RFR (S): JDK-6991197 G1: specialize deal_with_reference() for narrowOop* (original) (raw)

Stefan Karlsson stefan.karlsson at oracle.com
Mon Jan 27 13:15:21 UTC 2014

Thomas,

On 2014-01-27 12:51, Thomas Schatzl wrote:

Hi all,

can I have your opinion and optionally a review for the following small change? Like the bug title reads, the change manually specializes G1ParScanThreadState::dealwithreference() for the narrowOop*, allowing some additional specialization. Measurements indicate a small improvement (~1%) on object copy time for specjbb, however not significant. As I have the change on hand (I had to do it anyway for measurements) I would like to ask your opinion whether to keep it or not. If so, ask for review. I am actually undecided - personally I tend to like it (as it trims fat) and it seems small enough to keep it and does not distract too much, but it does not really show up on measurements. Testing: jprt, specjbb2005/13, specjvm98/2008, CRM Fuse Bug entry: https://bugs.openjdk.java.net/browse/JDK-6991197 Webrev: http://cr.openjdk.java.net/~tschatzl/6991197/webrev/

Have you considered specializing has_partial_array_mask() to always return false when T = narrowOop? If you do it that way, you would get rid of the code duplication.

thanks, StefanK

Thanks, Thomas

More information about the hotspot-gc-dev mailing list