[aarch64-port-dev ] RFR: backport volatile put/get and CAS optimization to JDK8

[Andrew Dinn]

On 13/10/15 16:19, Edward Nevill wrote:
> On Thu, 2015-10-08 at 17:11 +0100, Andrew Dinn wrote:
>> Ed, would you be able to check that jcstress works fine on other hardware?
>>
>> *** GC configs are
>>     -XX:+UseConcMarkSweepGC -XX:+UseCondCardMark
>>     -XX:+UseConcMarkSweepGC -XX:-UseCondCardMark
>>     -XX:+UseG1GC
>>     -XX:+UseParallelGC -XX:+UseCondCardMark
>>     -XX:+UseParallelGC -XX:-UseCondCardMark
> 
> All 5 configs tested fine on Cavium ThunderX.

All tests passed on my Mustang as well. So, I think we are ok to push this.

> For the -XX:+UseG1GC run I had to use the following patch as well
> 
> --- CUT HERE ---
> diff -r 2e58b602c89b src/share/vm/gc/g1/g1CodeCacheRemSet.cpp
> --- a/src/share/vm/gc/g1/g1CodeCacheRemSet.cpp	Tue Sep 08 14:08:58 2015 +0100
> +++ b/src/share/vm/gc/g1/g1CodeCacheRemSet.cpp	Tue Sep 15 08:06:31 2015 +0000
> @@ -197,7 +197,9 @@
>  }
>  
>  void G1CodeRootSet::allocate_small_table() {
> -  _table = new CodeRootSetTable(SmallSize);
> +  CodeRootSetTable* temp = new CodeRootSetTable(SmallSize);
> +
> +  OrderAccess::release_store_ptr(&_table, temp);
>  }
>  
>  void CodeRootSetTable::purge_list_append(CodeRootSetTable* table) {
> --- CUT HERE ---
> 
> Should this be backported as well.

Yes, I guess so, since g1 needs it. I'll push the upstream patch as well
as my patch so it doesn't cause any problems if we end up merging it
from upstream later.

> Your patch is for jdk8, are we developing on jdk8 or jdk8u now?

I just reviewed this with Andrew Haley and the story is as follows.

jdk8 is still being used as the development tree (yes, I know that's a
bit confusing modulo how it works upstream but nevertheless ...).

jdk8u will (soon) be set up to auto-merge upstream jdk8u updates daily.
That way we can identify merge issues separate from development. So,
this means that any changes which go into jdk8 should also -- at some
point -- be backported into jdk8u. I expect that most of the time that
will simply involve applying the same patch although, no doubt, in some
cases we may need to wield the crowbar.

regards,


Andrew Dinn
-----------