MethodHandle lookup&invocation performance

[Alexander Turner]

Remi,

OK - that is quite amazing!

Thanks for the explanation.

- AJ

On 10 July 2011 11:29, Rémi Forax <forax at univ-mlv.fr> wrote:
> On 07/10/2011 10:14 AM, Alexander Turner wrote:
>> John,
>>
>> Thanks for your fascinating reply. I would just like to clarify something:
>>
>> "
>> Another use case which benefits from this particular optimization
>> trick is "static final" method handles.  I.e., if you want to get
>> invokedynamic-like performance in plain Java code, consider doing your
>> work by invoking MH constants.  This means your bootstrap logic has to
>> run in the static initializer of the class containing the constants,
>> which is less flexible (more eager).  But it's an option.
>> "
>>
>> Is what you are saying here that making a method handle final static
>> will bring invoke performance closer to invokeExact because the
>> matching logic only happens once. IE the handle cannot be repatched? I
>> am not sure I fully understand this because (as I understand it)
>> making a field final does not do change the way fields in the
>> reference graph below the field behave. Does a mh 'know' is it final
>> and so not attempt to mutate itself or anything to which it refers?
>
> Currently the JIT tries to replace method handle calls
> by a direct call (or chain of calls) to after inline them.
> This occurs if the method handle is the target of a CallSite
> of an invokedynamic, if the method handle is stored
> in a final static field or if the method handle is a constant
> method handle loaded by a LDC.
>
> Basically, it works if the method handle is reachable from a static root.
>
> Now, if the method handle is reachable from a static root,
> them JIT knows its type (its method type) so because
>   mh.invoke(args) <=> mh.asType(methodType(args)).invokeExact(args)
> the JIT can fully inline mh.invoke(args).
>
>> On another note - it is great to keep hearing that so much work on
>> performance is in the pipeline!
>>
>> Best wishes - AJ
>
> cheers,
> Rémi
>
> _______________________________________________
> mlvm-dev mailing list
> mlvm-dev at openjdk.java.net
> http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
>



-- 
Dr Alexander J Turner
http://nerds-central.blogspot.com/2008/01/about-me.html