<snip> > I have applied the patch and rebuild LinBox and started running the > test 500 times to see. Can you guess if/how much this patch does > affect performance for charpoly mod p? > For the dimensions you are considering (and up to a thousand) I don't expect any performance loss. But the probabilistic alg improves on larger matrices and gets asymptotically better (the best algorithm indeed!)
I'll let you know when I've made progress on this one. Clement >> mabshoff a écrit : >> >> >> >>> On Jul 17, 10:34 pm, mabshoff <[EMAIL PROTECTED]> wrote: >>>> Ok, here is what I found out last night: >>>> * 3.0.3 runs the test 200 times without failing it once >>>> * 3.0.4 with the new FLINT 1.0.13 fails 8 ought of 500 tests. >>>> So we are given a couple possibilities: >>>> * There is an algorithmic issue in ssmod somewhere or some >>>> algorithmic issue got exposed somehow in 3.0.4+ >>>> * There is an undiscovered bug in LinBox >>>> * There is an undiscovered bug in FLINT >>>> * none of the above >>>> * all of the above >>> After looking at the code William has conjectured that it is very >>> likely charpoly mod p that fails here. We update Linbox in 3.0.3- >>>> 3.0.4, so that fits the bill. This issue is now #3671. To debug this >>> we can compute the charpoly with LinBox and the generic code for a >>> large number of random inputs and compare. According to William the >>> speed difference between generic code and LinBox for the example in >>> ssmod (32 by 32 matrices) won't be too large. >>> If you have any (alternate) theories what goes wrong here please let >>> us know. >>>> Cheers, >>>> Michael >>> Cheers, >>> Michae > > > --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
