Re: Putting an end to "compiled closures"

[Ludovic Courtès]

Hello!

address@hidden (Ludovic Courtès) writes:

> The attached patch creates a new type tag, `scm_tc7_gsubr', whereby the
> 24 MSBs are used to store gsubr arity information as returned by
> `SCM_GSUBR_MAKTYPE ()'.  This makes cclos useless, which simplifies the
> code and reduces the overhead when creating and invoking such
> procedures.

I committed it:

  
http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commitdiff;h=e20d7001c3f7150400169fecb0bf0eefdf122fe2
  
http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commitdiff;h=54d14084e229f90b75475a866e3f458be30fa233

I also committed a simple benchmark for subr invocation [0].  The result
is as follows (truncated for clarity):

  - before patch

    ("subr.bm: subr invocation: simple subr" 700000 total 0.99)
    ("subr.bm: subr invocation: generic subr" 700000 total 1.77)
    ("subr.bm: subr application: simple subr" 700000 total 1.75)
    ("subr.bm: subr application: generic subr" 700000 total 2.31)

  - after patch

    ("subr.bm: subr invocation: simple subr" 700000 total 0.98)
    ("subr.bm: subr invocation: generic subr" 700000 total 1.38)
    ("subr.bm: subr application: simple subr" 700000 total 1.76)
    ("subr.bm: subr application: generic subr" 700000 total 2.16)

That is, roughly a 20% improvement on gsubr invocation.

Thanks,
Ludo'.

[0] 
http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commitdiff;h=b786a5bbf825f61e04ccd9a54f93cb1e40ac67d9