How high are he memory costs of polymorphic inline caches?
john.r.rose at oracle.com
Tue Aug 19 19:50:24 UTC 2014
Thanks for pointing that out, Mark.
As the JEP notes, Vladimir Ivanov is working on a number of approaches for LF reduction.
Caching similarly-shaped instances is relatively straightforward, although changing from our "full custom" approach is difficult.
The tricky part is to split just enough to avoid polluted profiles, especially the hot GWT nodes.
For the curious, some patches in progress may be found here:
Independently, Duncan McGregor has found some replicated DMHs that can be cache.
In the long term, our work in Valhalla to build frameworks for specializing code (shrinking Object refs to primitives and value types) should be applicable to method handles.
On Aug 19, 2014, at 12:30 PM, Mark Roos <mroos at roos.com> wrote:
> Related to the overhead of Lambda forms from Mark Reinhold on the Hotspot list.
> Looks like some efforts to improve them.
> New JEP Candidate: http://openjdk.java.net/jeps/210
More information about the mlvm-dev