Request for review: 8003848: Make ConstMethod::generic_signature_index optional and move Method::_max_stack to ConstMethod
Jiangli Zhou
jiangli.zhou at oracle.com
Tue Nov 27 11:31:56 PST 2012
Hi Coleen,
Thanks. Will do.
Jiangli
On 11/27/2012 10:37 AM, Coleen Phillimore wrote:
>
> Hi Jiangli,
>
> On 11/27/2012 11:06 AM, Jiangli Zhou wrote:
>> Hi Coleen,
>>
>> I remember now you suggested moving those fields as well when I
>> proposed the change. I'll move max_locals and size_of_parameters into
>> constMethod. Would it be okay if I push this as is and work on the
>> other two fields on the same time?
>
> Sure, that is fine. Can you file a bug for it? I don't think there
> is one yet.
> Thanks,
> Coleen
>
>>
>> Thanks!
>> Jiangli
>>
>> On 11/26/2012 07:35 PM, Coleen Phillimore wrote:
>>>
>>> Jiangli,
>>>
>>> This code looks fine, although I don't like that max_stack is being
>>> moved without moving max_locals. They're usually used together in
>>> the same place. If you moved max_locals and size_of_parameters
>>> also with this change into constMethod, I think it would be
>>> better. Also, we want to move these fields anyway for more sharing
>>> with CDS.
>>>
>>> Thanks,
>>> Coleen
>>>
>>> On 11/26/2012 9:08 PM, serguei.spitsyn at oracle.com wrote:
>>>> Jiangli,
>>>>
>>>> It looks good.
>>>>
>>>> Thanks,
>>>> Serguei
>>>>
>>>> On 11/21/12 11:53 AM, Jiangli Zhou wrote:
>>>>> Hi,
>>>>>
>>>>> Please review the following memory reduction change for Method and
>>>>> ConstMethod:
>>>>>
>>>>> http://cr.openjdk.java.net/~jiangli/8003848/webrev/
>>>>>
>>>>> Majority of java methods do not have generic signature. For those
>>>>> methods, ConstMethod::generic_signature_index are always 0.
>>>>> ConstMethod::generic_signature_index can be optional and only
>>>>> allocated for methods with generic signature.
>>>>>
>>>>> The Method class has 2-byte unused due to padding for alignment.
>>>>> Method::_max_stack can be moved to ConstMethod after changing
>>>>> ConstMethod::generic_signature_index to be optionally allocated.
>>>>> This removes an u2 field in Method, and can avoid the 2-byte
>>>>> unused space waste due to padding.
>>>>>
>>>>> Together, above would save 4-bytes for each java method without
>>>>> generic signature.
>>>>>
>>>>> Tested with jprt, runthese and vm.quick.testlist. No measurable
>>>>> performance difference with the change for specjvm98 and specjbb2005.
>>>>>
>>>>> ==============================================================================
>>>>>
>>>>> logs.method_baseline.jbb:
>>>>> Benchmark Samples Mean Stdev Geomean Weight
>>>>> specjbb2005 3 56252.53 284.79
>>>>> ==============================================================================
>>>>>
>>>>> logs.method.jbb.1:
>>>>> Benchmark Samples Mean Stdev %Diff P
>>>>> Significant
>>>>> specjbb2005 3 56350.15 626.79 0.17
>>>>> 0.823 *
>>>>> ==============================================================================
>>>>>
>>>>>
>>>>> ==============================================================================
>>>>>
>>>>> logs.method_baseline.spec:
>>>>> Benchmark Samples Mean Stdev Geomean Weight
>>>>> specjvm98 8 543.39 11.93
>>>>> ==============================================================================
>>>>>
>>>>> logs.method.spec.1:
>>>>> Benchmark Samples Mean Stdev %Diff P
>>>>> Significant
>>>>> specjvm98 8 544.21 9.56 0.15
>>>>> 0.881 *
>>>>> ==============================================================================
>>>>>
>>>>>
>>>>>
>>>>> Thanks,
>>>>> Jiangli
>>>>
>>>
>>
>
More information about the hotspot-runtime-dev
mailing list