Re: indirect threading for bytecode interpreter

[Dan Nicolaescu <dann@ics.uci.edu>]

Tom Tromey <tromey@redhat.com> writes:

  > >>>>> "Dan" == Dan Nicolaescu <dann@ics.uci.edu> writes:
  > 
  > Dan> The overhead of mark_object should be lower in that version (it's
  > Dan> higher in 23 because of the way charsets are represented).
  > 
  > I would be interested if you, or anybody, has ideas on how to improve
  > the performance of mark_object.  That is, without a whole GC rewrite :-)

Not sure if much can be done about mark_object itself, the issue is
the amount of things that need to be marked.

Making the GC generational might help.  The first GC after startup does
200K mark_object calls, so if most of those don't need to be scanned for
each GC, then that might help.

If you have a testcase that shows GC problems, maybe posting the test
case would enticemore people look into it...