Re: Windows performance

[Keith OHara]

Trevor Daniels <t.daniels <at> treda.co.uk> writes:

> >>> Phil Holmes wrote Thursday, April 16, 2015 2:00 PM
> >>>
> >>>> The performance of LilyPond 2.19.18 on Windows is _much_ better than
> >>>> previous versions.  Some examples:
> >>>>
> >>>> A 26 page multi-score piece I've been working on:
> >>>> 2.19.16: 114s to compile
> >>>> 2.19.18: 52s
> >>> ... [etc]

> 
> Further to this, I first established that the speed-up on Windows happened 
between 2.19.17 and 2.19.18. 
> One possibly relevant change between those two releases was the move from 
gs 8.70 to gs 9.15.  This
> increased the size of usr/bin from 49.2Mb to 58.1Mb, virtually all of the 
increase being in the three gs
> files, so this looks like a pretty significant upgrade.  Checking the 
GhostScript website for release 9
> suggests it includes performance improvements and major changes to font 
handling.  This looks a possible
> candidate, installed courtesy of Masamichi HOSODA-san.  If so, yet more 
thanks and kudos to you,
> Masamichi! But wouldn't this show a speed-up on systems other than Windows?
> 

I downloaded 2.19.18 (seeing this thread title and assuming there was 
another slowdown) and find it completes a score in 2/3 the time that
2.19.17 takes.   

None of our changes to LilyPond itself could explain this.

Ghostscript changed from 8.7 to 9.15, and Pango from 1.24 to 1.28,
between releases 2.19.17 and .18

The change in seconds is the same when I skip Ghostscript by using 
the --ps option on Lilypond.
On Linux, I see no measurable change (on music without lyrics).

Windows font lookups were very slow, causing a significant slowdown
once when we accidentally double the number of lookups.  The change
to the version of Pango is probably the cause of the speedup -- or 
maybe an underlying version change in fontforge; I'm not sure how 
fontforge is incorporated into LilyPond.