Re: OMG - performance issue with Scheme engraver

[Aaron Hill]

On 2018-07-10 02:56, Urs Liska wrote:
> The situation is that a three-page score that takes ~ 2.1 seconds
> without the engraver and with the old state of the engraver now needs
> 5.8 seconds. And a >20 page score that I recall using around 20-25
> seconds now needs 57 - so that's obviously nothing to ignore.

On 2018-07-10 03:22, David Kastrup wrote:
> Quadratic behavior or worse.

Based on the durations posted, it seems much more linear, as 5.8/2.1 is 
close enough to 57/22.5.  If it were quadratic, then one would have 
expected the second test case to take closer to 570 seconds.  As it 
stands, the original code was taking about one second per page and now 
is taking around two seconds per page.

So it sounds like the modification is roughly doubling the amount of 
work needed overall, but this extra work is more-or-less fixed and not 
dependent on the size of the problem.  This might help narrow the focus 
by looking at things that have more of a global impact, independent of 
the amount of markup.

Doing a quick web search turned up Statprof, but sadly that was 
introduced with Guile 2.0.  And nothing much else turned up regarding 
general purpose profiling tools for Guile other than this section in the 
docs:

https://www.gnu.org/software/guile/docs/docs-1.8/guile-ref/Profiling-or-Tracing-a-Procedure_0027s-Code.html

I did find a thread on LilyPond performance from over a year ago when it 
looks like folks were evaluating the newer version of Guile as it was 
resulting in significant slowdown.  So, I would have suggested trying a 
branch of LilyPond that runs against Guile 2.0 just to be able to make 
use of the profiling tools; but that sounds like it would be adding more 
muck to the mix.

-- Aaron Hill