review request (S) - 6821693: TaskQueue capacity

John Coomes John.Coomes at sun.com
Mon Aug 3 17:52:57 PDT 2009


I (John.Coomes at sun.com) wrote:
> I'd appreciate some reviews of 
> 
> http://cr.openjdk.java.net/~jcoomes/6821693-taskqueue64/
> 
> 	6821693 64-bit TaskQueue capacity still too small
> 	6821507 Alignment problem in GC taskqueue
> 
> Those w/good memories may recall an earlier review request, but enough
> code has changed and time has passed to require new reviews.
> 
> A micro-test that pushes/pops items on/off the queue is ~10% faster
> w/the new code on intel, but there's no noticeable effect on
> real-world apps.

I've updated the webrev (at the location above) based on review
comments from Tony Printezis and Andrey Petrusenko (thanks).

Changes:

	- use upper case for constant names
	- fix two comment typos
	- declare the TaskQueueSuper::_age field volatile and make
	  all the necessary accessor methods volatile.  This
	  eliminates the need for separate top() and top_volatile()
	  methods.  It costs a tiny bit (~1%) in performance on the
	  micro test, but is cleaner and the result is still
	  measurably faster than the reference VM.

The prior webrev is now at

	http://cr.openjdk.java.net/~jcoomes/6821693-taskqueue64.old.02
	
-John




More information about the hotspot-gc-dev mailing list