RFR (S) JDK-8139922 Get rid of dead code in ConcurrentMark

Thomas Schatzl thomas.schatzl at oracle.com
Wed Nov 18 15:30:35 UTC 2015


Hi,

On Wed, 2015-11-18 at 07:16 -0800, Jon Masamitsu wrote:
> 
> On 11/18/2015 02:31 AM, Thomas Schatzl wrote:
> > Hi,
> >
> > On Tue, 2015-11-17 at 13:51 -0800, Jon Masamitsu wrote:
> >> Anyone think that parClear() is worth saving?  If just for
> >> symmetry with parMark()?
> >    I would prefer to remove unused code. The CMBitmap class does not feel
> > like being in any way reusable but a thin wrapper for use with G1
> > concurrent marking. This method would be simple to recreate too.
> 
> Yes, it would be easy to recreate and I have no problem with
> removing it.  I actually was surprised that it was not being
> used.  I have a recollection that clearing the bit maps after
> a full GC on a gigantic heap was noticeable.  But if we need
> it, as you say, easily recreated.

  I think we do that in parallel without any synchronization on the
bitmap, splitting the whole bitmaps by region. Within the bitmap
corresponding to a single region, there is no kind of work stealing or
helping.

It would also be very costly to clear the whole bitmap bit by bit (which
is what the method does).

Thanks,
  Thomas




More information about the hotspot-gc-dev mailing list