Our tuning code seems to give about the same cutoff from Toom 3 to Toom 4 as from Toom 3 to Toom 7.
>From gmp-mparam on k8. #define MUL_TOOM4_THRESHOLD 767 #define MUL_TOOM7_THRESHOLD 767 The cutoff for Windows was much lower ~ 200 if I recall correctly. Neither implementation has been particularly well optimised. Though we did a lot to optimise Toom 3. We are still working on optimising the assembly code used in our Toom 4. Presumably GMP is ahead in that regard, given their lower cutoff. At the time I wrote my notes, the cutoffs were much lower in MPIR (as indicated in my notes), but probably wrong. We later fixed our tuning code and changed the method of tuning. Bill. 2009/4/15 David Harvey <[email protected]>: > > > On Apr 15, 4:42 pm, William Stein <[email protected]> wrote: > >> You might find my notes from Bill's class right now interesting: >> >> http://wiki.wstein.org/09/583e >> >> Click on the "schedule" link. > > Hmmm, the notes are a bit hard to decipher, but I see things like > > "For Toom 7, cutoff is abut 190 limbs, and it always beats Toom 4/5" > > I find this a bit hard to believe. It suggests that the implementation > of toom 4 is inefficient. In GMP 4.3 on K8 the toom3 -> toom4 > threshold is about 400 limbs. You are seriously claiming toom 7 can > beat that at 190 limbs? What evaluation points are being used? > > david > > > > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en -~----------~----~----~----~------~----~------~--~---
