Re: [Sbcl-devel] Re: improving SBCL compiler speed, for money.

[David L. <da...@li...>]

Quoting Nikodemus Siivola (nik...@ra...):
>    - There are probably several sources of nasty-big-Oh lurking in the
>      compiler, but finding and eliminating them is a lot easier said
>      then done. Hard.

Christophe's patch for this is a little buggy but helps a lot for me (at
least together with some other little workarounds that might be specific
to the kind of extremely long code I am generating):

http://article.gmane.org/gmane.lisp.steel-bank.devel/3467/

(Why buggy?  SBCL compiles, but contribs don't build because slot-boundp
in asdf.lisp returns the wrong value, or something like that.  You would
want to fix that before enabling it unconditionally for a large code
base.)

>    - There should be a better SSET implementation sans bottlenecks
>      in the list archive. Effect unknown, but simplish to try at least.

Not really.  That was an exercise in AVL tree implementation, but as an
SSET it turned out to be rather slow.  The sorted-list implementation
has very good behaviour when things are inserted in the right order and
very low constant factors.


David