Request for review:7133260:Remove -Xaprof and the usage of Klass:_alloc_count and ArrayKlass::_alloc_size
jiangli.zhou at oracle.com
Wed Jun 5 13:29:42 PDT 2013
Here is updated webrev with object_iterate_since_last_GC and
On 06/05/2013 09:44 AM, Jiangli Zhou wrote:
> Hi Stefan,
> On 06/05/2013 01:48 AM, Stefan Karlsson wrote:
>> On 06/05/2013 04:15 AM, Jiangli Zhou wrote:
>>> Please review following change for 7133260.
>> I looked at:
> Thanks for doing that! Sorry for the wrong link.
>> Looks good. Thanks for doing this cleanup.
>> You remove the last usage of:
>> so there's even more cleanups that can be done here.
> That does appear to be the case. Thanks for noticing that. I'll try to
> remove them as well.
>>> In JDK8, -Xaprof support is outdated and broken. A new feature,
>>> GC.class_stats command has been added to jcmd in JDK8. It provides
>>> detailed class metadata memory statistics information. -Xaprof uses
>>> Klass:_alloc_count and ArrayKlass::_alloc_size. Removing -Xaprof and
>>> related fields would save 4bytes per class and 8bytes per array class.
>> FTR, I don't think you save 8 bytes for the array classes. The size
>> of ArrayKlass increased so that the vtable offset matches the vtable
>> offset of the InstanceKlasses. See ArrayKlass::static_size.
> Thanks for pointing out the ArrayKlass::static_size. You are right,
> the array class size is increased to be the same as InstanceKlass size
> in ArrayKlass::static_size. I will fix CCC. So removing fields from
> ArrayKlass does not yield real saving. :( It also means any size
> reduction in InstanceKlass would help array class. :) Any reason why
> ArrayKlass and InstanceKlass vtable offset need to be matched?
> Thanks for the review!
>>> Please see more information in the CCC request:
>>> Tested with JPTR and vm.quick.testlist.
More information about the hotspot-runtime-dev