Reader performance regression

Reader performance regression

[Ludovic Courtès]

Hello,

A profile of ‘(compile-file "boot-9.scm")’ shows this:

  %     cumulative   self             
  time   seconds     seconds      name
   48.15      0.91      0.91  read
    8.15      0.15      0.15  #{\search-vector-rib\\\ 1944}#
    6.67      0.13      0.13  make-struct-layout
    5.93      1.88      0.11  eval
    ...

IOW, ‘read’ performs badly.

Commit 889975e51accb80491af76fc5db980aeb3edd342 may have been a
performance regression as it did away with the optimization that
consisted in using stack-allocated buffers in the common case, rather
than heap-allocated temporaries [0, 1].  (Indeed, Guile-Reader, whose
code is close to that of the pre-Unicode ‘scm_read ()’, is 1.4x faster
than current ‘read’.)

Mike: would you like to try to reintroduce that optimization?  :-)
What do you think?

Thanks,
Ludo’.

[0] Commit 7337d56d5714227865aeca2b40b6bd97cce296d2 (2007-07-22).
[1] http://lists.gnu.org/archive/html/guile-devel/2007-07/msg00013.html

Re: Reader performance regression

[Mike Gran]

> From: Ludovic Courtès <ludo@gnu.org>

> Hello,
> 

...

> IOW, ‘read’ performs badly.
> 
> Commit 889975e51accb80491af76fc5db980aeb3edd342 may have been a
> performance regression as it did away with the optimization that
> consisted in using stack-allocated buffers in the common case, rather
> than heap-allocated temporaries [0, 1].  (Indeed, Guile-Reader, whose
> code is close to that of the pre-Unicode ‘scm_read ()’, is 1.4x faster
> than current ‘read’.)
> 
> Mike: would you like to try to reintroduce that optimization?  :-)
> What do you think?

I'll see what I can do.

Thanks,

Mike