takes ages. > New Strassen implementation, cutoff=20: 427 seconds using 6.9% of my > computer's memory. > Old Strassen implementation, cutoff=20: 431 seconds using 9.3% of my > computer's memory. > > I don't know if it is worth it. Shall I open a ticket and post my > patch?
Concerning "worth it": Even without Strassen multiplication, MeatAxe only needs about 12 seconds for that example. So, one could argue that it is not worth it: One easily gets a better speed in a different way. Or one could argue that it is worth it: If one uses sage.matrix.strassen on top of MeatAxe then the speed might be even better (perhaps close to Magma speed). By the way, the MeatAxe version I started with is 2.2.4 -- it was a special release under GPL 2+ (otherwise identical to 2.2.3, which is GPL2 only). So, the licence problem might be solvable. Or one could argue that it is not worth it: LinBox already has Strassen. So, it might be better if Burcin and Martin revived the LinBox wrapper (so that we have full speed over finite prime fields) and then use the tricks discussed earlier in this thread to extend it to non-prime fields (which would be quite a lot of work). Anyway, I think Strassen multiplication should become the default for Matrix_modn_dense. What do people think? Cheers, 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
