Another paper on invokedynamic

[Fredrik Öhrström]

Interesting paper. It is good to have a test to verify that
Direct MethodHandles installed into the CallSites should
have (almost) identical performance as the original code.
Obviously we are not quite there yet.

I do think it is any use in making super() link very late.
It would only make sense if you could change the
inheritance hierarchy very late, which you can't.

//Fredrik

2010/9/27 Chanwit Kaewkasi <chanwit at gmail.com>

> Dear all,
>
> This is another paper on invokedynamic (second to John Rose's :)) to
> be appeared in VMIL '10:
>
> Abstract
> ========
> This paper presents a study of a Java Virtual Machine prototype
> from the Da Vinci Machine project, defined by JSR 292. It
> describes binary translation techniques to prepare benchmarks to
> run on the invokedynamic mode of the prototype, resulting
> in the  invokedynamic version of the SciMark 2.0 suite.
> Benchmark preparation techniques presented in this paper are
> proven to be useful as the  invokedynamic version of
> benchmark programs successfully identified strange slowness
> behavior of the  invokedynamic mode of the server virtual
> machine.
>
> Surprisingly, benchmarking results show that the  invoke-
> dynamic mode with direct method handles on the server
> virtual machine is just 2-5 times slower than native Java
> invocations, except the Monte Carlo benchmark. But this mode
> on the client virtual machine still requires further performance
> tuning.
> ========
>
> I hope you find it useful somehow and any feedback is very welcome!
>
> http://dl.dropbox.com/u/381580/vmil10_kaewkasi_revised.pdf
>
> Best regards,
>
> Chanwit
> _______________________________________________
> mlvm-dev mailing list
> mlvm-dev at openjdk.java.net
> http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/mlvm-dev/attachments/20100927/6c428cb4/attachment.html