Caches pure translations for full score.

Caches pure translations for full score.

[MikeSol, 2011-11-05 12:37:28]

Hey all,

A small patch to speed up LilyPond for large scores.  Some benchmarks:

\version "2.15.17"
foo = \relative c'' {
  \repeat unfold 1200 {
    <a b cis e>8-.^"foo"\pp^\espressivo\glissando[
    ( <a b cis e>-.^"foo"\p^\espressivo )]   
  }
} 
  
<<
\new Staff \foo
>>

CURRENT MASTER
real	0m36.986s
user 0m33.366s

WITH PATCH
real	0m36.393s
user 0m32.262s

\version "2.15.17"
foo = \relative c'' {
  \repeat unfold 1200 {
    <a b cis e>8-.^"foo"\pp^\espressivo\glissando[
    ( <a b cis e>-.^"foo"\p^\espressivo )]   
  }
} 
  
<<
\new Staff \foo
\new Staff \foo
\new Staff \foo
\new Staff \foo
\new Staff \foo
\new Staff \foo
>>

CURRENT MASTER
real	5m29.722s
user 5m17.332s
sys 0m4.544s

WITH PATCH
real	3m30.123s
user 3m21.257s

[MikeSol, 2011-11-05 15:45:31]

Just tried this out on a real score - the 4th movement of the Eroica on mutopia.

CURRENT MASTER

real	4m30.352s
user	4m21.088s

WITH PATCH

real	2m4.457s
user	2m0.148s

w00t!

[t.daniels_treda.co.uk, 2011-11-05 16:43:26]

mtsolo@gmail.com wrote Saturday, November 05, 2011 3:45 PM


> Just tried this out on a real score - the 4th movement of the 
> Eroica on
> mutopia.
>
> CURRENT MASTER
>
> real 4m30.352s
> user 4m21.088s
>
> WITH PATCH
>
> real 2m4.457s
> user 2m0.148s
>
> w00t!

W00t indeed!  That's some improvement!

(I don't know how well that translates from the Yorkshire expression 
when written out - it means "That's a tremendous improvement!", 
expressed with surprised admiration. :)

Getting a 10% improvement in speed would be remarkable; getting a 
speed-up of 125% is incredible!

Trevor

[mike_apollinemike.com, 2011-11-05 18:04:23]

On Nov 5, 2011, at 9:43 AM, Trevor Daniels wrote:

> 
> mtsolo@gmail.com wrote Saturday, November 05, 2011 3:45 PM
> 
> 
>> Just tried this out on a real score - the 4th movement of the Eroica on
>> mutopia.
>> 
>> CURRENT MASTER
>> 
>> real 4m30.352s
>> user 4m21.088s
>> 
>> WITH PATCH
>> 
>> real 2m4.457s
>> user 2m0.148s
>> 
>> w00t!
> 
> W00t indeed!  That's some improvement!
> 
> (I don't know how well that translates from the Yorkshire expression when
written out - it means "That's a tremendous improvement!", expressed with
surprised admiration. :)
> 
> Getting a 10% improvement in speed would be remarkable; getting a speed-up of
125% is incredible!
> 
> Trevor
> 

Thanks!  I haven't done exhaustive testing, but I'm guessing that the +100% is a
best-case scenario that only applies to large scores with several
VerticalAxisGroups.  I tried it on a few page long snippets w/ one system and it
shaves off around 0.03 seconds from a 1.5 second compilation.

Vive l'optimisation,
~Mike

[joeneeman, 2011-11-05 20:36:31]

looks good! I never realized this case was such a bottleneck.

http://codereview.appspot.com/5349043/diff/1/lily/align-interface.cc
File lily/align-interface.cc (right):

http://codereview.appspot.com/5349043/diff/1/lily/align-interface.cc#newcode320
lily/align-interface.cc:320: SCM l = SCM_EOL;
Given that you've gone to the trouble of writing float_list_to_vector, maybe you
could also write float_vector_to_list. And I think they belong somewhere more
public, too.

[Keith, 2011-11-05 22:52:21]

This looks very nice.

http://codereview.appspot.com/5349043/diff/1005/lily/align-interface.cc
File lily/align-interface.cc (right):

http://codereview.appspot.com/5349043/diff/1005/lily/align-interface.cc#newco...
lily/align-interface.cc:327: vector<Real> translates = start == 0 && end ==
INT_MAX
This happens to be exactly the condition for which I wanted to use natural
spacing rather than compressed spacing
 <http://codereview.appspot.com/5354043/>
in order to make my beloved skyline-vertical-padding work better.  I might just
hook into this, later.

[pkx166h, 2011-11-06 08:32:31]

Passes make and no reg test diffs

James