Request for reviews (M): 6614597: Performance variability in jvm2008 xml.validation

John Rose John.Rose at Sun.COM
Fri Jan 22 19:47:20 PST 2010

On Jan 22, 2010, at 6:33 PM, Tom Rodriguez wrote:

> ...The compiler needs to be able to distinguish failure of the mono case from failure of the bi case.  If you want to smear it to a different counter then why not use a counter which isn't used at call sites at all like Reason_range_check?

Good point.

> For that matter why can't we does Reason_none get a counter?  If we corrected the limits of the per bci counter then we could really give Reason_bimorphic it's own per bci counter.

Another good point.  Let's fix the allocations so we can use all the bits:

    Reason_RECORDED_FIRST = Reason_null_check   // first reason to be recorded per bc
    Reason_RECORDED_LIMIT = Reason_unloaded   // first reason after that not to be recorded per bc


-- John

More information about the hotspot-compiler-dev mailing list