RFR 8049105: Move array component mirror to instance of java/lang/Class (hotspot part 2)

[Coleen Phillimore]

Thanks, Mikael!

On 8/28/14, 2:21 AM, Mikael Gerdin wrote:
> Coleen,
>
> On 08/26/2014 03:56 PM, Coleen Phillimore wrote:
>> Summary: This removes component mirrors from arrayKlass metadata and the
>> C2 intrinsic for them.
>>
>> I'm sending this out to hotspot-dev so I can get someone from the
>> compiler to review this (again).  Chris Thalinger, you already saw this
>> once, it's the same thing, but I had to hold this change until it was
>> okay'd by licensees.   Can you re-review?
>>
>> Also GC team note that Klass::oops_do() is now non-virtual. Will that
>> help?
>
> It may. But the oops_do code is riddled with closures and indirect 
> calls through callbacks so I don't think it will gain us any 
> significant performance boost.
>

Too bad.

Thanks for the code review.

Coleen
>>
>> Tested with jck, nsk and jtreg tests.  Also, ran refworkload with no
>> difference in performance.
>>
>> open webrev at http://cr.openjdk.java.net/~coleenp/8049105/
>
> Looks good to me too.
> /Mikael
>
>> bug link https://bugs.openjdk.java.net/browse/JDK-8049105
>>
>> Thanks,
>> Coleen
>