RFR: JDK-8153203: Remove liveRange.hpp

Bengt Rutisson bengt.rutisson at oracle.com
Thu Mar 31 15:46:28 UTC 2016


Hi everyone,

Could I have a couple of reviews for this change?

http://cr.openjdk.java.net/~brutisso/8153203/webrev.00/
https://bugs.openjdk.java.net/browse/JDK-8153203

The methods CompactibleSpace::scan_and_forward() and 
PSMarkSweepDecorator::precompact() are more or less copy-paste versions 
of each other. Both these use the LiveRange class.

The LiveRange class is used to write to the mark word of a dead object. 
But no one really cares that the LiveRange class is used. Instead it 
just gives an extra level of indirection to already complicated code. We 
also do unnecessary work to keep track of the end of the range even 
though no one ever gets the end value.

We also do duplicate stores to _first_dead in these methods.

The methods that consume the values from the LiveRange 
(CompactibleSpace::scan_and_adjust_pointers() and 
PSMarkSweepDecorator::adjust_pointers()) don't use the LiveRange class. 
Instead they use oop()->mark()->decode_pointer(), which is kind of odd 
considering that this is normally used for forwarded objects.

The code would be simpler if we just store and load directly from the 
memory addresses we are working with.

Improving the assert at the end of the 
CompactibleSpace::scan_and_adjust_pointers() and 
PSMarkSweepDecorator::adjust_pointers() methods to log the values of q 
and prev_q will hopefully improve the chances of understanding 
JDK-8073321 if that ever happens again.

Thanks,
Bengt


More information about the hotspot-gc-dev mailing list