Re: indirect threading for bytecode interpreter

[Stefan Monnier <monnier@iro.umontreal.ca>]

>>> I've profiled (using oprofile) Emacs running this a few times;
>>> typical results show mark_object at 10-12% of runtime.  It is
>>> probably pretty easy to reproduce this with a variety of elisp
>>> programs.
>> So no matter how hard we try, we won't gain more than about 10%
>> speed-up trying to optimize the GC.  Doesn't sound like a great
>> motivator.
> The 10% occur in bursts of unresponsiveness.  If such a burst is halved
> in size, the total runtime improvement will not be impressive, but the
> responsiveness gets quite better.

That is true.  To help solve this, we should aim to move towards
a concurrent GC (i.e. a GC that's actually likely to be slower, but
won't cause unresponsiveness).


        Stefan