Re: Array marking

[Bryce McKinlay <bryce at albatross dot co dot nz>]

Tom Tromey wrote:

> Maybe we're doing it to ensure the array's class gets marked.
> Right now I think (heap-allocated) vtables aren't scanned at all,
> so we have to mark the class explicitly.  We could fix this, though.
>
> We really do need nicer GC interfaces... I have some stuff here on
> that to look at, but I haven't been able to dig up the time.
>
> ISTR you have a GC benchmark of some sort.
> If you want to run it both ways (hacking out the array marker is
> probably easy), that would be cool.

I did that, and I got the bitmap-marking-descriptor stuff working.
Together they seem to be worth some pretty serious performance gains -
some of my gc tests are running 40 - 50% faster now, and ShellSort is now
twice as fast as the IBM JDK (using --no-bounds-check).

I'm getting some strange crashes though which look like they might be
vtables getting freed or some such weirdness. I'm investigating...

regards

  [ bryce ]