On Wednesday 02 May 2007 23:03, Arvid Grøtting wrote: > 2007/5/2, Joe Neeman <[EMAIL PROTECTED]>: > > With my working copy of lilypond (on a 1.6 GHz Opteron), I have this down > > to 3m22.321s. That's still about 3 times slower than 2.8 but it's > > progress. FWIW, I don't think 2.10 (and above) will ever be as fast as > > 2.8 when there are many scores -- the newer versions do a lot more work > > in the page-breaking stage and there is a limit to how much it can be > > optimised. > > Personally, I'm not at all worried about a constant-factor increase in > running time. The trouble, at least with 2.11.23, is that the running > time scales very badly with the number of scores (or possibly more > generally with the total length of the music). > > As I said, 26m5.898s user time for the problematic file (here), with > 29 score blocks. With roughly half the score blocks commented out, > I'm down to 2m44.806s user time for 14 score blocks. Halve again, and > I'm at 0m47.255s for 7 score blocks. A single (the first) score block > clocks in at 0m3.020s user time. > > Something tells me this is a bit worse than O(n).
With the optimisations, it scales like (left column is the number of scores) 24 2m0.017s 19 1m23.650s 14 0m52.611s 9 0m27.961s 4 0m11.139s 1 0m4.402s as you can see, it isn't O(n) but it isn't much worse either. The further optimisations I have planned should make it even closer to linear. _______________________________________________ bug-lilypond mailing list bug-lilypond@gnu.org http://lists.gnu.org/mailman/listinfo/bug-lilypond
