Re: Speed tips, again, for extremely large scores?

[Michael Käppler]

[Sorry, forgot to cc' the list]

Hi Trevor,
thanks for giving this discussion another try.
I'm also still very interested if there are possibilities to increase 
LilyPond performance further.
My machine is very slow, though and I cannot speak for many people when 
raising performance issues.
> In 2005 I put together a 64-staff score running 250 measures of music 
> (for untuned percussion). The LilyPond version was 2.7.x and the 
> compile time was 15 or 20 minutes on an Intel iMac running OS 
> 10.4.something with 1.5 GB ram and a 1.83 Ghz processor. Not super 
> fast, but certainly acceptable; the score compiled and I've been 
> distributing the resulting PDFs for years.
>
> Going back to that exact score *on the same machine* now hangs at 
> "Preprocessing graphical objects ...". I've let the process run to 
> more than an hour on two separate occasions with no end in sight. I'm 
> writing now to ask if there's a way to speed things up. But because 
> the question has been asked before, I'll first collect what I think 
> are the important parts from the archive and then add comments towards 
> the end.
I have also a file which I am unable to compile because it doesn't 
finish even in 12 hours and more. In the thread you mentioned I was 
pointed to my insufficient memory size. However, what I notice is that 
compiling time seems to increase in an exponential way with larger 
scores. Is this only a result of my bad system specs?

Three points:

1. It would be very interesting to create a "Lilypond benchmark suite" 
to help objectifying such questions as you, Trevor, have started.
In the nicest case it could be ran on different builds automatically to 
see whether there are huge changes between different releases.

2. I encounter a strange behaviour that nobody was able to explain up to 
now. Simply spoken: a lilypond foo1.ly foo2.ly foo3.ly foo4.ly is much 
>slower< than a lilypond foo1.ly && lilypond foo2.ly && lilypond 
foo3.ly && lilypond foo4.ly. Normally I would expect the opposite. (e.g. 
LilyPond does not need to load fonts several times ...)

3. I wonder if there is a debugging tool to show how much time Lilypond 
needs to perform this or that operation. I imagine something like:
System::fixup_refpoints (system.cc:231) ... 23 ms
blah blah

Regards,
Michael