Sv: Re: RFR(XXS): 8040244 compiler/whitebox/IsMethodCompilableTest.java fails
Nils Eliasson
nils.eliasson at oracle.com
Tue Jun 10 18:42:21 UTC 2014
Thanks for your reviews Vladimir and Christian!
//N
----- Ursprungligt meddelande -----
Från: nils.eliasson at oracle.com
Till: christian.thalinger at oracle.com
Kopia: hotspot-compiler-dev at openjdk.java.net
Skickat: tisdag, 10 jun 2014 11:19:16 GMT +01:00 Amsterdam/Berlin/Bern/Rom/Stockholm/Wien
Ämne: Re: RFR(XXS): 8040244 compiler/whitebox/IsMethodCompilableTest.java fails
Xcomp It honors the CompileCommand in not compiling any other methods, but the methods matching compileonly is compiled directly.
Reading the bug’s comments I think I get it:
" The test uses reflection to get the method being tested. -Xcomp makes the reflecting method inline the test method. That method is never invalidated. The test method never gets any invocations and never ends up as a c2.compile. ”
I did some additional expermiments and tried adding the dontinline command but it didn't help. Xcomp changes the behavior of the thresholdpolicy. After the invalidation after the first iteration - the methods get compiled at level 3 but then reverted to level 1 instead of promoted to 4. Anyway - the there is no good reason for running this test with Xcomp. The suggested change makes the test work as intended.
//N
On Jun 9, 2014, at 1:45 AM, Nils Eliasson < nils.eliasson at oracle.com > wrote:
Hi,
Please review this patch for bug 8040244.
Summary: The test uses CompileCommand compileonly and breaks when forced to run with Xcomp. Add -Xmixed to prevent -Xcomp from being used.
webrev: http://cr.openjdk.java.net/~neliasso/8040244/webrev.01/
bug: https://bugs.openjdk.java.net/browse/JDK-8040244
Regards,
Nils
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20140610/e8490d25/attachment.html>
More information about the hotspot-compiler-dev
mailing list