Migrate inheritance-based cache compiler to use arrays for more stuff
---------------------------------------------------------------------
Key: JRUBY-2807
URL: http://jira.codehaus.org/browse/JRUBY-2807
Project: JRuby
Issue Type: Bug
Components: Compiler
Reporter: Charles Oliver Nutter
Assignee: Charles Oliver Nutter
The current caching logic in the compiler leans on the fact that the superclass
of a Script, AbstractScript, defines a bunch of fields to use for caching
fixnums, symbols, regexps, blocks, and so on. Unfortunately this also wastes
memory for ever Script instance instantiated, since regardless of whether they
*need* 50 call site caching slots, they'll take up 4 bytes each in the object.
So that's 200 bytes for call site references, 200 bytes for symbol references,
and 200 bytes for fixnum references; 600 bytes of memory per instantiated
Script (including jitted methods, even the tiniest ones) that might never be
filled or used.
The inherited cache logic was originally added to reduce the amount of permgen
taken up by all those generated classes, since defining new fields adds new
metadata which eats up permgen memory on at least HotSpot. But it seems this
move may have traded some amount of permgen for more heap memory, and we could
certainly do better to reduce them.
It would be interesting to see what the actual in-memory size of Script
subclasses is and whether this inefficiency has a real impact on heap usage. At
any rate, it ought to be addressed.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.codehaus.org/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
