RFR: 8024927: Nashorn performance regression with CompressedOops
coleen.phillimore at oracle.com
Thu Oct 24 07:03:35 PDT 2013
On 10/24/2013 9:49 AM, Stefan Karlsson wrote:
>>> OK. Could you change the check to compare LogMinObjAlignmentInBytes
>>> to 3, or ObjAlignmentInBytes to 8, or OopEncodingHeapMax to 32GB?
>>> That would be much more clear too me.
>> I don't like constants in code, which is why I picked the symbolic
>> class alignment name for "3". I'll add a comment. How about that?
> But it's just wrong to use the symbolic class alignment
> (LogKlassAlignmentInBytes) here. What you need to check for is the
> heap size and not the klass alignments. Using
> LogMinObjAlignmentInBytes == LogKlassAlignmentInBytes makes the code
> hard for me (and probably others) to understand. It might even
> introduce bugs if someone changes LogKlassAlignmentInBytes to 4 and
> and uses 16 byte object aligment (yes, I know other places will break
You're right I want OopEncodingHeapMax to be 32*G which it might be
bigger with ObjectAlignmentInBytes. I'll just put 32*G since I can't
find a symbolic constant for it.
More information about the hotspot-dev