RFR(M): 8212243: More gc interface tweaks for arraycopy and some other smaller changes in preparation for Shenandoah

Erik Österlund erik.osterlund at oracle.com
Thu Oct 18 17:03:34 UTC 2018

Hi Roland,

On 2018-10-18 17:44, Roland Westrelin wrote:
>> To determine that, I wonder exactly what barriers you need here at the
>> write. All I know is what barriers you do not know (no SATB, no
>> write-resolve).
> As I mentioned earlier, storing is:
> 1- b' = wb(b);
> 2- pre_barrier();
> 3- b'.f = wb(v);
> 1- is emitted at parse time when the arraycopy is parsed, 2- is not
> needed so wb(v) is what we need.

Okay. In that case I think I prefer Roman's idea of having an 
ACCESS_ARRAYCOPY decorator there in both the load and store to make the 


> Roland.

More information about the hotspot-compiler-dev mailing list