Re: scheme night-mare...

[Arno Waschk]

On Mon, 12 Jul 2010 22:23:04 +0200, Joe Neeman <address@hidden> wrote:

> On Monday, July 12, 2010 06:51:15 am Arno Waschk wrote:
> > On Mon, 12 Jul 2010 15:02:26 +0200, Carl Sorensen <address@hidden>
> >
> > wrote:
> > > On 7/12/10 4:48 AM, "Arno Waschk" <address@hidden> wrote:
> > >>> Just finished a profile run with a larger score- ly_scm2interval is
> > >>> reported to have consumd 16% of computation time. There must be
> > >>> something wrong.
> > >>
> > >> ... which appears in a loop, which is performed >2 billions times (!)
> > >> for
> > >> a 18 a3 page test score.
> > >> Says gprof...
> > >>
> > >> Is that possible/necessary?
> > >
> > > Well, if it's an 18 page score, then there are lots of page break
> > > options,
> > > and so there would be lots of calculations.   2 billion seems like a  
> > lot,
> > > but there are a lot of ways to figure out line and page breaks in 18
> > > pages
> > > of score....
> > >
> > > ly_scm2interval is part of the beam scoring code.  So that might be  
> > where
> > > many of the calls come from.
> >
> > no, from the beam thing it is only called ~1000 tmes, the two billion  
> > come
> >  from a loop in Axis_group_interface::combine_pure_heights (...)
>
> This sounds excessive; if we cache more sensibly, this should reduce by  
> at
> least a factor of 100. Does your score have many staves? Also, how many  
> times
> is Align_interface::get_pure_child_y_translation called?
>
> Cheers,
> Joe


Hi, the mentioned routine is called ~ 230000 times and does hardly consume  
time according to gprof.
it is a cello and piano score with quite lot systems squeezed onto the a3  
pages. it contains ~ 1600 bars, and breaks.size() is around 1200 IIRC

Yours, Arno